chore: normalize formatting (2-space → tabs) in stream-error-patterns files

Re-tab only — no logic changes. Follows the repo's tab-based formatting for these files, matching the CommandCode executor/translator style. Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
docs(open-sse): in-stream error handling recipe + CHANGELOG
2026-08-05 09:03:41 +07:00 · 2026-08-04 23:35:57 +07:00 · 2026-08-04 23:35:25 +07:00 · 2026-08-04 23:34:21 +07:00 · 2026-08-04 23:34:00 +07:00 · 2026-08-04 23:33:12 +07:00
502 changed files with 58500 additions and 12891 deletions
--- a/.env.example
+++ b/.env.example
@@ -34,5 +34,8 @@ NEXT_PUBLIC_CLOUD_URL=https://9router.com
 # ALL_PROXY=socks5://127.0.0.1:7890
 # NO_PROXY=localhost,127.0.0.1

+# Optional SearXNG endpoint for the built-in unauthenticated web-search provider.
+# SEARXNG_URL=http://searxng:8080/search
+
 # Currently unused by application runtime (kept as reference)
 # INSTANCE_NAME=9router
--- a/.gitignore
+++ b/.gitignore
@@ -75,3 +75,17 @@ gitbook/README.md

 # Refactor backup reference (do not bundle/lint)
 open-sse.old/
+.graphifyignore
+graphify-out/*
+
+# Local-only working dirs (notes, vendored repos, scripts, skills)
+.claude/
+.docs/
+.repo/
+.script/
+.codegraph/
+.PR/
+.next-analyze/*
+
+# CommandCode CLI local state (auth/taste/projects)
+.commandcode/
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,6 +1,212 @@
+# Unreleased
+
+## Features
+
+- **Stream error patterns**: per-provider `streamErrorPatterns` setting (UI: provider page → Stream Error Patterns) — HTTP-200 streams whose first bytes match configured patterns (plain text or `/regex/`) are treated as failed requests: fallback works for non-streaming and early stream errors, and late streaming errors are logged as FAILED. Zero overhead when unconfigured.
+
+## Fixes
+
+- **CommandCode**: in-stream `{"type":"error"}` events now emit OpenAI error chunks + an executor early-peek → 502 fallback instead of fake success content (`[CommandCode error: ...]`).
+
+# v0.5.45 (2026-07-30)
+
+## Features
+
+- **Providers**: add Poolside (OpenAI-compatible)
+- **Providers**: add api-airforce, baidu, bazaarlink, bluesminds, kilo-gateway, llm7, morph, sambanova, tencent
+- **OAuth**: zed / trae / windsurf providers + harden callback proxies
+- **CLI tools**: set Claude Code max context tokens
+- **Qoder**: PAT auth + refresh model list
+- **Gemini**: Gemini 3.6 Flash tier routing + Gemini 3.5 Flash Lite
+- **Claude**: bump default Opus to `claude-opus-5`
+- **Kiro**: add Claude Opus 5 models
+- **Usage**: Kimi and DeepSeek usage handlers
+- **Usage**: SuperGrok weekly pool via gRPC-web
+
+## Fixes
+
+- **Refresh**: rotate `refresh_token` between retry attempts
+- **Kiro**: canonicalize tool history and route API keys correctly
+- **Kiro**: normalize dashboard thinking intensity models
+- **Cursor**: stop leaking agent tool errors as text
+- **Gemini**: fill empty tool schemas after `$ref` strip
+- **Antigravity**: strip `stream_options` from non-stream requests
+- **Jina-reader**: recover after transient errors, use JSON POST API
+- **Usage**: record exact embedding tokens
+- **Tunnel**: preserve successor cloudflared PID
+- **Console-log**: initialize capture at server boot + prevent SSE proxy buffering
+- **Dashboard**: count dual-auth, free-tier OAuth and API-key connections correctly
+- **Dashboard**: flex quota rows, thin global scrollbars, no hidden-row overflow
+
+## Docs
+
+- **i18n**: expand pt-BR translation to 986 terms
+- README: Indonesian translation
+
+# v0.5.40 (2026-07-20)
+
+## Features
+
+- **i18n**: add Khmer (km) translations
+- **CLI tools**: configure Grok Build subagent models
+- **Kimi**: merge OAuth into dual-auth provider, add K3 / K2.7 models
+- **Dashboard**: ProviderTopology flow animation
+
+## Fixes
+
+- **DB**: resolve better-sqlite3 parameter binding crash
+- **Translator**: pass `service_tier` through OpenAI → Responses conversion
+- **Kiro**: map GPT-5.6 reasoning effort fields
+- **Kiro**: validate terminal streams before emitting output
+- **Kiro**: map GPT reasoning effort fields
+- **Codex**: current `client_version` + refresh-aware model sync
+- **Alicode-intl**: split into Coding Plan + Model Studio providers
+- **Cursor**: HTTP/2 AgentService support + version bump 3.12.17
+- **Dashboard**: cut duplicate API/icon spam, lazy-load provider assets
+
+# v0.5.35 (2026-07-16)
+
+## Features
+
+- **xAI**: Grok Imagine video generation (`/v1/videos`) + CLI
+- **CLI tools**: Grok Build setup — choose separate main/general-purpose/explore/plan models and preserve each model's context window
+- **GitHub Copilot**: route Claude models through Copilot's native `/v1/messages`
+- **Kiro**: add GPT-5.6 model family (#2596)
+- **RTK**: `X-9Router-Token-Saver` header to bypass token savers per request
+- **Providers**: quota visibility settings
+- **Translator**: drop temperature for all Claude models
+- **i18n**: Thai (th) + Persian (fa) translations / README
+
+## Fixes
+
+- **Providers**: bulk-add API keys no longer overwrite existing keys (gap-fill `Key N`)
+- **Anthropic**: lowercase `anthropic-version` header to prevent duplication on `/v1/messages`
+- **Alicode-intl**: use DashScope compatible-mode endpoint so standard keys work
+- **Grok CLI**: align Grok Build with current subscription protocol (#2590)
+- **Grok CLI**: surface `expiresAt` so proactive token refresh fires (#2546)
+- **Kiro**: improve direct session cache reuse
+- **Models**: populate capabilities for live-catalog LLM models
+- **Models**: list compatible provider models in `/v1/models`
+- **Thinking**: send explicit `thinking:{type:adaptive}` alongside `output_config.effort`
+- **Translator**: strip `client_metadata` when converting openai-responses → openai
+
+## Improvements
+
+- **Perf**: skip inactive background services on startup
+
+## Docs
+
+- README: Persian YouTube tutorial
+
+# v0.5.30 (2026-07-10)
+
+## Features
+
+- **Perplexity**: add Agent API provider (#2492)
+- **Grok CLI**: add Grok CLI / Grok Build provider with OAuth device-code flow (#2502)
+- **Featherless**: add OpenAI-compatible provider presets
+- **SearXNG**: configure endpoint via SEARXNG_URL env (#2499)
+- **Providers**: add max thinking level for gpt-5.6-sol (#2500)
+- **Headroom**: add extras detection and install UI (#2403)
+- **Headroom**: activate/uninstall extras + fix interpreter detection
+- **PXPipe**: PXPIPE token saver — multimodal prompt compression (#2465)
+- **Proxy-Pools**: auto-rotate strategy for no-auth providers (#2409)
+
+## Fixes
+
+- **Cloudflare-AI**: support accountId in bulk key import (#2449)
+- **DB**: backup on schema change, MCP child cleanup, codex models, usage providers OOM
+- **Codex**: avoid bare-email OAuth dedup (#2477)
+- **CLI**: allow staged app bundle builds (#2479)
+- **Headroom**: compress Kiro conversation state (#2488)
+- **Gemini-CLI**: raise output floor for thinking and add validated toolConfig (#2486)
+- **GitHub**: label Copilot profiles by account identity (#2498)
+- **OpenAI-to-Claude**: unwrap bare {function:{…}} tools without parent type (#2473)
+- **Translator**: clamp thinking effort max->xhigh for OpenAI format (#2466)
+- **RTK/find**: detect and group Windows backslash-style find output (#2448)
+- **Codex**: handle fast tier and capacity SSE (#2452)
+- **Volcengine-ark**: clamp Kimi max_tokens to 32768 endpoint cap
+- **Antigravity**: align provider fingerprint with IDE Desktop 2.1.1 (#2389)
+- **Pricing**: update Claude/Codex model rates and add new models
+
+## Improvements
+
+- **i18n(zh-CN)**: complete Chinese translations for all UI strings (#2436)
+- **API**: caching for tunnel and version status endpoints
+- **Perf**: faster dev startup and lighter bundle
+
+# v0.5.20 (2026-07-07)
+
+## Features
+
+- **Thinking**: per-model thinking level picker on provider page — appends `(level)` suffix to copied model names for forced reasoning effort across all formats (openai, claude, gemini, deepseek, kimi, qwen, zai, minimax, hunyuan, step)
+- **RTK**: add JS-native git-log filter (#2423)
+- **Caveman**: add targeted upstream-aligned style rules (#2424)
+- **i18n**: add Farsi (fa) language support (#2385)
+
+## Fixes
+
+- **Thinking**: strip `(level)` suffix from upstream `body.model` so providers no longer reject requests
+- **Translator**: preserve developer instructions in openai-responses conversion (#2434)
+- **count_tokens**: count structured Anthropic blocks (#2419)
+- **Volcengine-ark**: clamp GLM-5 max_tokens to model output ceiling (#2428)
+- **Kimi**: normalize reasoning_effort to backend enum (#2427)
+- **Claude**: reconcile max_tokens vs thinking budget and lift per-model ceiling (#2381)
+- **Kiro**: deliver system prompt natively, add Opus 4.5/4.7/4.8, tolerate dash version ids (#2366)
+- **Headroom**: proxy dashboard through app (#2372)
+- **MITM**: recover from stale lock file on server start
+
+# v0.5.18 (2026-07-03)
+
+## Features
+
+- **Usage**: track cached tokens + correct input/output/cache cost (#2209) — hodtien
+- **Codex**: show reset credit expiry details (#2290) — Rafli Ahmad Zulfikar
+- **NVIDIA**: add new models and capabilities — decolua
+- **ClinePass**: add provider support — sternelee
+
+## Fixes
+
+- **Usage**: dedupe streaming request-details log entries — Qin Li
+- **Claude**: drop foreign thinking signatures in passthrough — decolua
+- Prevent non-SSE stream pipe crash and cross-IdP account overwrites (#2244) — KunN-21
+- **Kiro**: route IdC auth to regional CodeWhisperer surface (#2297) — Volodymyr Saakian
+- **Kiro**: add Claude Sonnet 5 model support (#2264) — Edison42
+- **Xiaomi-tokenplan**: region selector, key validation, multi-connection (#2251) — MiQieR
+- **Translator**: strict Anthropic content block compliance (#2225) — Sahrul Ramadhan Hardiansyah
+- **Kimchi**: strip reasoning_content echo to bound multi-turn input tokens — KunN-21
+- **Kimchi**: bump User-Agent to kimchi/0.1.40 (#2256) — Ansh7473
+- **Codebuddy-cn**: strip empty tool_calls arrays to preserve reasoning — zmf
+- **Antigravity**: preserve Claude tool delta index (#2223) — Sutarto Jordan Chrisfivo
+- **MITM**: generate root CA on server startup (#2228) — Sutarto Jordan Chrisfivo
+
+# v0.5.15 (2026-06-29)
+
+## Features
+
+- Add Kimchi OAuth provider — Nant361
+- Refine Qwen vision/video + thinking model patterns — decolua
+- Opt-in Codex auto-ping quota keep-alive — Emirhan
+
+## Fixes
+
+- **Responses**: handle response.done terminal events (#2142) — rifuki
+- **Headroom**: skip unsafe responses tool history (#2132) — Sutarto Jordan Chrisfivo
+- **Translator**: map mid-conversation system message to user (claude→openai) — decolua
+- **Gemini**: normalize contents to prevent 400 invalid_argument (#2192) — warelik
+- **Gemini**: backfill thoughtSignature + suppress stream done sentinel — WARELIK
+- **Alicode**: preserve cache_control for DashScope providers (#2069) — Rex
+- **Antigravity**: strip deprecated/readOnly/writeOnly from tool schemas — iletai, Yudhistira-Official
+- **CodeBuddy CN**: show bonus packs as one-time, not monthly-replenishing — whale9820
+- **Kiro**: strip leaked <thinking> tags from content stream (#2158) — hamsa0x7
+- **Tray**: make Windows context menu DPI-aware — Emirhan
+- **Kilocode**: expose full gateway catalog in combo model picker — jellylarper
+- **OpenCode**: fix Go GLM — decolua
+
 # v0.5.12 (2026-06-26)

 ## Features
+
 - Add token-saver dashboard page — decolua
 - Add bulk delete for provider connections — teddytkz
 - Resolve GitHub Copilot model catalog from upstream — caiqinzhou
@@ -9,6 +215,7 @@
 - Overhaul Blackbox provider catalog + WebUI test support — suryacagur

 ## Fixes
+
 - Provider thinking compatibility (DeepSeek/Gemini) — Mink Nguyen
 - Stop double-counting streaming usage at source — decolua
 - Usage logging dedupe to reduce stats churn — Mink Nguyen
@@ -20,6 +227,7 @@
 - Support Gemini native TTS generateContent endpoint — nguyenha935
 - Add missing zh-CN endpoint key label (i18n) — weimaozhen
 - CodeBuddy: only send reasoning params when client requests reasoning (#2071) — Rex
+- CodeBuddy CN: show one-shot bonus packs as expiring, not monthly-replenishing
 - Show custom provider models in combo picker — Sapto
 - Docker: add docker-compose.yml with headroom enabled by default — nitsuahlabs
 - Clarify token diagnostics vs provider billing (headroom, #1998) — Sutarto Jordan Chrisfivo
@@ -36,11 +244,13 @@
 # v0.5.8 (2026-06-21)

 ## Features
+
 - **Antigravity**: native image generation support (image models tagged kind:image, hiển thị trong media-providers UI)
 - **CodeBuddy CN**: API key auth + credit quota tracker
 - **CodeBuddy CN**: short model prefix alias "cbcn"

 ## Fixes
+
 - **MiniMax-M3**: enable vision capability
 - **Headroom**: support Docker sidecar proxy
 - **Antigravity**: image executor fixes
@@ -54,12 +264,14 @@
 # v0.5.6 (2026-06-20)

 ## Features
+
 - **Ponytail**: minimalist code generation feature
 - **Headroom**: proxy lifecycle management + dashboard UI (one-click start/stop, install detection, status probing, token saver, claude↔openai shape conversion)
 - **CodeBuddy CN**: new OAuth provider (copilot.tencent.com) — 15-model catalog, /v2 inference, forced streaming, OpenAI-style reasoning
 - **OpenCode-Go**: align models with official endpoints; route Qwen 3.7 MiniMax via /v1/messages, GLM/Kimi/DeepSeek/MiMo via /chat/completions

 ## Fixes
+
 - **Anthropic-compatible validation**: use POST /v1/messages (GET /models not spec, false "invalid" for valid keys)
 - **CLI tools**: tolerate JSONC configs in all 8 settings routes (opencode, openclaw, kilo, droid, cowork, copilot, claude, cline)
 - **Gemini/Antigravity**: preserve 'pattern' in tool schema translation (glob/grep)
@@ -70,6 +282,7 @@
 # v0.5.4 (2026-06-18)

 ## Fixes
+
 - **Kiro**: honor thinking effort budgets
 - **AG/Kiro/Xiaomi**: provider fixes
 - **Combo/Fusion**: flatten tool history in panel calls to prevent 503
@@ -79,6 +292,7 @@
 # v0.5.2 (2026-06-17)

 ## Features
+
 - **Combo Fusion strategy** — fans the prompt out to all member models in parallel, then a configurable judge model synthesizes one final answer (quorum-grace, anonymized sources, graceful degradation)
 - **Per-combo strategy selector** — pick `fallback` / `round-robin` / `fusion` / `capacity` per combo (replaces the old round-robin toggle), with a judge picker for fusion
 - **Capacity auto-switch** — reorders models per request so images/PDFs route to capable models first
@@ -86,6 +300,7 @@
 - **Claude auto-ping** — warms the 5h quota window right after reset so a fresh window starts immediately (per-connection toggle)

 ## Fixes
+
 - **Claude 429**: stop hammering the OAuth usage endpoint — cache resetAt, throttle quota refresh to 3 min, cool down after a 429 (chat unaffected)
 - **Usage logs always empty**: missing `await` on `getAdapter()` in `getRecentLogs` made `/api/usage/logs` & `/api/usage/request-logs` return nothing
 - **Executors**: strip params unsupported by the provider/model (drops deprecated `temperature` for claude-opus-4 → Anthropic 400)
@@ -97,11 +312,13 @@
 - **Security**: SSRF hardening on web fetch

 ## Internal
+
 - Large **open-sse / translator refactor** (~40 commits): unified provider/model registry (LiteLLM-style `models[]` + `kind` field, 100 co-located registry files), single-sourced media/OAuth/refresh/token URLs, registry-based dispatch for usage & token-refresh, DRY translator concerns (buildUsage, encodeDataUri, finishReasonMap, chunkBuilder, reasoningDelta…), ESM-safe registry init, large-file splits, dead-code removal, and golden/no-regression test gates

 # v0.4.80 (2026-06-13)

 ## Features
+
 - Vercel AI Gateway: support embeddings, images and credit usage (#1183)
 - Add MiMo Free no-auth provider (#1789)
 - Vertex: support ADC `authorized_user` credential
@@ -110,6 +327,7 @@
 - Kiro: enable multi-endpoint failover for GenerateAssistantResponse (#1722)

 ## Fixes
+
 - Security: re-auth on DB export/import + SSRF guard on web fetch
 - Auth: real client IP rate-limiting + remote default-password guard
 - Cerebras/Mistral: strip unsupported `client_metadata` from downstream requests (#1742)
@@ -128,11 +346,13 @@
 - Dashboard: show provider node name instead of connection name in topology (#1770) + show explicit `kind="llm"` combos on combos page (#1684)

 ## Docs
+
 - README: add Indonesian 9Router tutorial video (#1709)

 # v0.4.71 (2026-06-06)

 ## Features
+
 - Caveman: add wenyan classical Chinese levels and sync upstream prompts; locale-based visibility on endpoint page
 - i18n: endpoint exposure notice across multiple languages + Russian README
 - Antigravity: add gemini-3.5-flash-extra-low (Low) model
@@ -141,6 +361,7 @@
 - MiniMax: add MiniMax-M3 + update Quota Tracker coding/CN (#1631)

 ## Fixes
+
 - Codex: harden streaming timeouts (stall/connect raised to 60s, configurable per-provider), accept `response.done` event, and always emit a terminal `response.failed` + `[DONE]` for Responses passthrough when a stream closes, stalls, or aborts before a terminal event — prevents codex clients from hanging (#1648, #1680, #1688, #1618)
 - Codex: durable OAuth refresh lifecycle (#1664)
 - Tunnel: skip virtual interfaces to prevent false netchange watchdog
@@ -154,21 +375,25 @@
 - Model-test: route image/STT probes to their real endpoints, harden STT ping; add opencode-go + xiaomi-tokenplan to connection test (#1576, #1628)

 ## Improvements
+
 - Dashboard: reorganize menu actions across sidebar/header/profile
 - Translator: add data-driven coverage, bug-exposing cases, and real provider smoke tests

 # v0.4.66 (2026-05-29)

 ## Features
+
 - Add Qoder provider: device-flow OAuth, COSY signing, WAF-bypass body encoding, live model catalog, dashboard quota tracker, 11 models (#1372)
 - Add new models: Claude Opus 4.8 (Claude Code), GPT 5.4 Mini (Codex)

 ## Fixes
+
 - DeepSeek thinking mode: echo `reasoning_content` back on follow-up/tool-call turns so OpenCode-free and custom providers no longer 400 with "reasoning_content must be passed back" (#1543)
 - Reasoning injector: match deepseek/kimi model ids case-insensitively (covers custom providers using capitalized model names)
 - OpenCode suggested-models: include free models without the `-free` suffix, e.g. `big-pickle` (#1535)

 ## Improvements
+
 - Codex: trim sunset models, keep gpt-5.5 / gpt-5.4 / gpt-5.3-codex family, add gpt-5.4-mini
 - volcengine-ark: refresh model list (add DeepSeek-V4-Flash/Pro, drop EOL entries)
 - Lower stream stall timeout 35s → 30s for faster hang detection
@@ -176,18 +401,21 @@
 # v0.4.63 (2026-05-26)

 ## Fixes
+
 - GitHub Copilot: never route Gemini/Claude models to the `/responses` endpoint; prevents misleading "does not support Responses API" 400s (#1062)
 - proxyFetch: restore missing `Readable` import causing runtime `ReferenceError` in DNS-bypass fetch path

 ## Improvements
+
 - Lower stream stall timeout from 60s → 35s for faster hang detection

 # v0.4.62 (2026-05-26)

 ## Fixes
+
 - Codex: auto-retry when upstream drops mid-stream (no more hangs)
 - Codex: fix random 400/404 errors, tool-calling failures, and unstable prompt cache
- MITM: support Antigravity 2.x 
+- MITM: support Antigravity 2.x
 - Sanitize Read tool args to prevent retry loops from non-Anthropic models (#1144)
 - Implement json_schema fallback for OpenAI-compatible providers without native Structured Output (#1343)
 - Strip empty Read pages argument in OpenAI-to-Claude translator (#1354)
@@ -196,25 +424,30 @@
 - Gemini CLI: reuse stored OAuth project IDs for quota checks and show clearer setup guidance when the project is missing (#1271, #1428)

 ## Features
+
 - Add Cloudflare Workers proxy deployer and pool integration (#1360)
 - Add Deno Deploy relays support and improved proxy pools dashboard layout (#1437)

 ## Improvements
+
 - Refactor Tunnel into dedicated Cloudflare and Tailscale manager modules
 - Refactor tokenRefresh service with in-flight dedup to prevent refresh_token_reused errors

 # v0.4.59 (2026-05-21)

 ## Fixes
+
 - OAuth: fix login flow on Windows

 # v0.4.58 (2026-05-21)

 ## Features
+
 - xAI Grok provider (OAuth, API key, image)
 - Provider limits: paginated accounts with page size controls

 ## Fixes
+
 - Tailscale: fix connection status on Windows (#1300)
 - Tunnel: fix false "checking" when tunnel URL is reachable
 - Stream: fix pipe errors on client disconnect/abort
@@ -222,11 +455,13 @@
 # v0.4.55 (2026-05-18)

 ## Features
+
 - Xiaomi MiMo Token Plan: region selector (Singapore / China / Europe) — keys are cluster-specific
 - Antigravity: risk confirmation dialog before first connection
 - Gemini CLI: surface upstream retry delay on 429 errors

 ## Fixes
+
 - MITM: cannot kill process on macOS under sudo (lsof not found in PATH)
 - Stream: false-positive stall timeout on Claude reasoning / Kiro responses
 - Tunnel: cannot re-enable after disable (stuck state)
@@ -235,16 +470,19 @@
 - Antigravity OAuth: metadata now matches the official client

 ## Improvements
+
 - Gemini CLI: bump engine to 0.34.0
 - Re-hide `qwen` (OAuth EOL) and `iflow` (not ready) providers

 # v0.4.52 (2026-05-17)

 ## Features
+
 - Add Vercel AI Gateway provider support (#1183)
 - rtk: Kiro format tool result compression — handle conversationState.history & currentMessage, preserve error results, ~13.6% savings (#1194)

 ## Fixes
+
 - openclaw: normalize agent.model object form `{primary, fallbacks}` before .startsWith → fix TypeError & 'not configured' status (#1216)
 - Usage Details pagination: stay inside mobile viewport <640px (#1218)
 - Fix test model error
@@ -254,6 +492,7 @@
 # v0.4.50 (2026-05-16)

 ## Fixes
+
 - Fix duplicate tray icon on macOS when hiding to tray
 - Fix tray not showing in background mode on macOS
 - Fix hide to tray broken on Windows/Linux
@@ -262,11 +501,13 @@
 # v0.4.49 (2026-05-16)

 ## Features
+
 - Add Kiro provider support: full request/response translation, live model listing, reasoning content support
 - Add `buildOutput` RTK filter with autodetect for npm/yarn/cargo build logs
 - Add MITM warning notification in tray and dashboard

 ## Improvements
+
 - Add modalities (input/output) to model configuration for OpenCode
 - Fix tray hide-to-tray: keep current process alive instead of spawning detached child (fixes macOS NSStatusItem ghost icon)
 - Fix tray kill: graceful shutdown with SIGTERM/SIGKILL escalation
@@ -275,49 +516,11 @@
 - Update i18n across 32 languages

 ## Fixes
+
 - Fix model check (test-models) blocked by dashboardGuard: pass machineId-based CLI token in internal self-calls

 # v0.4.46 (2026-05-15)

 ## Breaking Changes
+
 - Tunnel public URL changed — old tunnel links no longer work, please reconnect to get the new URL
-
-# v0.4.44 (2026-05-15)
-
-## Features
- Add Blackbox provider with `bb` alias (#1143)
- Add Xiaomi token plan provider
- Enhance model select modal UX + modal traffic lights (#1111)
- Default Usage dashboard period to Today (#1141)
-
-## Fixes
- Fix Cowork model selection and Windows CLI packaging (#1129)
- Update provider name retrieval for compatibility provider (#1135)
- Update JWT_SECRET handling
-
-# v0.4.41 (2026-05-14)
-
-## Features
- Add jcode CLI tool integration with auto-configuration (#1047)
- Redesign CLI Tools dashboard: grid layout (1/2/3 cols) + dedicated detail page per tool
- Add drag-and-drop reordering for combo models (#1108)
- Add Today period option to Usage & Analytics (#1063)
- Add DeepSeek V4 Pro effort aliases (#950)
-
-## Fixes
- fix(autostart): work on nvm + npm 9/10, actually register with launchctl (#1104, fixes #1082)
- Fix Ollama usage not tracked/shown in UI (#1102)
- fix(opencode): preserve DeepSeek reasoning content (#1099, fixes #1093)
- Fix TUI input lag (replace enquirer with native readline, persistent raw mode)
- fix(ui): show API key row actions on mobile (#1112)
-
-## Improvements
- Sync DeepSeek TUI card style with other CLI tools (badges, layout, manual config modal)
- Add official logos for Amp CLI, jcode, Qwen Code (replace generic icons)
- Resize deepseek-tui icon 1024→128 with padding for visual consistency
-
-# v0.4.39 (2026-05-14)
-
-## Fixes
- fix(docker): restore `/app/server.js` (v0.4.38 regression)
-
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -0,0 +1,91 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## What this is
+
+9Router (`9router-app`) — a local AI routing gateway + Next.js dashboard. It exposes one OpenAI-compatible endpoint (`/v1/*`) and routes traffic across 40+ upstream providers with format translation, model-combo fallback, multi-account fallback, OAuth/API-key credential management, token refresh, quota/usage tracking, and optional cloud sync.
+
+Two published artifacts live in this one repo:
+- The **dashboard + gateway** (root `package.json`, `9router-app`) — the Next.js server that does the actual routing.
+- The **CLI launcher** (`cli/`, published to npm as `9router`) — a separate package that installs/starts the server and manages the tray. It has its own `package.json`, version, and build.
+
+The code lives in `src/` (Next.js app + dashboard/compat APIs), `open-sse/` (the provider-agnostic routing/translation engine), `cli/` (the launcher package), and `tests/`.
+
+## Commands
+
+Dashboard/gateway (run from repo root):
+```bash
+cp .env.example .env
+npm install
+PORT=20128 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run dev   # dev (webpack, port 20127 by default via next dev)
+npm run build && PORT=20128 HOSTNAME=0.0.0.0 npm run start           # production
+```
+- Bun variants: `npm run dev:bun` / `build:bun` / `start:bun`.
+- Default runtime port is **20128** (dashboard at `/dashboard`, API at `/v1`).
+- Lint: `npx eslint .` (config `eslint.config.mjs`, extends `eslint-config-next`).
+
+CLI package (`cli/`):
+```bash
+npm run cli:pack       # build + npm pack from root
+cd cli && npm run dev  # nodemon watch
+```
+
+Tests (vitest, in `tests/`, an **independent** ESM package — not wired into root `npm test`):
+```bash
+npm install                             # ROOT deps first — tests import from src/ which needs `open`, `undici`, etc.
+cd tests && npm install                 # then tests' own deps (vitest) → tests/node_modules (allowed by tests/.gitignore)
+npx vitest run                          # all tests; auto-discovers tests/vitest.config.js
+npx vitest run unit/capabilities.test.js   # single file (path relative to tests/)
+```
+> The committed `tests/package.json` `test` script hardcodes Unix paths (`NODE_PATH=/tmp/node_modules …`) — a shared-install workaround from upstream. On Windows (or anywhere), ignore it and use the `npx vitest` form above; `vitest.config.js` resolves the `open-sse`/`@/` aliases from the repo root regardless of where vitest lives.
+>
+> **The suite is NOT expected to be all-green on a plain checkout.** ~938 pass, ~64 fail. Judge regressions with `tests/__baseline__/verify-no-regression.mjs`, not a raw run. Expected red:
+> - 26 catalogued in `tests/__baseline__/known-fails.txt` (rtk, oauth-cursor-auto-import, translator-request-normalization, …).
+> - `unit/embeddings.cloud.test.js` imports `cloud/src/handlers/embeddings.js` — the `cloud/` worker dir is **not in this repo**, so it always fails here.
+> - `unit/xai-oauth-service.test.js` times out (5s) when the xAI endpoint-discovery fetch isn't reachable/mocked.
+> - `real/*.real.test.js` make live provider calls — need credentials, skip otherwise.
+- `*.real.test.js` under `tests/translator/real/` make live provider calls — skip unless credentials are set.
+- Regression baselines: `tests/__baseline__/verify-*.mjs` compare against committed snapshots (providers, aliases, OAuth URLs). Run these after touching provider registry / alias logic.
+
+## Architecture
+
+Two authoritative docs already exist — read them before working in these areas rather than re-deriving:
+- `docs/ARCHITECTURE.md` — full system: request lifecycle, combo/account fallback, OAuth + token refresh, cloud sync, data model.
+- `open-sse/AGENTS.md` — the routing/translation engine's own conventions and "how to add a provider/executor/translator". **Read this before editing anything under `open-sse/`.**
+
+### Request flow (the thing to understand first)
+`src/app/api/v1/*` route (Next rewrite maps `/v1/*` → `/api/v1/*` in `next.config.mjs`)
+→ `src/sse/handlers/chat.js` (parse, combo expansion, account-selection loop)
+→ `open-sse/handlers/chatCore.js` (detect source format, translate request, dispatch to executor, retry/refresh, stream setup)
+→ `open-sse/executors/*` (per-provider upstream call; `default.js` handles any OpenAI-compatible provider)
+→ `open-sse/translator/*` (client format ↔ provider format)
+→ SSE back to client.
+
+`src/sse/` is the app-side entry glue; `open-sse/` is the provider-agnostic engine (also usable standalone). Cross that boundary consciously.
+
+### Translator engine (`open-sse/translator/`)
+- Pivots through **OpenAI as the intermediate format**. A translator registered on an exact `source:target` pair (e.g. `claude:kiro`) runs as a **direct route**, skipping the lossy double-hop. Prefer a direct route for fragile pairs (thinking blocks, tool ids, non-base64 images, `is_error`).
+- Translators **self-register** via `register(from, to, reqFn, resFn)` as an import side effect — a new translator file MUST be imported in `open-sse/translator/index.js` or it never runs.
+- Never hardcode role/block/model strings — use `open-sse/translator/schema/` and `open-sse/config/` constants. Config-driven and DRY is enforced by convention here.
+
+### Provider registry (`open-sse/providers/registry/*`)
+- One file per provider. `providers/registry/index.js` is an **auto-generated** static import list — regenerate it with `scripts/migrate-registry.mjs` / `injectDisplayToRegistry.mjs`, don't hand-edit.
+- Add a provider: copy `providers/REGISTRY_TEMPLATE.js`, add models to `config/providerModels.js`. Only add an executor for non-OpenAI-compatible upstreams.
+
+### Persistence — IMPORTANT (ARCHITECTURE.md is stale here)
+State is **no longer `db.json`**. It's a SQLite layer under `src/lib/db/` with an adapter fallback chain (`driver.js`): `bun:sqlite` → `better-sqlite3` (optional native dep) → `node:sqlite` (Node ≥22.5) → `sql.js` (pure-JS fallback, always works). `better-sqlite3` is deliberately in `optionalDependencies` so install never fails without build tools.
+- `src/lib/localDb.js` is a **backward-compat shim** re-exporting `src/lib/db/index.js`. New code should import from `@/lib/db/index.js`; per-entity logic lives in `src/lib/db/repos/*`. Schema/migrations in `src/lib/db/migrations/`.
+- DB file location resolves via `src/lib/db/paths.js` (`DATA_DIR`, else `~/.9router/`).
+- Usage/logs (`src/lib/usageDb.js`, `usage.json` + `log.txt`) still live under `~/.9router` and do **not** follow `DATA_DIR`.
+
+### RTK token saver (`open-sse/rtk/`)
+Pre-translate hooks that compress `tool_result` content in-place to cut tokens. **Fail-open**: any error returns null and leaves the body untouched — never throw out of them. Skips `is_error`/`status:"error"` results to preserve traces.
+
+## Conventions & gotchas
+
+- Plain JavaScript (ESM), no TypeScript. `@/*` path alias → `src/*` (`jsconfig.json`).
+- `custom-server.js` wraps the Next standalone server to derive client IP from the TCP socket and strip attacker-controlled `X-Forwarded-For` — trusting forwarding headers only from a loopback reverse proxy. Preserve this when touching request/IP/rate-limit code.
+- Security-sensitive env: `JWT_SECRET` (session cookie), `INITIAL_PASSWORD` (default `123456` — must override), `API_KEY_SECRET`, `MACHINE_ID_SALT`. Full env contract in `.env.example` and ARCHITECTURE.md's env matrix.
+- Binary/protobuf upstreams (kiro EventStream, cursor protobuf, commandcode NDJSON) don't round-trip through OpenAI — they're handled inside their own executor, not the translator.
+- Versioning: root and `cli/` are versioned independently; changes are logged in `CHANGELOG.md`. Commit style is Conventional Commits (`fix(translator): …`, `feat(...)`).
--- a/README.md
+++ b/README.md
@@ -13,11 +13,12 @@
  [![GHCR](https://img.shields.io/badge/GHCR-decolua%2F9router-blue?logo=github)](https://github.com/decolua/9router/pkgs/container/9router)
  [![License](https://img.shields.io/npm/l/9router.svg)](https://github.com/decolua/9router/blob/main/LICENSE)

-  <a href="https://trendshift.io/repositories/22628" target="_blank"><img src="https://trendshift.io/api/badge/repositories/22628" alt="decolua%2F9router | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
-  
-  [🚀 Quick Start](#-quick-start) • [💡 Features](#-key-features) • [📖 Setup](#-setup-guide) • [🌐 Website](https://9router.com)
+<a href="https://trendshift.io/repositories/22628" target="_blank"><img src="https://trendshift.io/api/badge/repositories/22628" alt="decolua%2F9router | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>
+
+[🚀 Quick Start](#-quick-start) • [💡 Features](#-key-features) • [📖 Setup](#-setup-guide) • [🌐 Website](https://9router.com)
+
+[🇻🇳 Tiếng Việt](./i18n/README.vi.md) • [🇨🇳 中文](./i18n/README.zh-CN.md) • [🇯🇵 日本語](./i18n/README.ja-JP.md) • [🇷🇺 Русский](./i18n/README.ru.md) • [🇹🇭 ไทย](./i18n/README.th.md) • [🇮🇷 فارسی](./i18n/README.fa_IR.md) • [🇮🇩 Indonesia](./i18n/README.id-ID.md)

-  [🇻🇳 Tiếng Việt](./i18n/README.vi.md) • [🇨🇳 中文](./i18n/README.zh-CN.md) • [🇯🇵 日本語](./i18n/README.ja-JP.md) • [🇷🇺 Русский](./i18n/README.ru.md)
 </div>

 ---
@@ -83,7 +84,7 @@ npm install -g 9router

 **2. Connect a FREE provider (no signup needed):**

-Dashboard → Providers → Connect **Kiro AI** (free Claude unlimited) or **OpenCode Free** (no auth) → Done!
+Dashboard → Providers → Connect **Kiro AI** (~50 credits/month free: Claude 4.5 + GLM-5 + MiniMax) or **OpenCode Free** (no auth) → Done!

 **3. Use in your CLI tool:**

@@ -114,6 +115,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 ```

 Default URLs:
+
 - Dashboard: `http://localhost:20128/dashboard`
 - OpenAI-compatible API: `http://localhost:20128/v1`

@@ -125,6 +127,20 @@ Default URLs:

 <table>
  <tr>
+    <td align="center" width="320">
+      <a href="https://www.youtube.com/watch?v=X69n5Lm06Yw">
+        <img src="https://img.youtube.com/vi/X69n5Lm06Yw/maxresdefault.jpg" alt="Tiết kiệm chi phí LLM với 9Router" width="300"/>
+      </a><br/>
+      <b>🇻🇳 Tiếng Việt</b><br/>
+      <sub>Tiết kiệm chi phí LLM cho OpenClaw với 9Router<br/>by <a href="https://www.youtube.com/c/M%C3%ACAIblog">Mì AI</a></sub>
+    </td>
+    <td align="center" width="320">
+      <a href="https://youtu.be/VQAw612S27Y">
+        <img src="https://img.youtube.com/vi/VQAw612S27Y/maxresdefault.jpg" alt="9Router + Claude Code FREE Unlimited Setup" width="300"/>
+      </a><br/>
+      <b>🇵🇰 اردو / हिन्दी</b><br/>
+      <sub>9Router + Claude Code FREE Unlimited Setup<br/>by <a href="https://www.youtube.com/@BuildAIWithHamid">Build AI With Hamid</a></sub>
+    </td>
    <td align="center" width="320">
      <a href="https://www.youtube.com/watch?v=raEyZPg5xE0">
        <img src="https://img.youtube.com/vi/raEyZPg5xE0/maxresdefault.jpg" alt="9Router Setup Tutorial" width="300"/>
@@ -133,11 +149,11 @@ Default URLs:
      <sub>9Router + Claude Code FREE Setup<br/>by <a href="https://www.youtube.com/@BuildAIWithHamid">Build AI With Hamid</a></sub>
    </td>
    <td align="center" width="320">
-      <a href="https://www.youtube.com/watch?v=X69n5Lm06Yw">
-        <img src="https://img.youtube.com/vi/X69n5Lm06Yw/maxresdefault.jpg" alt="Tiết kiệm chi phí LLM với 9Router" width="300"/>
+      <a href="https://youtu.be/3dF5GIYMrcQ?si=bAyfyiHbARJQAHj_">
+        <img src="https://img.youtube.com/vi/3dF5GIYMrcQ/hqdefault.jpg" alt="9Router Setup Tutorial" width="300"/>
      </a><br/>
-      <b>🇻🇳 Tiếng Việt</b><br/>
-      <sub>Tiết kiệm chi phí LLM cho OpenClaw với 9Router<br/>by <a href="https://www.youtube.com/c/M%C3%ACAIblog">Mì AI</a></sub>
+      <b>🇺🇸 English</b><br/>
+      <sub>9Router + Claude Code FREE Setup<br/>by <a href="https://www.youtube.com/@BuildAIWithHamid">Build AI With Hamid</a></sub>
    </td>
    <td align="center" width="320">
      <a href="https://www.youtube.com/watch?v=o3qYCyjrFYg">
@@ -169,8 +185,6 @@ Default URLs:
      <b>🇺🇸 English</b><br/>
      <sub>FREE OpenClaw + Claude Opus 4.6<br/>by <a href="https://www.youtube.com/@BuildAIWithHamid">Build AI With Hamid</a></sub>
    </td>
-  </tr>
-  <tr>
    <td align="center" width="320">
      <a href="https://www.youtube.com/watch?v=CkVZZUSTXAI">
        <img src="https://img.youtube.com/vi/CkVZZUSTXAI/mqdefault.jpg" alt="Claude CLI Free Setup" width="300"/>
@@ -186,6 +200,25 @@ Default URLs:
      <sub>Cara Deploy 9Router di Hugging Face GRATIS Non-Stop! | Alternatif VPS RAM 16GB<br/>by <a href="https://www.youtube.com/@krisswuh">Krisswuh</a></sub>
    </td>
  </tr>
+  <tr>
+    <td align="center" width="320">
+      <a href="https://www.youtube.com/watch?v=GyX-DLvePW8">
+        <img src="https://img.youtube.com/vi/GyX-DLvePW8/hqdefault.jpg" alt="این شکلی از هر API ای استفاده کن برای هوش مصنوعی" width="300"/>
+      </a><br/>
+      <b>🇮🇷 Persian-فارسی</b><br/>
+      <sub dir="rtl">این شکلی از هر API ای استفاده کن برای هوش مصنوعی<br/>by <a href="https://www.youtube.com/@Matin_SenPai">Matin SenPai</a></sub>
+    </td>
+    <td align="center" width="320">
+      <a href="https://www.youtube.com/watch?v=hPusYX-5Pmw">
+        <img src="https://img.youtube.com/vi/hPusYX-5Pmw/maxresdefault.jpg" alt="Hướng Dẫn Setup OpenClaw + 9Router: Tạo Bot Zalo AI Tự Động Từ A-Z" width="300"/>
+      </a><br/>
+      <b>🇻🇳 Tiếng Việt</b><br/>
+      <sub>Hướng Dẫn Setup OpenClaw + 9Router: Tạo Bot Zalo AI Tự Động Từ A-Z<br/>by <a href="https://github.com/tuanminhhole">tuanminhhole</a></sub>
+    </td>
+    <td align="center" width="320"></td>
+    <td align="center" width="320"></td>
+    <td align="center" width="320"></td>
+  </tr>
 </table>

 </div>
@@ -284,6 +317,10 @@ Default URLs:
        <img src="./public/providers/cursor.png" width="60" alt="Cursor"/><br/>
        <b>Cursor</b>
      </td>
+      <td align="center" width="120">
+        <img src="./public/providers/kimchi.png" width="60" alt="Kimchi"/><br/>
+        <b>Kimchi</b>
+      </td>
    </tr>
  </table>
 </div>
@@ -296,12 +333,12 @@ Default URLs:
      <td align="center" width="150">
        <img src="./public/providers/kiro.png" width="70" alt="Kiro"/><br/>
        <b>Kiro AI</b><br/>
-        <sub>Claude 4.5 + GLM-5 + MiniMax<br/>Unlimited FREE</sub>
+        <sub>Claude 4.5 + GLM-5 + MiniMax<br/>50 credits/month free</sub>
      </td>
      <td align="center" width="150">
        <img src="./public/providers/opencode.png" width="70" alt="OpenCode Free"/><br/>
        <b>OpenCode Free</b><br/>
-        <sub>No auth • Auto-fetch models<br/>Unlimited FREE</sub>
+        <sub>No auth • Auto-fetch models<br/>Free (model list varies)</sub>
      </td>
      <td align="center" width="150">
        <img src="./public/providers/gemini.png" width="70" alt="Vertex AI"/><br/>
@@ -312,7 +349,11 @@ Default URLs:
  </table>
 </div>

-> **Note:** iFlow, Qwen and Gemini CLI free tiers were discontinued in 2026. Use Kiro / OpenCode Free / Vertex instead.
+> **Note:** iFlow, Qwen Code and Gemini CLI free tiers were discontinued in 2026. Use Kiro / OpenCode Free / Vertex instead.
+>
+> **Kiro AI** moved to a paid model in Sep 2025 — the free tier is now capped at **50 credits/month** (plus 500 trial credits for new accounts in the first 30 days). Paid tiers: Pro $20/mo (1,000 credits), Pro+ $40/mo (2,000), Pro Max $100/mo (5,000), Power $200/mo (10,000).
+> **OpenCode Free** model list fluctuates over time (some models free only for limited promos) — subject to change without notice.
+> **Vertex AI**: the $300 free credit for new GCP accounts is still valid, but since Mar 2026 the **Gemini API endpoint no longer consumes these credits** — call the **Vertex AI Studio** endpoint instead.

 ### 🔑 API Key Providers (40+)

@@ -404,22 +445,24 @@ Default URLs:

 ## 💡 Key Features

-| Feature | What It Does | Why It Matters |
-|---------|--------------|----------------|
-| 🚀 **RTK Token Saver** ([RTK](https://github.com/rtk-ai/rtk) ⭐40K) | Compress tool outputs (`git diff`, `grep`, `ls`, `tree`...) before sending to LLM | Save **20-40% input tokens** per request |
-| 🧠 **Headroom Token Saver** ([Headroom](https://github.com/chopratejas/headroom)) | Optional external `/v1/compress` proxy before provider routing | Save more context tokens without changing clients |
-| 🪨 **Caveman Mode** ([Caveman](https://github.com/JuliusBrussee/caveman) ⭐52K) | Inject caveman-speak prompt → LLM replies terse, technical substance preserved | Save **up to 65% output tokens** |
-| 🐴 **Ponytail** ([Ponytail](https://github.com/DietrichGebert/ponytail)) | Inject "lazy senior dev" prompt → LLM writes minimal, YAGNI-first code (Lite/Full/Ultra) | **Fewer output tokens, less refactoring** |
-| 🎯 **Smart 3-Tier Fallback** | Auto-route: Subscription → Cheap → Free | Never stop coding, zero downtime |
-| 📊 **Real-Time Quota Tracking** | Live token count + reset countdown | Maximize subscription value |
-| 🔄 **Format Translation** | OpenAI ↔ Claude ↔ Gemini ↔ Cursor ↔ Kiro ↔ Vertex | Works with any CLI tool |
-| 👥 **Multi-Account Support** | Multiple accounts per provider | Load balancing + redundancy |
-| 🔄 **Auto Token Refresh** | OAuth tokens refresh automatically | No manual re-login needed |
-| 🎨 **Custom Combos** | Create unlimited model combinations | Tailor fallback to your needs |
-| 📝 **Request Logging** | Debug mode with full request/response logs | Troubleshoot issues easily |
-| 💾 **Cloud Sync** | Sync config across devices | Same setup everywhere |
-| 📊 **Usage Analytics** | Track tokens, cost, trends over time | Optimize spending |
-| 🌐 **Deploy Anywhere** | Localhost, VPS, Docker, Cloudflare Workers | Flexible deployment options |
+| Feature                                                                           | What It Does                                                                             | Why It Matters                                    |
+| --------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------- | ------------------------------------------------- |
+| 🚀 **RTK Token Saver** ([RTK](https://github.com/rtk-ai/rtk) ⭐40K)               | Compress tool outputs (`git diff`, `grep`, `ls`, `tree`...) before sending to LLM        | Save **20-40% input tokens** per request          |
+| 🧠 **Headroom Token Saver** ([Headroom](https://github.com/chopratejas/headroom)) | Optional external `/v1/compress` proxy before provider routing                           | Save more context tokens without changing clients |
+| 🪨 **Caveman Mode** ([Caveman](https://github.com/JuliusBrussee/caveman) ⭐52K)   | Inject caveman-speak prompt → LLM replies terse, technical substance preserved           | Save **up to 65% output tokens**                  |
+| 🐴 **Ponytail** ([Ponytail](https://github.com/DietrichGebert/ponytail))          | Inject "lazy senior dev" prompt → LLM writes minimal, YAGNI-first code (Lite/Full/Ultra) | **Fewer output tokens, less refactoring**         |
+| 🎯 **Smart 3-Tier Fallback**                                                      | Auto-route: Subscription → Cheap → Free                                                  | Never stop coding, zero downtime                  |
+| 📊 **Real-Time Quota Tracking**                                                   | Live token count + reset countdown                                                       | Maximize subscription value                       |
+| 🔄 **Format Translation**                                                         | OpenAI ↔ Claude ↔ Gemini ↔ Cursor ↔ Kiro ↔ Vertex                                        | Works with any CLI tool                           |
+| 👥 **Multi-Account Support**                                                      | Multiple accounts per provider                                                           | Load balancing + redundancy                       |
+| 🔄 **Auto Token Refresh**                                                         | OAuth tokens refresh automatically                                                       | No manual re-login needed                         |
+| 🎨 **Custom Combos**                                                              | Create unlimited model combinations                                                      | Tailor fallback to your needs                     |
+| 📝 **Request Logging**                                                            | Debug mode with full request/response logs                                               | Troubleshoot issues easily                        |
+| 💾 **Cloud Sync**                                                                 | Sync config across devices                                                               | Same setup everywhere                             |
+| 📊 **Usage Analytics**                                                            | Track tokens, cost, trends over time                                                     | Optimize spending                                 |
+| 🌐 **Deploy Anywhere**                                                            | Localhost, VPS, Docker, Cloudflare Workers                                               | Flexible deployment options                       |
+
+Set `X-9Router-Token-Saver: off` to bypass all token savers for one chat request.

 <details>
 <summary><b>📖 Feature Details</b></summary>
@@ -470,7 +513,7 @@ If Headroom is down or returns an error, 9Router fails open and sends the origin

 ### 🐴 Ponytail (Lazy Senior Dev)

-Ponytail injects a *"lazy senior dev"* system prompt into every request, biasing the LLM toward minimal, YAGNI-first code — deletion over addition, stdlib over new deps, one-liners over abstractions. Adapted from [DietrichGebert/ponytail](https://github.com/DietrichGebert/ponytail).
+Ponytail injects a _"lazy senior dev"_ system prompt into every request, biasing the LLM toward minimal, YAGNI-first code — deletion over addition, stdlib over new deps, one-liners over abstractions. Adapted from [DietrichGebert/ponytail](https://github.com/DietrichGebert/ponytail).

 - **Lite** — Build what's asked, name the lazier alternative.
 - **Full** — YAGNI ladder enforced: stdlib → native → existing deps → one-liner → minimal code.
@@ -506,6 +549,7 @@ Combo: "my-coding-stack"
 ### 🔄 Format Translation

 Seamless translation between formats:
+
 - **OpenAI** ↔ **Claude** ↔ **Gemini** ↔ **Cursor** ↔ **Kiro** ↔ **Vertex** ↔ **Antigravity** ↔ **Ollama** ↔ **OpenAI Responses**
 - Your CLI tool sends OpenAI format → 9Router translates → Provider receives native format
 - Works with any tool that supports custom OpenAI endpoints
@@ -559,14 +603,14 @@ Seamless translation between formats:
 - Optimize your AI spending

 > **💡 IMPORTANT - Understanding Dashboard Costs:**
-> 
-> The "cost" displayed in Usage Analytics is **for tracking and comparison purposes only**. 
+>
+> The "cost" displayed in Usage Analytics is **for tracking and comparison purposes only**.
 > 9Router itself **never charges** you anything. You only pay providers directly (if using paid services).
-> 
-> **Example:** If your dashboard shows "$290 total cost" while using iFlow models, this represents 
-> what you would have paid using paid APIs directly. Your actual cost = **$0** (iFlow is free unlimited).
-> 
-> Think of it as a "savings tracker" showing how much you're saving by using free models or 
+>
+> **Example:** If your dashboard shows "$290 total cost" while using Kiro free models, this represents
+> what you would have paid using paid APIs directly. Your actual cost = **$0** (Kiro free tier: ~50 credits/mo).
+>
+> Think of it as a "savings tracker" showing how much you're saving by using free models or
 > routing through 9Router!

 ### 🌐 Deploy Anywhere
@@ -582,19 +626,19 @@ Seamless translation between formats:

 ## 💰 Pricing at a Glance

-| Tier | Provider | Cost | Quota Reset | Best For |
-|------|----------|------|-------------|----------|
-| **🚀 TOKEN SAVER** | **RTK (built-in)** | **FREE** | Always on | **Save 20-40% tokens on EVERY request** |
-| **💳 SUBSCRIPTION** | Claude Code (Pro/Max) | $20-200/mo | 5h + weekly | Already subscribed |
-| | Codex (Plus/Pro) | $20-200/mo | 5h + weekly | OpenAI users |
-| | GitHub Copilot | $10-19/mo | Monthly | GitHub users |
-| | Cursor IDE | $20/mo | Monthly | Cursor users |
-| **💰 CHEAP** | GLM-5.1 / GLM-4.7 | $0.6/1M | Daily 10AM | Budget backup |
-| | MiniMax M2.7 | $0.2/1M | 5-hour rolling | Cheapest option |
-| | Kimi K2.5 | $9/mo flat | 10M tokens/mo | Predictable cost |
-| **🆓 FREE** | Kiro AI | $0 | Unlimited | Claude 4.5 + GLM-5 + MiniMax free |
-| | OpenCode Free | $0 | Unlimited | No auth, auto-fetch models |
-| | Vertex AI | $300 credits | New GCP accounts | Gemini 3 Pro + DeepSeek + GLM-5 |
+| Tier                | Provider              | Cost         | Quota Reset      | Best For                                |
+| ------------------- | --------------------- | ------------ | ---------------- | --------------------------------------- |
+| **🚀 TOKEN SAVER**  | **RTK (built-in)**    | **FREE**     | Always on        | **Save 20-40% tokens on EVERY request** |
+| **💳 SUBSCRIPTION** | Claude Code (Pro/Max) | $20-200/mo   | 5h + weekly      | Already subscribed                      |
+|                     | Codex (Plus/Pro)      | $20-200/mo   | 5h + weekly      | OpenAI users                            |
+|                     | GitHub Copilot        | $10-19/mo    | Monthly          | GitHub users                            |
+|                     | Cursor IDE            | $20/mo       | Monthly          | Cursor users                            |
+| **💰 CHEAP**        | GLM-5.1 / GLM-4.7     | $0.6/1M      | Daily 10AM       | Budget backup                           |
+|                     | MiniMax M2.7          | $0.2/1M      | 5-hour rolling   | Cheapest option                         |
+|                     | Kimi K2.5             | $9/mo flat   | 10M tokens/mo    | Predictable cost                        |
+ | **🆓 FREE**         | Kiro AI               | $0           | 50 credits/mo    | Claude 4.5 + GLM-5 + MiniMax free (paid tiers above) |
+ |                     | OpenCode Free         | $0           | Varies*          | No auth, auto-fetch models (list changes over time) |
+ |                     | Vertex AI             | $300 credits | New GCP accounts | Gemini 3 Pro + DeepSeek + GLM-5 (use Vertex AI Studio endpoint for free credits) |

 **💡 Pro Tip:** RTK + Kiro AI + OpenCode Free combo = **$0 cost + 20-40% token savings**!

@@ -607,7 +651,7 @@ Seamless translation between formats:
 ✅ **9Router software = FREE forever** (open source, never charges)  
 ✅ **Dashboard "costs" = Display/tracking only** (not actual bills)  
 ✅ **You pay providers directly** (subscriptions or API fees)  
-✅ **FREE providers stay FREE** (iFlow, Kiro, Qwen = $0 unlimited)  
+✅ **FREE providers stay FREE** (Kiro ~50 credits/mo, OpenCode Free, Vertex $300 credits = $0 within free-tier limits) — note iFlow/Qwen/Gemini CLI free tiers were discontinued in 2026
 ❌ **9Router never sends invoices** or charges your card

 **How Cost Display Works:**
@@ -615,6 +659,7 @@ Seamless translation between formats:
 The dashboard shows **estimated costs** as if you were using paid APIs directly. This is **not billing** - it's a comparison tool to show your savings.

 **Example Scenario:**
+
 ```
 Dashboard Display:
 • Total Requests: 1,662
@@ -622,12 +667,13 @@ Dashboard Display:
 • Display Cost: $290

 Reality Check:
-• Provider: iFlow (FREE unlimited)
+• Provider: Kiro (free tier: ~50 credits/mo)
 • Actual Payment: $0.00
 • What $290 Means: Amount you SAVED by using free models!
 ```

 **Payment Rules:**
+
 - **Subscription providers** (Claude Code, Codex): Pay them directly via their websites
 - **Cheap providers** (GLM, MiniMax): Pay them directly, 9Router just routes
 - **FREE providers** (iFlow, Kiro, Qwen): Genuinely free forever, no hidden charges
@@ -642,6 +688,7 @@ Reality Check:
 **Problem:** Quota expires unused, rate limits during heavy coding

 **Solution:**
+
 ```
 Combo: "maximize-claude"
  1. cc/claude-opus-4-7        (use subscription fully)
@@ -657,9 +704,10 @@ vs. $20 + hitting limits = frustration
 **Problem:** Can't afford subscriptions, need reliable AI coding

 **Solution:**
+
 ```
 Combo: "free-forever"
-  1. kr/claude-sonnet-4.5      (Claude 4.5 free unlimited)
+  1. kr/claude-sonnet-4.5      (Claude 4.5 free via Kiro, ~50 credits/mo)
  2. kr/glm-5                  (GLM-5 free via Kiro)
  3. oc/<auto>                 (OpenCode Free, no auth)

@@ -672,13 +720,14 @@ Quality: Production-ready models + RTK saves 20-40% tokens
 **Problem:** Deadlines, can't afford downtime

 **Solution:**
+
 ```
 Combo: "always-on"
  1. cc/claude-opus-4-7        (best quality)
  2. cx/gpt-5.5                (second subscription)
  3. glm/glm-5.1               (cheap, resets daily)
  4. minimax/MiniMax-M2.7      (cheapest, 5h reset)
-  5. kr/claude-sonnet-4.5      (free unlimited)
+  5. kr/claude-sonnet-4.5      (free via Kiro, ~50 credits/mo)

 Result: 5 layers of fallback = zero downtime
 Monthly cost: $20-200 (subscriptions) + $10-20 (backup)
@@ -689,6 +738,7 @@ Monthly cost: $20-200 (subscriptions) + $10-20 (backup)
 **Problem:** Need AI assistant in messaging apps (WhatsApp, Telegram, Slack...), completely free

 **Solution:**
+
 ```
 Combo: "openclaw-free"
  1. kr/claude-sonnet-4.5      (Claude 4.5 free)
@@ -709,8 +759,9 @@ Access via: WhatsApp, Telegram, Slack, Discord, iMessage, Signal...
 The dashboard tracks your token usage and displays **estimated costs** as if you were using paid APIs directly. This is **not actual billing** - it's a reference to show how much you're saving by using free models or existing subscriptions through 9Router.

 **Example:**
+
 - **Dashboard shows:** "$290 total cost"
- **Reality:** You're using iFlow (FREE unlimited)
+- **Reality:** You're using Kiro free models (~50 credits/mo)
 - **Your actual cost:** **$0.00**
 - **What $290 means:** Amount you **saved** by using free models instead of paid APIs!

@@ -724,6 +775,7 @@ The cost display is a "savings tracker" to help you understand your usage patter
 **No.** 9Router is free, open-source software that runs on your own computer. It never charges you anything.

 **You only pay:**
+
 - ✅ **Subscription providers** (Claude Code $20/mo, Codex $20-200/mo) → Pay them directly on their websites
 - ✅ **Cheap providers** (GLM, MiniMax) → Pay them directly, 9Router just routes your requests
 - ❌ **9Router itself** → **Never charges anything, ever**
@@ -735,19 +787,21 @@ The cost display is a "savings tracker" to help you understand your usage patter
 <details>
 <summary><b>🆓 Are FREE providers really unlimited?</b></summary>

-**Yes!** The current FREE providers (Kiro, OpenCode Free, Vertex) are genuinely free with **no hidden charges**.
+**Mostly!** The current FREE providers (Kiro, OpenCode Free, Vertex) are genuinely free, but free tiers have limits:

 These are free services offered by those respective companies:
- **Kiro AI**: Free unlimited Claude 4.5 + GLM-5 + MiniMax via AWS Builder ID / Google / GitHub OAuth
- **OpenCode Free**: No-auth passthrough proxy, models auto-fetched from `opencode.ai/zen/v1/models`
- **Vertex AI**: $300 free credits for new Google Cloud accounts (90 days)

-9Router just routes your requests to them - there's no "catch" or future billing. They're truly free services, and 9Router makes them easy to use with fallback support.
+- **Kiro AI**: ~50 credits/month free (plus 500 trial credits for new accounts in the first 30 days) via AWS Builder ID / Google / GitHub OAuth. Paid tiers available above that.
+- **OpenCode Free**: No-auth passthrough proxy, models auto-fetched from `opencode.ai/zen/v1/models`. The free model list fluctuates over time (some models free only for limited promos) — subject to change without notice.
+- **Vertex AI**: $300 free credits for new Google Cloud accounts (90 days). Since Mar 2026 the Gemini API endpoint no longer consumes these credits — use the **Vertex AI Studio** endpoint instead.
+
+9Router just routes your requests to them - there's no "catch" or future billing from 9Router itself. They're truly free services, and 9Router makes them easy to use with fallback support.

 **Discontinued free tiers (no longer recommended):**
+
 - ❌ **iFlow**: Was free unlimited, now changed to paid (2026)
- ❌ **Qwen Code**: Free OAuth tier discontinued by Alibaba on 2026-04-15
- ❌ **Gemini CLI**: Still works, but using it with non-CLI tools (Claude, Codex, Cursor...) may result in account bans — only use if you stick to Gemini CLI itself
+- ❌ **Qwen Code**: Free OAuth tier fully discontinued by Alibaba on 2026-04-15
+- ❌ **Gemini CLI**: Service fully shut down by Google on 2026-06-18 (replaced by the closed-source Antigravity CLI). Discontinued — do not use.

 </details>

@@ -757,17 +811,21 @@ These are free services offered by those respective companies:
 **Free-First Strategy:**

 1. **Start with 100% free combo:**
+
   ```
-   1. gc/gemini-3-flash (180K/month free from Google)
-   2. if/kimi-k2-thinking (unlimited free from iFlow)
-   3. qw/qwen3-coder-plus (unlimited free from Qwen)
+   1. kr/glm-5 (GLM-5 free via Kiro, ~50 credits/mo)
+   2. OpenCode Free models (no auth, auto-fetched)
+   3. Vertex AI Gemini 3 Pro (using the Vertex AI Studio endpoint with $300 credits)
   ```
-   **Cost: $0/month**
+
+   **Cost: $0/month** (within Kiro's free credit cap; OpenCode/Vertex subject to their free-tier limits)

 2. **Add cheap backup** only if you need it:
+
   ```
   4. glm/glm-4.7 ($0.6/1M tokens)
   ```
+
   **Additional cost: Only pay for what you actually use**

 3. **Use subscription providers last:**
@@ -786,10 +844,12 @@ These are free services offered by those respective companies:
 **Scenario:** You're on a coding sprint and blow through your quotas

 **Without 9Router:**
+
 - ❌ Hit rate limit → Work stops → Frustration
 - ❌ Or: Accidentally rack up huge API bills

 **With 9Router:**
+
 - ✅ Subscription hits limit → Auto-fallback to cheap tier
 - ✅ Cheap tier gets expensive → Auto-fallback to free tier
 - ✅ Never stop coding → Predictable costs
@@ -983,7 +1043,7 @@ Monthly cost example (100M tokens):
 ```
 Name: free-combo
 Models:
-  1. kr/claude-sonnet-4.5 (Claude 4.5 free unlimited)
+  1. kr/claude-sonnet-4.5 (Claude 4.5 free via Kiro, ~50 credits/mo)
  2. kr/glm-5 (GLM-5 free via Kiro)
  3. vertex/gemini-3.1-pro-preview ($300 free credits)

@@ -1113,6 +1173,7 @@ pm2 startup
 ### Docker

 Published images (multi-platform `linux/amd64` + `linux/arm64`):
+
 - Docker Hub: [`decolua/9router`](https://hub.docker.com/r/decolua/9router)
 - GHCR: [`ghcr.io/decolua/9router`](https://github.com/decolua/9router/pkgs/container/9router)

@@ -1140,6 +1201,7 @@ docker run -d --name 9router -p 20128:20128 \
 ```

 **Container defaults:**
+
 - `PORT=20128`
 - `HOSTNAME=0.0.0.0`

@@ -1156,26 +1218,28 @@ docker pull decolua/9router:latest   # update to latest

 ### Environment Variables

-| Variable | Default | Description |
-|----------|---------|-------------|
-| `JWT_SECRET` | Auto-generated (`~/.9router/jwt-secret`) | JWT signing secret for dashboard auth cookie (override to share across instances) |
-| `INITIAL_PASSWORD` | `123456` | First login password when no saved hash exists |
-| `DATA_DIR` | `~/.9router` | Main app data location (SQLite at `$DATA_DIR/db/data.sqlite`) |
-| `PORT` | framework default | Service port (`20128` in examples) |
-| `HOSTNAME` | framework default | Bind host (Docker defaults to `0.0.0.0`) |
-| `NODE_ENV` | runtime default | Set `production` for deploy |
-| `BASE_URL` | `http://localhost:20128` | Server-side internal base URL used by cloud sync jobs |
-| `CLOUD_URL` | `https://9router.com` | Server-side cloud sync endpoint base URL |
-| `NEXT_PUBLIC_BASE_URL` | `http://localhost:3000` | Backward-compatible/public base URL (prefer `BASE_URL` for server runtime) |
-| `NEXT_PUBLIC_CLOUD_URL` | `https://9router.com` | Backward-compatible/public cloud URL (prefer `CLOUD_URL` for server runtime) |
-| `API_KEY_SECRET` | `endpoint-proxy-api-key-secret` | HMAC secret for generated API keys |
-| `MACHINE_ID_SALT` | `endpoint-proxy-salt` | Salt for stable machine ID hashing |
-| `ENABLE_REQUEST_LOGS` | `false` | Enables request/response logs under `logs/` |
-| `AUTH_COOKIE_SECURE` | `false` | Force `Secure` auth cookie (set `true` behind HTTPS reverse proxy) |
-| `REQUIRE_API_KEY` | `false` | Enforce Bearer API key on `/v1/*` routes (recommended for internet-exposed deploys) |
-| `HTTP_PROXY`, `HTTPS_PROXY`, `ALL_PROXY`, `NO_PROXY` | empty | Optional outbound proxy for upstream provider calls |
+| Variable                                             | Default                                  | Description                                                                         |
+| ---------------------------------------------------- | ---------------------------------------- | ----------------------------------------------------------------------------------- |
+| `JWT_SECRET`                                         | Auto-generated (`~/.9router/jwt-secret`) | JWT signing secret for dashboard auth cookie (override to share across instances)   |
+| `INITIAL_PASSWORD`                                   | `123456`                                 | First login password when no saved hash exists                                      |
+| `DATA_DIR`                                           | `~/.9router`                             | Main app data location (SQLite at `$DATA_DIR/db/data.sqlite`)                       |
+| `PORT`                                               | framework default                        | Service port (`20128` in examples)                                                  |
+| `HOSTNAME`                                           | framework default                        | Bind host (Docker defaults to `0.0.0.0`)                                            |
+| `NODE_ENV`                                           | runtime default                          | Set `production` for deploy                                                         |
+| `BASE_URL`                                           | `http://localhost:20128`                 | Server-side internal base URL used by cloud sync jobs                               |
+| `CLOUD_URL`                                          | `https://9router.com`                    | Server-side cloud sync endpoint base URL                                            |
+| `NEXT_PUBLIC_BASE_URL`                               | `http://localhost:3000`                  | Backward-compatible/public base URL (prefer `BASE_URL` for server runtime)          |
+| `NEXT_PUBLIC_CLOUD_URL`                              | `https://9router.com`                    | Backward-compatible/public cloud URL (prefer `CLOUD_URL` for server runtime)        |
+| `API_KEY_SECRET`                                     | `endpoint-proxy-api-key-secret`          | HMAC secret for generated API keys                                                  |
+| `MACHINE_ID_SALT`                                    | `endpoint-proxy-salt`                    | Salt for stable machine ID hashing                                                  |
+| `ENABLE_REQUEST_LOGS`                                | `false`                                  | Enables request/response logs under `logs/`                                         |
+| `AUTH_COOKIE_SECURE`                                 | `false`                                  | Force `Secure` auth cookie (set `true` behind HTTPS reverse proxy)                  |
+| `REQUIRE_API_KEY`                                    | `false`                                  | Enforce Bearer API key on `/v1/*` routes (recommended for internet-exposed deploys) |
+| `HTTP_PROXY`, `HTTPS_PROXY`, `ALL_PROXY`, `NO_PROXY` | empty                                    | Optional outbound proxy for upstream provider calls                                 |
+| `SEARXNG_URL`                                        | `http://localhost:8888/search`           | Endpoint for the built-in unauthenticated SearXNG web-search provider               |

 Notes:
+
 - Lowercase proxy variables are also supported: `http_proxy`, `https_proxy`, `all_proxy`, `no_proxy`.
 - `.env` is not baked into Docker image (`.dockerignore`); inject runtime config with `--env-file` or `-e`.
 - On Windows, `APPDATA` can be used for local storage path resolution.
@@ -1198,6 +1262,7 @@ Notes:
 <summary><b>View all available models</b></summary>

 **Claude Code (`cc/`)** - Pro/Max:
+
 - `cc/claude-opus-4-7`
 - `cc/claude-opus-4-6`
 - `cc/claude-sonnet-4-6`
@@ -1205,6 +1270,7 @@ Notes:
 - `cc/claude-haiku-4-5-20251001`

 **Codex (`cx/`)** - Plus/Pro:
+
 - `cx/gpt-5.5`
 - `cx/gpt-5.4`
 - `cx/gpt-5.3-codex`
@@ -1212,6 +1278,7 @@ Notes:
 - `cx/gpt-5.1-codex-max`

 **GitHub Copilot (`gh/`)**:
+
 - `gh/gpt-5.4`
 - `gh/claude-opus-4.7`
 - `gh/claude-sonnet-4.6`
@@ -1219,25 +1286,30 @@ Notes:
 - `gh/grok-code-fast-1`

 **Cursor (`cu/`)** - Subscription:
+
 - `cu/claude-4.6-opus-max`
 - `cu/claude-4.5-sonnet-thinking`
 - `cu/gpt-5.3-codex`
 - `cu/kimi-k2.5`

 **GLM (`glm/`)** - $0.6/1M:
+
 - `glm/glm-5.1`
 - `glm/glm-5`
 - `glm/glm-4.7`

 **MiniMax (`minimax/`)** - $0.2/1M:
+
 - `minimax/MiniMax-M2.7`
 - `minimax/MiniMax-M2.5`

 **Kimi (`kimi/`)** - $9/mo flat:
+
 - `kimi/kimi-k2.5`
 - `kimi/kimi-k2.5-thinking`

-**Kiro (`kr/`)** - FREE unlimited:
+**Kiro (`kr/`)** - Free (~50 credits/month, paid tiers above):
+
 - `kr/claude-sonnet-4.5`
 - `kr/claude-haiku-4.5`
 - `kr/glm-5`
@@ -1246,9 +1318,11 @@ Notes:
 - `kr/deepseek-3.2`

 **OpenCode Free (`oc/`)** - FREE no-auth:
+
 - Auto-fetched from `opencode.ai/zen/v1/models`

 **Vertex AI (`vertex/`)** - $300 free credits:
+
 - `vertex/gemini-3.1-pro-preview`
 - `vertex/gemini-3-flash-preview`
 - `vertex/gemini-2.5-flash`
@@ -1262,31 +1336,38 @@ Notes:
 ## 🐛 Troubleshooting

 **"Language model did not provide messages"**
+
 - Provider quota exhausted → Check dashboard quota tracker
 - Solution: Use combo fallback or switch to cheaper tier

 **Rate limiting**
+
 - Subscription quota out → Fallback to GLM/MiniMax
 - Add combo: `cc/claude-opus-4-7 → glm/glm-5.1 → kr/claude-sonnet-4.5`

 **OAuth token expired**
+
 - Auto-refreshed by 9Router
 - If issues persist: Dashboard → Provider → Reconnect

 **High costs**
+
 - Enable RTK in Dashboard → Endpoint settings (default ON, saves 20-40% tokens)
 - Check usage stats in Dashboard
 - Switch primary model to GLM/MiniMax
 - Use free tier (Kiro, OpenCode Free, Vertex) for non-critical tasks

 **Dashboard opens on wrong port**
+
 - Set `PORT=20128` and `NEXT_PUBLIC_BASE_URL=http://localhost:20128`

 **First login not working**
+
 - Check `INITIAL_PASSWORD` in `.env`
 - If unset, fallback password is `123456`

 **No request logs under `logs/`**
+
 - Set `ENABLE_REQUEST_LOGS=true`

 ---
@@ -1349,8 +1430,6 @@ Thanks to all contributors who helped make 9Router better!

 [![Star Chart](https://starchart.cc/decolua/9router.svg?variant=adaptive)](https://starchart.cc/decolua/9router)

-
-
 ## 🔀 Forks

 **[OmniRoute](https://github.com/diegosouzapw/OmniRoute)** — A full-featured TypeScript fork of 9Router. Adds 36+ providers, 4-tier auto-fallback, multi-modal APIs (images, embeddings, audio, TTS), circuit breaker, semantic cache, LLM evaluations, and a polished dashboard. 368+ unit tests. Available via npm and Docker.
@@ -1363,8 +1442,8 @@ Built on the shoulders of giants:

 - **[CLIProxyAPI](https://github.com/router-for-me/CLIProxyAPI)** — original Go implementation that inspired this JavaScript port.
 - **[RTK](https://github.com/rtk-ai/rtk)** ![Stars](https://img.shields.io/github/stars/rtk-ai/rtk?style=flat&color=yellow) — Rust token-saver. 9Router ports its compression pipeline to JS → **−20-40% input tokens** on every request.
- **[Caveman](https://github.com/JuliusBrussee/caveman)** ![Stars](https://img.shields.io/github/stars/JuliusBrussee/caveman?style=flat&color=yellow) by **[@JuliusBrussee](https://github.com/JuliusBrussee)** — viral *"why use many token when few token do trick"*. 9Router adapts its prompt → **−65% output tokens**.
- **[Ponytail](https://github.com/DietrichGebert/ponytail)** ![Stars](https://img.shields.io/github/stars/DietrichGebert/ponytail?style=flat&color=yellow) by **[@DietrichGebert](https://github.com/DietrichGebert)** — *"lazy senior dev"* skill. 9Router injects its YAGNI-first ladder → **fewer tokens, less code, shorter diffs**.
+- **[Caveman](https://github.com/JuliusBrussee/caveman)** ![Stars](https://img.shields.io/github/stars/JuliusBrussee/caveman?style=flat&color=yellow) by **[@JuliusBrussee](https://github.com/JuliusBrussee)** — viral _"why use many token when few token do trick"_. 9Router adapts its prompt → **−65% output tokens**.
+- **[Ponytail](https://github.com/DietrichGebert/ponytail)** ![Stars](https://img.shields.io/github/stars/DietrichGebert/ponytail?style=flat&color=yellow) by **[@DietrichGebert](https://github.com/DietrichGebert)** — _"lazy senior dev"_ skill. 9Router injects its YAGNI-first ladder → **fewer tokens, less code, shorter diffs**.

 Huge thanks to these authors — without their work, 9Router's token-saving features wouldn't exist. ⭐ them on GitHub!

--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -82,7 +82,7 @@ npm install -g 9router

 **2. 连接免费提供商（无需注册）：**

-控制面板 → 提供商 → 连接 **Kiro AI**（免费 Claude 无限量）或 **OpenCode Free**（无需认证）→ 完成！
+控制面板 → 提供商 → 连接 **Kiro AI**（约 50 积分/月免费：Claude 4.5 + GLM-5 + MiniMax）或 **OpenCode Free**（无需认证）→ 完成！

 **3. 在 CLI 工具中使用：**

@@ -279,12 +279,12 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
      <td align="center" width="150">
        <img src="./public/providers/kiro.png" width="70" alt="Kiro"/><br/>
        <b>Kiro AI</b><br/>
-        <sub>Claude 4.5 + GLM-5 + MiniMax<br/>无限免费</sub>
+        <sub>Claude 4.5 + GLM-5 + MiniMax<br/>每月 50 积分免费</sub>
      </td>
      <td align="center" width="150">
        <img src="./public/providers/opencode.png" width="70" alt="OpenCode Free"/><br/>
        <b>OpenCode Free</b><br/>
-        <sub>无需认证 • 自动获取模型<br/>无限免费</sub>
+        <sub>无需认证 • 自动获取模型<br/>免费（模型列表会变）</sub>
      </td>
      <td align="center" width="150">
        <img src="./public/providers/gemini.png" width="70" alt="Vertex AI"/><br/>
@@ -295,7 +295,11 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
  </table>
 </div>

-> **注意：** iFlow、Qwen 和 Gemini CLI 的免费等级已于 2026 年停止。请改用 Kiro / OpenCode Free / Vertex。
+> **注意：** iFlow、Qwen Code 和 Gemini CLI 的免费等级已于 2026 年停止。请改用 Kiro / OpenCode Free / Vertex。
+>
+> **Kiro AI** 于 2025 年 9 月转为付费模式 — 免费等级现在上限为**每月 50 积分**（新账户前 30 天另加 500 试用积分）。付费档位：Pro $20/月（1,000 积分）、Pro+ $40/月（2,000）、Pro Max $100/月（5,000）、Power $200/月（10,000）。
+> **OpenCode Free** 的模型列表会随时间变化（部分模型仅限时免费）— 可能随时变更，恕不另行通知。
+> **Vertex AI**：新 GCP 账户的 $300 免费额度仍然有效，但自 2026 年 3 月起 **Gemini API 端点不再消耗这些额度** — 请改用 **Vertex AI Studio** 端点。

 ### 🔑 API Key 提供商（40+）

@@ -500,7 +504,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 > 使用分析中显示的"成本"**仅用于追踪和比较目的**。
 > 9Router 本身**永远不会向你收费**。你只直接向提供商付款（如果使用付费服务）。
 > 
-> **示例：** 如果你的控制面板显示使用 iFlow 模型时"总成本 $290"，这代表你如果直接使用付费 API 需要支付的金额。你的实际成本 = **$0**（iFlow 免费无限量）。
+> **示例：** 如果你的控制面板显示使用 Kiro 免费模型时"总成本 $290"，这代表你如果直接使用付费 API 需要支付的金额。你的实际成本 = **$0**（Kiro 免费等级：约 50 积分/月）。
 > 
 > 把它想象成一个"节省追踪器"，展示你通过使用免费模型或通过 9Router 路由节省了多少钱！

@@ -527,9 +531,9 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 | **💰 低价** | GLM-5.1 / GLM-4.7 | $0.6/1M | 每日 10AM | 预算备份 |
 | | MiniMax M2.7 | $0.2/1M | 5小时滚动 | 最便宜选项 |
 | | Kimi K2.5 | $9/月固定 | 10M tokens/月 | 可预测成本 |
-| **🆓 免费** | Kiro AI | $0 | 无限量 | Claude 4.5 + GLM-5 + MiniMax 免费 |
-| | OpenCode Free | $0 | 无限量 | 无需认证，自动获取模型 |
-| | Vertex AI | $300 额度 | 新 GCP 账户 | Gemini 3 Pro + DeepSeek + GLM-5 |
+ | **🆓 免费** | Kiro AI | $0 | 50 积分/月 | Claude 4.5 + GLM-5 + MiniMax 免费（之上为付费档位） |
+ | | OpenCode Free | $0 |  varies* | 无需认证，自动获取模型（列表会变化） |
+ | | Vertex AI | $300 额度 | 新 GCP 账户 | Gemini 3 Pro + DeepSeek + GLM-5（使用 Vertex AI Studio 端点消耗免费额度） |

 **💡 专业提示：** RTK + Kiro AI + OpenCode Free 组合 = **$0 成本 + 节省 20-40% tokens**！

@@ -542,7 +546,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 ✅ **9Router 软件 = 永久免费**（开源，绝不收费）  
 ✅ **控制面板"成本" = 仅用于显示/追踪**（不是实际账单）  
 ✅ **你直接向提供商付款**（订阅或 API 费用）  
-✅ **免费提供商保持免费**（iFlow、Kiro、Qwen = $0 无限量）  
+✅ **免费提供商保持免费**（Kiro 约 50 积分/月、OpenCode Free、Vertex $300 额度 = 在免费额度内 $0）— 注意 iFlow/Qwen/Gemini CLI 免费等级已于 2026 年停止
 ❌ **9Router 永不发送发票** 或扣款

 **成本显示如何工作：**
@@ -557,7 +561,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 • 显示成本：$290

 实际检查：
-• 提供商：iFlow（免费无限量）
+• 提供商：Kiro（免费等级：约 50 积分/月）
 • 实际支付：$0.00
 • $290 意味着什么：通过使用免费模型节省的金额！
 ```
@@ -565,7 +569,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 **付款规则：**
 - **订阅提供商**（Claude Code、Codex）：通过他们的网站直接付款
 - **低价提供商**（GLM、MiniMax）：直接付款，9Router 只做路由
- **免费提供商**（iFlow、Kiro、Qwen）：真正的永久免费，无隐藏费用
+- **免费提供商**（Kiro、OpenCode Free、Vertex）：真正的免费，在免费额度内无隐藏费用
 - **9Router**：从不收取任何费用，永远不会

 ---
@@ -594,7 +598,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 **解决方案：**
 ```
 组合："free-forever"
-  1. kr/claude-sonnet-4.5      （Claude 4.5 免费无限量）
+  1. kr/claude-sonnet-4.5      （通过 Kiro 免费使用 Claude 4.5，约 50 积分/月）
  2. kr/glm-5                  （通过 Kiro 免费使用 GLM-5）
  3. oc/<auto>                 （OpenCode Free，无需认证）

@@ -613,7 +617,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
  2. cx/gpt-5.5                （第二个订阅）
  3. glm/glm-5.1               （低价，每日重置）
  4. minimax/MiniMax-M2.7      （最便宜，5小时重置）
-  5. kr/claude-sonnet-4.5      （免费无限量）
+  5. kr/claude-sonnet-4.5      （通过 Kiro 免费使用，约 50 积分/月）

 结果：5 层切换 = 零停机时间
 月成本：$20-200（订阅）+ $10-20（备份）
@@ -645,7 +649,7 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run

 **示例：**
 - **控制面板显示：** "$290 总成本"
- **实际情况：** 你在使用 iFlow（免费无限量）
+- **实际情况：** 你在使用 Kiro 免费模型（约 50 积分/月）
 - **你的实际成本：** **$0.00**
 - **$290 的含义：** 你通过使用免费模型而不是付费 API **节省**的金额！

@@ -670,19 +674,19 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run
 <details>
 <summary><b>🆓 免费提供商真的是无限量的吗？</b></summary>

-**是的！** 当前的免费提供商（Kiro、OpenCode Free、Vertex）是真正的免费，**无隐藏费用**。
+**基本上是！** 当前的免费提供商（Kiro、OpenCode Free、Vertex）是真正的免费，但免费等级有上限：

 这些是各公司提供的免费服务：
- **Kiro AI**：通过 AWS Builder ID / Google / GitHub OAuth 免费无限量使用 Claude 4.5 + GLM-5 + MiniMax
- **OpenCode Free**：无认证直连代理，模型从 `opencode.ai/zen/v1/models` 自动获取
- **Vertex AI**：新 Google Cloud 账户可获得 $300 免费额度（90 天）
+- **Kiro AI**：通过 AWS Builder ID / Google / GitHub OAuth 使用，免费等级约**每月 50 积分**（新账户前 30 天另加 500 试用积分）。之上提供付费档位。
+- **OpenCode Free**：无认证直连代理，模型从 `opencode.ai/zen/v1/models` 自动获取。免费模型列表会随时间变化（部分模型仅限时免费）— 可能随时变更。
+- **Vertex AI**：新 Google Cloud 账户可获得 $300 免费额度（90 天）。自 2026 年 3 月起 Gemini API 端点不再消耗这些额度 — 请改用 **Vertex AI Studio** 端点。

 9Router 只是路由你的请求到它们 — 没有"陷阱"或未来的计费。它们是真正的免费服务，9Router 让它们易于使用并支持切换。

 **已停止的免费等级（不再推荐）：**
 - ❌ **iFlow**：曾是免费无限量，现在改为付费（2026）
- ❌ **Qwen Code**：阿里巴巴于 2026-04-15 停止免费 OAuth 等级
- ❌ **Gemini CLI**：仍可用，但与非 CLI 工具（Claude、Codex、Cursor...）一起使用可能会导致账户被封 — 仅在你坚持使用 Gemini CLI 本身时才使用
+- ❌ **Qwen Code**：阿里巴巴于 2026-04-15 完全停止免费 OAuth 等级
+- ❌ **Gemini CLI**：Google 已于 2026-06-18 完全停止服务（由闭源的 Antigravity CLI 取代）。已停止 — 请勿使用。

 </details>

@@ -693,11 +697,11 @@ PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run

 1. **从 100% 免费组合开始：**
   ```
-   1. gc/gemini-3-flash (Google 每月 180K 免费)
-   2. if/kimi-k2-thinking (iFlow 无限量免费)
-   3. qw/qwen3-coder-plus (Qwen 无限量免费)
+   1. kr/glm-5 (通过 Kiro 免费使用 GLM-5，约 50 积分/月)
+   2. OpenCode Free 模型（无认证，自动获取）
+   3. Vertex AI Gemini 3 Pro（使用 Vertex AI Studio 端点 + $300 额度）
   ```
-   **成本：$0/月**
+   **成本：$0/月**（在 Kiro 免费积分上限内；OpenCode/Vertex 受各自免费等级限制）

 2. **仅在需要时添加低价备份：**
   ```
@@ -918,7 +922,7 @@ Vertex 合作伙伴（通过 Vertex 提供 Anthropic / DeepSeek / GLM / Qwen）
 ```
 名称：free-combo
 模型：
-  1. kr/claude-sonnet-4.5 (Claude 4.5 免费无限量)
+  1. kr/claude-sonnet-4.5 (通过 Kiro 免费使用 Claude 4.5，约 50 积分/月)
  2. kr/glm-5 (通过 Kiro 免费使用 GLM-5)
  3. vertex/gemini-3.1-pro-preview ($300 免费额度)

@@ -1168,7 +1172,7 @@ docker stop 9router && docker rm 9router
 - `kimi/kimi-k2.5`
 - `kimi/kimi-k2.5-thinking`

-**Kiro（`kr/`）** - 免费无限量：
+**Kiro（`kr/`）** - 免费（约 50 积分/月，之上为付费档位）：
 - `kr/claude-sonnet-4.5`
 - `kr/claude-haiku-4.5`
 - `kr/glm-5`
--- a/cli/cli.js
+++ b/cli/cli.js
@@ -4,8 +4,28 @@ const { spawn, exec, execSync } = require("child_process");
 const path = require("path");
 const fs = require("fs");
 const https = require("https");
+const net = require("net");
 const os = require("os");

+// Poll until the server accepts TCP connections on port, or timeout — avoids blind fixed waits.
+function waitServerReady(port, { timeoutMs = 15000, intervalMs = 150 } = {}) {
+  const deadline = Date.now() + timeoutMs;
+  return new Promise((resolve) => {
+    const tryConnect = () => {
+      const socket = net.connect({ host: "127.0.0.1", port }, () => {
+        socket.destroy();
+        resolve(true);
+      });
+      socket.on("error", () => {
+        socket.destroy();
+        if (Date.now() >= deadline) return resolve(false);
+        setTimeout(tryConnect, intervalMs);
+      });
+    };
+    tryConnect();
+  });
+}
+
 // Native spinner - no external dependency
 function createSpinner(text) {
  const frames = ["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"];
@@ -47,6 +67,19 @@ const { ensureSqliteRuntime, buildEnvWithRuntime } = require("./hooks/sqliteRunt
 const { ensureTrayRuntime } = require("./hooks/trayRuntime");
 const args = process.argv.slice(2);

+// Subcommands (`9router xai video …`) run against an already-running gateway
+// and bypass the launcher flow (no runtime self-heal, no server spawn).
+if (args[0] === "xai" && args[1] === "video") {
+  const { run } = require("./src/cli/commands/xaiVideo");
+  run(args.slice(2))
+    .then((code) => process.exit(code))
+    .catch((err) => {
+      console.error(`❌ ${err?.message || err}`);
+      process.exit(1);
+    });
+  return;
+}
+
 // Self-heal SQLite runtime deps (sql.js + better-sqlite3) into ~/.9router/runtime
 // so the server can resolve them via NODE_PATH. Best-effort — sql.js is required,
 // better-sqlite3 is optional. Logs to stderr only on failure.
@@ -119,6 +152,11 @@ Options:
  --skip-update       Skip auto-update check
  -h, --help          Show this help message
  -v, --version       Show version
+
+Commands:
+  xai video --prompt "..." --output video.mp4
+                      Generate a Grok Imagine video via the running gateway
+                      (see: ${APP_NAME} xai video --help)
 `);
    process.exit(0);
  } else if (args[i] === "--version" || args[i] === "-v") {
@@ -212,17 +250,18 @@ function killCloudflaredByAppPort(appPort) {
 function killAllAppProcesses(appPort) {
  return new Promise((resolve) => {
    try {
-      // Kill MIT first (privileged process, needs special handling)
-      killProxyByPidFile();
-      // Kill cloudflared/tailscale by PID file (precise, only this app's tunnel)
-      killTunnelByPidFile();
+      // Background: MITM + tunnel/cloudflared run on separate ports/processes —
+      // killing them doesn't free the app port, so don't block the critical path.
+      // Server-side MITM manager has stale-lock recovery and starts deferred (~3s).
+      setImmediate(() => {
+        try { killProxyByPidFile(); } catch {}
+        try { killTunnelByPidFile(); } catch {}
+        try { killCloudflaredByAppPort(appPort); } catch {}
+      });

      const platform = process.platform;
      let pids = [];

-      // Catch stale PID files: kill cloudflared bound to this app's port
-      pids.push(...killCloudflaredByAppPort(appPort));
-
      if (platform === "win32") {
        // Windows: use WMI to get full CommandLine (tasklist /V doesn't include it)
        try {
@@ -499,14 +538,11 @@ if (!fs.existsSync(serverPath)) {
  process.exit(1);
 }

-// Check for updates FIRST, then start server
-checkForUpdate().then((latestVersion) => {
-  killAllAppProcesses(port).then(() => {
-    return killProcessOnPort(port);
-  }).then(() => {
-    startServer(latestVersion);
-  });
-});
+// Start server immediately; run update check in parallel (not on the critical path).
+const updatePromise = checkForUpdate();
+killAllAppProcesses(port)
+  .then(() => killProcessOnPort(port))
+  .then(() => startServer(updatePromise));

 // Show interface selection menu
 async function showInterfaceMenu(latestVersion) {
@@ -556,7 +592,9 @@ async function showInterfaceMenu(latestVersion) {
 const MAX_RESTARTS = 2;
 const RESTART_RESET_MS = 30000; // Reset counter if alive > 30s

-function startServer(latestVersion) {
+function startServer(updatePromise) {
+  // Accept either a Promise (parallel update check) or a resolved value.
+  const latestVersionPromise = Promise.resolve(updatePromise);
  const displayHost = getDisplayHost();
  const url = `http://${displayHost}:${port}/dashboard`;
  // Surface real network exposure when bound to all interfaces (default 0.0.0.0).
@@ -574,7 +612,7 @@ function startServer(latestVersion) {
  function spawnServer() {
    serverStartTime = Date.now();
    crashLog = [];
-    const child = spawn(RUNTIME, ["--max-old-space-size=6144", serverPath], {
+    const child = spawn(RUNTIME, ["--dns-result-order=ipv4first", "--max-old-space-size=6144", serverPath], {
      cwd: standaloneDir,
      stdio: showLog ? "inherit" : ["ignore", "ignore", "pipe"],
      detached: true,
@@ -677,17 +715,19 @@ function startServer(latestVersion) {
    console.log(`\n🚀 ${pkg.name} v${pkg.version}`);
    console.log(`Server: http://${displayHost}:${port}`);

-    setTimeout(() => {
+    waitServerReady(port).then(() => {
      initTrayIcon();
      console.log("\n💡 Router is now running in system tray. Close this terminal if you want.");
      console.log("   Right-click tray icon to open dashboard or quit.\n");
-    }, 2000);
+    });

    return;
  }

  // Wait for server to be ready, then show interface menu loop + tray
-  setTimeout(async () => {
+  waitServerReady(port).then(async () => {
+    // Resolve parallel update check (already running); don't block server start on it.
+    const latestVersion = await latestVersionPromise;
    // Start tray icon alongside TUI
    initTrayIcon();

@@ -745,7 +785,7 @@ function startServer(latestVersion) {
          // Windows/Linux: spawn detached bgProcess (systray works fine in child)
          console.log(`\n⏳ Starting background process... (tray icon will appear in ~3s)`);

-          const bgProcess = spawn(process.execPath, [__filename, "--tray", "--skip-update", "-p", port.toString()], {
+          const bgProcess = spawn(process.execPath, ["--dns-result-order=ipv4first", __filename, "--tray", "--skip-update", "-p", port.toString()], {
            detached: true,
            stdio: "ignore",
            windowsHide: true,
@@ -772,7 +812,7 @@ function startServer(latestVersion) {
      cleanup();
      process.exit(1);
    }
-  }, 3000);
+  });

  function attachServerEvents() {
    server.on("error", (err) => {
--- a/cli/package.json
+++ b/cli/package.json
@@ -1,6 +1,6 @@
 {
  "name": "9router",
-  "version": "0.5.12",
+  "version": "0.5.45",
  "description": "9Router CLI - Start and manage 9Router server",
  "bin": {
    "9router": "./cli.js"
--- a/cli/scripts/build-cli.js
+++ b/cli/scripts/build-cli.js
@@ -7,7 +7,7 @@ const { execSync } = require("child_process");
 const cliDir = path.resolve(__dirname, "..");
 const appDir = path.resolve(cliDir, "..");
 const rootDir = path.resolve(appDir, "..");
-const cliAppDir = path.join(cliDir, "app");
+const cliAppDir = process.env.NINEROUTER_CLI_APP_DIR || path.join(cliDir, "app");
 const buildHomeDir = path.join(cliDir, ".build-home");
 const buildDistDirName = ".next-cli-build";
 const buildDistDir = path.join(appDir, buildDistDirName);
--- a/cli/scripts/buildMitm.js
+++ b/cli/scripts/buildMitm.js
@@ -12,7 +12,8 @@ const BUILD_CONFIG = {

 const cliDir = path.resolve(__dirname, "..");
 const appDir = path.resolve(cliDir, "..");
-const cliMitmDir = path.join(cliDir, "app", "src", "mitm");
+const cliAppDir = process.env.NINEROUTER_CLI_APP_DIR || path.join(cliDir, "app");
+const cliMitmDir = path.join(cliAppDir, "src", "mitm");
 // Bundle everything — no externals. This keeps MITM runtime self-contained so
 // it can be copied to DATA_DIR/runtime/ and spawned from there (escapes
 // node_modules file locks that block `npm i -g 9router@latest` on Windows).
--- a/cli/src/cli/commands/xaiVideo.js
+++ b/cli/src/cli/commands/xaiVideo.js
@@ -0,0 +1,300 @@
+/**
+ * `9router xai video` — generate a Grok Imagine video through the local
+ * 9router gateway and save the result as an MP4 file.
+ *
+ * Flow: POST /v1/videos/generations → poll GET /v1/videos/{request_id}
+ * until done/failed/timeout → download video.url → atomic rename.
+ *
+ * No OAuth tokens or Authorization headers are ever printed.
+ */
+
+const http = require("http");
+const https = require("https");
+const fs = require("fs");
+const path = require("path");
+
+const DEFAULT_PORT = 20128;
+const DEFAULT_HOST = "127.0.0.1";
+const DEFAULT_MODEL = "xai/grok-imagine-video";
+const DEFAULT_TIMEOUT_SEC = 600;
+const DEFAULT_POLL_INTERVAL_MS = 5000;
+
+const TERMINAL_STATUSES = new Set(["done", "failed", "completed", "error", "expired", "cancelled"]);
+const FAILED_STATUSES = new Set(["failed", "error", "expired", "cancelled"]);
+
+const HELP = `
+Usage: 9router xai video --prompt "..." [options]
+
+Generate a Grok Imagine video via your local 9router gateway
+(requires a connected xAI account — Grok Build OAuth or API key).
+
+Options:
+  --prompt <text>         Video description (required)
+  --output <file>         Output MP4 path (default: video.mp4)
+  --model <id>            Model (default: ${DEFAULT_MODEL})
+  --duration <seconds>    Video duration
+  --aspect-ratio <ratio>  e.g. 16:9, 9:16, 1:1
+  --resolution <res>      480p | 720p | 1080p
+  --image <path-or-url>   Image input for image-to-video
+  --timeout <seconds>     Max wait for the job (default: ${DEFAULT_TIMEOUT_SEC})
+  --port <port>           Gateway port (default: ${DEFAULT_PORT})
+  --host <host>           Gateway host (default: ${DEFAULT_HOST})
+  --api-key <key>         9router API key (or env NINE_ROUTER_API_KEY)
+  -h, --help              Show this help
+`;
+
+function sanitizeText(text) {
+  return String(text ?? "").replace(/Bearer\s+[A-Za-z0-9._~+/=-]{8,}/gi, "Bearer [redacted]");
+}
+
+function parseArgs(argv) {
+  const opts = {
+    model: DEFAULT_MODEL,
+    output: "video.mp4",
+    timeoutSec: DEFAULT_TIMEOUT_SEC,
+    port: DEFAULT_PORT,
+    host: DEFAULT_HOST,
+    apiKey: process.env.NINE_ROUTER_API_KEY || null,
+    pollIntervalMs: DEFAULT_POLL_INTERVAL_MS,
+  };
+  for (let i = 0; i < argv.length; i++) {
+    const a = argv[i];
+    const next = () => argv[++i];
+    if (a === "--prompt") opts.prompt = next();
+    else if (a === "--output" || a === "-o") opts.output = next();
+    else if (a === "--model") opts.model = next();
+    else if (a === "--duration") opts.duration = parseInt(next(), 10);
+    else if (a === "--aspect-ratio") opts.aspectRatio = next();
+    else if (a === "--resolution") opts.resolution = next();
+    else if (a === "--image") opts.image = next();
+    else if (a === "--timeout") opts.timeoutSec = parseInt(next(), 10) || DEFAULT_TIMEOUT_SEC;
+    else if (a === "--port" || a === "-p") opts.port = parseInt(next(), 10) || DEFAULT_PORT;
+    else if (a === "--host" || a === "-H") opts.host = next() || DEFAULT_HOST;
+    else if (a === "--api-key") opts.apiKey = next();
+    else if (a === "--poll-interval-ms") opts.pollIntervalMs = parseInt(next(), 10) || DEFAULT_POLL_INTERVAL_MS;
+    else if (a === "-h" || a === "--help") opts.help = true;
+    else {
+      throw new Error(`Unknown option: ${a}`);
+    }
+  }
+  return opts;
+}
+
+/** Local file path → base64 data URL; URLs pass through untouched. */
+function imageInputToUrl(input) {
+  if (/^(https?:|data:)/i.test(input)) return input;
+  const buf = fs.readFileSync(input);
+  const ext = path.extname(input).toLowerCase();
+  const mime = ext === ".png" ? "image/png" : ext === ".webp" ? "image/webp" : "image/jpeg";
+  return `data:${mime};base64,${buf.toString("base64")}`;
+}
+
+/** Minimal JSON request against the local gateway. Returns { status, headers, body }. */
+function gatewayRequest({ host, port, apiKey, method, reqPath, body, signal }) {
+  return new Promise((resolve, reject) => {
+    const payload = body ? JSON.stringify(body) : null;
+    const headers = { Accept: "application/json" };
+    if (payload) {
+      headers["Content-Type"] = "application/json";
+      headers["Content-Length"] = Buffer.byteLength(payload);
+    }
+    if (apiKey) headers.Authorization = `Bearer ${apiKey}`;
+
+    const req = http.request({ hostname: host, port, path: reqPath, method, headers, signal }, (res) => {
+      let data = "";
+      res.on("data", (c) => (data += c));
+      res.on("end", () => {
+        let parsed = null;
+        try { parsed = data ? JSON.parse(data) : null; } catch { /* keep raw */ }
+        resolve({ status: res.statusCode, headers: res.headers, body: parsed, raw: data });
+      });
+    });
+    req.on("error", reject);
+    if (payload) req.write(payload);
+    req.end();
+  });
+}
+
+const sleep = (ms, signal) =>
+  new Promise((resolve, reject) => {
+    const t = setTimeout(resolve, ms);
+    signal?.addEventListener?.("abort", () => { clearTimeout(t); reject(new Error("aborted")); }, { once: true });
+  });
+
+/**
+ * Poll GET /v1/videos/{id} until a terminal status or deadline.
+ * @returns {Promise<object>} final poll body (status done) — throws on failed/timeout.
+ */
+async function pollUntilDone({ host, port, apiKey, requestId, connectionId, timeoutSec, pollIntervalMs, signal, onProgress }) {
+  const deadline = Date.now() + timeoutSec * 1000;
+  while (true) {
+    if (signal?.aborted) throw new Error("aborted");
+    if (Date.now() > deadline) {
+      throw new Error(`Timed out after ${timeoutSec}s waiting for video job ${requestId}`);
+    }
+
+    const res = await gatewayRequestWithConnection({ host, port, apiKey, requestId, connectionId, signal });
+    if (res.status === 200 && res.body) {
+      const status = String(res.body.status || "").toLowerCase();
+      onProgress?.(status || "pending", res.body.progress);
+      if (FAILED_STATUSES.has(status)) {
+        const msg = res.body.error?.message || res.body.error || "video generation failed";
+        throw new Error(`Job ${requestId} failed: ${sanitizeText(typeof msg === "string" ? msg : JSON.stringify(msg))}`);
+      }
+      if (TERMINAL_STATUSES.has(status)) return res.body;
+    } else if (res.status >= 400 && res.status !== 429 && res.status !== 503) {
+      throw new Error(`Polling failed (HTTP ${res.status}): ${sanitizeText(res.raw?.slice(0, 300))}`);
+    }
+    await sleep(pollIntervalMs, signal);
+  }
+}
+
+function gatewayRequestWithConnection({ host, port, apiKey, requestId, connectionId, signal }) {
+  return new Promise((resolve, reject) => {
+    const headers = { Accept: "application/json" };
+    if (apiKey) headers.Authorization = `Bearer ${apiKey}`;
+    if (connectionId) headers["x-connection-id"] = connectionId;
+    const req = http.request(
+      { hostname: host, port, path: `/v1/videos/${encodeURIComponent(requestId)}`, method: "GET", headers, signal },
+      (res) => {
+        let data = "";
+        res.on("data", (c) => (data += c));
+        res.on("end", () => {
+          let parsed = null;
+          try { parsed = data ? JSON.parse(data) : null; } catch { /* keep raw */ }
+          resolve({ status: res.statusCode, body: parsed, raw: data });
+        });
+      }
+    );
+    req.on("error", reject);
+    req.end();
+  });
+}
+
+/**
+ * Download a URL to `outputPath` via a `.part` temp file with atomic rename.
+ * The temp file is removed on any failure.
+ */
+async function downloadToFile(url, outputPath, { signal } = {}) {
+  const partPath = `${outputPath}.part`;
+  await new Promise((resolve, reject) => {
+    const cleanupAnd = (fn) => (err) => {
+      try { fs.unlinkSync(partPath); } catch { /* not created yet */ }
+      fn(err);
+    };
+    const get = (target, redirectsLeft) => {
+      const mod = target.startsWith("https:") ? https : http;
+      const req = mod.get(target, { signal }, (res) => {
+        if (res.statusCode >= 300 && res.statusCode < 400 && res.headers.location && redirectsLeft > 0) {
+          res.resume();
+          return get(new URL(res.headers.location, target).toString(), redirectsLeft - 1);
+        }
+        if (res.statusCode !== 200) {
+          res.resume();
+          return cleanupAnd(reject)(new Error(`Download failed: HTTP ${res.statusCode}`));
+        }
+        const out = fs.createWriteStream(partPath);
+        res.pipe(out);
+        out.on("finish", () => out.close(resolve));
+        out.on("error", cleanupAnd(reject));
+        res.on("error", cleanupAnd(reject));
+      });
+      req.on("error", cleanupAnd(reject));
+    };
+    get(url, 5);
+  });
+  fs.renameSync(partPath, outputPath);
+}
+
+async function run(argv) {
+  let opts;
+  try {
+    opts = parseArgs(argv);
+  } catch (err) {
+    console.error(`❌ ${err.message}`);
+    console.log(HELP);
+    return 1;
+  }
+  if (opts.help) {
+    console.log(HELP);
+    return 0;
+  }
+  if (!opts.prompt) {
+    console.error("❌ --prompt is required");
+    console.log(HELP);
+    return 1;
+  }
+
+  const controller = new AbortController();
+  const partPath = `${opts.output}.part`;
+  const onSigint = () => {
+    controller.abort();
+    try { fs.unlinkSync(partPath); } catch { /* absent */ }
+    console.error("\n✋ Cancelled");
+    process.exit(130);
+  };
+  process.on("SIGINT", onSigint);
+
+  try {
+    const body = { model: opts.model, prompt: opts.prompt };
+    if (opts.duration) body.duration = opts.duration;
+    if (opts.aspectRatio) body.aspect_ratio = opts.aspectRatio;
+    if (opts.resolution) body.resolution = opts.resolution;
+    if (opts.image) body.image = { url: imageInputToUrl(opts.image) };
+
+    console.log(`🎬 Requesting video (${opts.model})…`);
+    const create = await gatewayRequest({
+      host: opts.host, port: opts.port, apiKey: opts.apiKey,
+      method: "POST", reqPath: "/v1/videos/generations", body, signal: controller.signal,
+    });
+
+    if (create.status !== 200 || !create.body?.request_id) {
+      const detail = create.body?.error?.message || create.body?.error || create.raw || `HTTP ${create.status}`;
+      console.error(`❌ Create failed: ${sanitizeText(typeof detail === "string" ? detail : JSON.stringify(detail)).slice(0, 500)}`);
+      if (create.status === 400 && /No credentials/i.test(String(detail))) {
+        console.error("   Connect an xAI account first: dashboard → Providers → xAI (Grok).");
+      }
+      return 1;
+    }
+
+    const requestId = create.body.request_id;
+    const connectionId = create.headers["x-9router-connection-id"] || null;
+    console.log(`📋 Job accepted: ${requestId}`);
+
+    let lastLine = "";
+    const result = await pollUntilDone({
+      host: opts.host, port: opts.port, apiKey: opts.apiKey,
+      requestId, connectionId,
+      timeoutSec: opts.timeoutSec, pollIntervalMs: opts.pollIntervalMs,
+      signal: controller.signal,
+      onProgress: (status, progress) => {
+        const line = `⏳ ${status}${Number.isFinite(progress) ? ` ${progress}%` : ""}`;
+        if (line !== lastLine) {
+          lastLine = line;
+          if (process.stdout.isTTY) process.stdout.write(`\r\x1b[K${line}`);
+          else console.log(line);
+        }
+      },
+    });
+    if (process.stdout.isTTY) process.stdout.write("\n");
+
+    const videoUrl = result.video?.url || result.video?.file_output?.public_url;
+    if (!videoUrl) {
+      console.error("❌ Job finished but no video URL was returned");
+      return 1;
+    }
+
+    console.log("⬇️  Downloading…");
+    await downloadToFile(videoUrl, opts.output, { signal: controller.signal });
+    console.log(`✅ Saved ${opts.output}`);
+    return 0;
+  } catch (err) {
+    if (process.stdout.isTTY) process.stdout.write("\n");
+    console.error(`❌ ${sanitizeText(err?.message || String(err))}`);
+    return 1;
+  } finally {
+    process.removeListener("SIGINT", onSigint);
+  }
+}
+
+module.exports = { run, parseArgs, pollUntilDone, downloadToFile, imageInputToUrl, sanitizeText };
--- a/cli/src/cli/menus/providers.js
+++ b/cli/src/cli/menus/providers.js
@@ -53,6 +53,9 @@ const PROVIDER_MODELS = {
    { id: "glm-4.7" },
  ],
  ag: [
+    { id: "gemini-3.6-flash-high" },
+    { id: "gemini-3.6-flash-medium" },
+    { id: "gemini-3.6-flash-low" },
    { id: "gemini-3-flash-agent" },
    { id: "gemini-3.5-flash-low" },
    { id: "gemini-3.5-flash-extra-low" },
@@ -78,6 +81,7 @@ const PROVIDER_MODELS = {
    { id: "grok-code-fast-1" },
  ],
  kr: [
+    { id: "claude-sonnet-5" },
    { id: "claude-sonnet-4.5" },
    { id: "claude-haiku-4.5" },
  ],
@@ -94,6 +98,8 @@ const PROVIDER_MODELS = {
    { id: "claude-3-5-sonnet-20241022" },
  ],
  gemini: [
+    { id: "gemini-3.6-flash" },
+    { id: "gemini-3.5-flash-lite" },
    { id: "gemini-3-pro-preview" },
    { id: "gemini-2.5-pro" },
    { id: "gemini-2.5-flash" },
@@ -130,7 +136,7 @@ const APIKEY_PROVIDERS = {
  openrouter: { id: "openrouter", name: "OpenRouter" },
  glm: { id: "glm", name: "GLM Coding" },
  minimax: { id: "minimax", name: "Minimax Coding" },
-  kimi: { id: "kimi", name: "Kimi Coding" },
+  kimi: { id: "kimi", name: "Kimi" },
  openai: { id: "openai", name: "OpenAI" },
  anthropic: { id: "anthropic", name: "Anthropic" },
  gemini: { id: "gemini", name: "Gemini" },
--- a/cli/src/cli/tray/tray.ps1
+++ b/cli/src/cli/tray/tray.ps1
@@ -2,14 +2,65 @@
 # IPC: stdin JSON commands, stdout JSON events
 param([string]$IconPath, [string]$Tooltip)

+$ErrorActionPreference = "Stop"
+
+Add-Type @"
+using System;
+using System.Runtime.InteropServices;
+
+public static class WinDpiAwareness {
+  public static IntPtr PerMonitorAwareV2 { get { return new IntPtr(-4); } }
+  public static IntPtr PerMonitorAware { get { return new IntPtr(-3); } }
+
+  [DllImport("user32.dll")]
+  public static extern bool SetProcessDpiAwarenessContext(IntPtr value);
+
+  [DllImport("user32.dll")]
+  public static extern IntPtr SetThreadDpiAwarenessContext(IntPtr value);
+
+  [DllImport("shcore.dll")]
+  public static extern int SetProcessDpiAwareness(int value);
+
+  [DllImport("user32.dll")]
+  public static extern bool SetProcessDPIAware();
+}
+"@
+
+function Enable-HighDpiAwareness {
+  $contexts = @(
+    [WinDpiAwareness]::PerMonitorAwareV2,
+    [WinDpiAwareness]::PerMonitorAware
+  )
+
+  foreach ($context in $contexts) {
+    try {
+      if ([WinDpiAwareness]::SetProcessDpiAwarenessContext($context)) { break }
+    } catch {}
+  }
+
+  try { [WinDpiAwareness]::SetProcessDpiAwareness(2) | Out-Null } catch {}
+  try { [WinDpiAwareness]::SetProcessDPIAware() | Out-Null } catch {}
+
+  foreach ($context in $contexts) {
+    try {
+      $previous = [WinDpiAwareness]::SetThreadDpiAwarenessContext($context)
+      if ($previous -ne [IntPtr]::Zero) { break }
+    } catch {}
+  }
+}
+
+Enable-HighDpiAwareness
+
 Add-Type -AssemblyName System.Windows.Forms
 Add-Type -AssemblyName System.Drawing

-$ErrorActionPreference = "Stop"
 [Console]::OutputEncoding = [System.Text.Encoding]::UTF8
 [Console]::InputEncoding = [System.Text.Encoding]::UTF8
 $OutputEncoding = [System.Text.Encoding]::UTF8

+[System.Windows.Forms.Application]::EnableVisualStyles()
+[System.Windows.Forms.Application]::SetCompatibleTextRenderingDefault($false)
+
 $script:notifyIcon = New-Object System.Windows.Forms.NotifyIcon
 $script:notifyIcon.Icon = New-Object System.Drawing.Icon($IconPath)
 $script:notifyIcon.Text = $Tooltip
--- a/i18n/README.fa_IR.md
+++ b/i18n/README.fa_IR.md
--- a/i18n/README.id-ID.md
+++ b/i18n/README.id-ID.md
@@ -0,0 +1,951 @@
+<div align="center">
+  <img src="../images/9router.png?1" alt="9Router Dashboard" width="800"/>
+
+  # 9Router - Router AI Gratis
+
+  **Jangan berhenti ngoding. Otomatis dialihkan ke model AI gratis & murah dengan smart fallback.**
+
+  **Hubungkan semua tool AI coding (Claude Code, Cursor, Antigravity, Copilot, Codex, Gemini, OpenCode, Cline, OpenClaw...) ke 40+ provider AI dan 100+ model.**
+
+  [![npm](https://img.shields.io/npm/v/9router.svg)](https://www.npmjs.com/package/9router)
+  [![Downloads](https://img.shields.io/npm/dm/9router.svg)](https://www.npmjs.com/package/9router)
+  [![License](https://img.shields.io/npm/l/9router.svg)](https://github.com/decolua/9router/blob/main/LICENSE)
+
+  [🚀 Mulai Cepat](#-mulai-cepat) • [💡 Fitur](#-fitur-utama) • [📖 Setup](#-panduan-setup) • [🌐 Website](https://9router.com)
+
+  [🇻🇳 Tiếng Việt](./README.vi.md) • [🇨🇳 中文](./README.zh-CN.md) • [🇯🇵 日本語](./README.ja-JP.md) • [🇮🇩 Bahasa Indonesia](./README.id-ID.md)
+</div>
+
+---
+
+## 🤔 Kenapa 9Router?
+
+**Berhenti buang-buang uang dan terhambat limit:**
+
+- ❌ Kuota langganan hangus tiap bulan tanpa terpakai
+- ❌ Rate limit bikin ngoding berhenti di tengah jalan
+- ❌ API mahal ($20–50/bulan per provider)
+- ❌ Harus gonta-ganti provider secara manual
+
+**9Router menyelesaikan itu semua:**
+
+- ✅ **Maksimalkan langganan** - lacak kuota dan habiskan sebelum reset
+- ✅ **Fallback otomatis** - langganan → murah → gratis, tanpa downtime
+- ✅ **Multi-akun** - round-robin antar akun untuk tiap provider
+- ✅ **Universal** - mendukung Claude Code, Codex, Gemini CLI, Cursor, Cline, dan tool CLI apa pun
+
+---
+
+## 🔄 Cara Kerja
+
+```
+┌─────────────┐
+│   Tool CLI  │  (Claude Code, Codex, Gemini CLI, OpenClaw, Cursor, Cline...)
+│    kamu     │
+└──────┬──────┘
+       │ http://localhost:20128/v1
+       ↓
+┌─────────────────────────────────────────┐
+│         9Router (Smart Router)          │
+│  • Konversi format (OpenAI ↔ Claude)    │
+│  • Pelacakan kuota                      │
+│  • Refresh token otomatis               │
+└──────┬──────────────────────────────────┘
+       │
+       ├─→ [Tier 1: Langganan] Claude Code, Codex, Gemini CLI
+       │   ↓ kuota habis
+       ├─→ [Tier 2: Murah] GLM ($0.6/1M), MiniMax ($0.2/1M)
+       │   ↓ batas budget tercapai
+       └─→ [Tier 3: Gratis] iFlow, Qwen, Kiro (unlimited)
+
+Hasil: ngoding tanpa berhenti, biaya minimum
+```
+
+---
+
+## ⚡ Mulai Cepat
+
+**1. Install secara global:**
+
+```bash
+npm install -g 9router
+9router
+```
+
+🎉 Dashboard terbuka di `http://localhost:20128`
+
+**2. Hubungkan provider gratis (tanpa perlu daftar):**
+
+Dashboard → Providers → hubungkan **Claude Code** atau **Antigravity** → login OAuth → selesai!
+
+**3. Pakai di tool CLI kamu:**
+
+```
+Konfigurasi Claude Code/Codex/Gemini CLI/OpenClaw/Cursor/Cline:
+  Endpoint: http://localhost:20128/v1
+  API Key: [salin dari dashboard]
+  Model: if/kimi-k2-thinking
+```
+
+**Cuma itu!** Mulai ngoding dengan model AI gratis.
+
+**Alternatif: jalankan dari source (repo ini):**
+
+Paket repo ini bersifat privat (`9router-app`), jadi menjalankan dari source/Docker adalah jalur yang diharapkan untuk pengembangan lokal.
+
+```bash
+cp .env.example .env
+npm install
+PORT=20128 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run dev
+```
+
+Mode produksi:
+
+```bash
+npm run build
+PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run start
+```
+
+URL default:
+- Dashboard: `http://localhost:20128/dashboard`
+- API kompatibel OpenAI: `http://localhost:20128/v1`
+
+---
+
+## 🎥 Video Tutorial
+
+<div align="center">
+
+### 📺 Panduan Setup Lengkap - 9Router + Claude Code Gratis
+
+[![9Router + Claude Code Setup](https://img.youtube.com/vi/raEyZPg5xE0/maxresdefault.jpg)](https://www.youtube.com/watch?v=raEyZPg5xE0)
+
+**🎬 Tonton tutorial langkah demi langkah:**
+- ✅ Install dan setup 9Router
+- ✅ Konfigurasi Claude Sonnet 4.5 gratis
+- ✅ Integrasi dengan Claude Code
+- ✅ Demo live coding
+
+**⏱️ Durasi:** 20 menit | **👥 Dibuat oleh:** Developer Community
+
+[▶️ Tonton di YouTube](https://www.youtube.com/watch?v=o3qYCyjrFYg)
+
+</div>
+
+---
+
+## 🛠️ Tool CLI yang Didukung
+
+9Router bekerja mulus dengan semua tool AI coding utama:
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/claude.png" width="60" alt="Claude Code"/><br/>
+        <b>Claude-Code</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/openclaw.png" width="60" alt="OpenClaw"/><br/>
+        <b>OpenClaw</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/codex.png" width="60" alt="Codex"/><br/>
+        <b>Codex</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/opencode.png" width="60" alt="OpenCode"/><br/>
+        <b>OpenCode</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/cursor.png" width="60" alt="Cursor"/><br/>
+        <b>Cursor</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/antigravity.png" width="60" alt="Antigravity"/><br/>
+        <b>Antigravity</b>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/cline.png" width="60" alt="Cline"/><br/>
+        <b>Cline</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/continue.png" width="60" alt="Continue"/><br/>
+        <b>Continue</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/droid.png" width="60" alt="Droid"/><br/>
+        <b>Droid</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/roo.png" width="60" alt="Roo"/><br/>
+        <b>Roo</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/copilot.png" width="60" alt="Copilot"/><br/>
+        <b>Copilot</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/kilocode.png" width="60" alt="Kilo Code"/><br/>
+        <b>Kilo Code</b>
+      </td>
+    </tr>
+  </table>
+</div>
+
+---
+
+## 🌐 Provider yang Didukung
+
+### 🔐 Provider OAuth
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/claude.png" width="60" alt="Claude Code"/><br/>
+        <b>Claude-Code</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/antigravity.png" width="60" alt="Antigravity"/><br/>
+        <b>Antigravity</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/codex.png" width="60" alt="Codex"/><br/>
+        <b>Codex</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/github.png" width="60" alt="GitHub"/><br/>
+        <b>GitHub</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/cursor.png" width="60" alt="Cursor"/><br/>
+        <b>Cursor</b>
+      </td>
+    </tr>
+  </table>
+</div>
+
+### 🆓 Provider Gratis
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="150">
+        <img src="../public/providers/iflow.png" width="70" alt="iFlow"/><br/>
+        <b>iFlow AI</b><br/>
+        <sub>8+ model • unlimited</sub>
+      </td>
+      <td align="center" width="150">
+        <img src="../public/providers/qwen.png" width="70" alt="Qwen"/><br/>
+        <b>Qwen Code</b><br/>
+        <sub>3+ model • unlimited</sub>
+      </td>
+      <td align="center" width="150">
+        <img src="../public/providers/gemini-cli.png" width="70" alt="Gemini CLI"/><br/>
+        <b>Gemini CLI</b><br/>
+        <sub>180 ribu request/bulan gratis</sub>
+      </td>
+      <td align="center" width="150">
+        <img src="../public/providers/kiro.png" width="70" alt="Kiro"/><br/>
+        <b>Kiro AI</b><br/>
+        <sub>Claude • unlimited</sub>
+      </td>
+    </tr>
+  </table>
+</div>
+
+### 🔑 Provider API Key (40+)
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/openrouter.png" width="50" alt="OpenRouter"/><br/>
+        <sub>OpenRouter</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/glm.png" width="50" alt="GLM"/><br/>
+        <sub>GLM</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/kimi.png" width="50" alt="Kimi"/><br/>
+        <sub>Kimi</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/minimax.png" width="50" alt="MiniMax"/><br/>
+        <sub>MiniMax</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/openai.png" width="50" alt="OpenAI"/><br/>
+        <sub>OpenAI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/anthropic.png" width="50" alt="Anthropic"/><br/>
+        <sub>Anthropic</sub>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/gemini.png" width="50" alt="Gemini"/><br/>
+        <sub>Gemini</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/deepseek.png" width="50" alt="DeepSeek"/><br/>
+        <sub>DeepSeek</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/groq.png" width="50" alt="Groq"/><br/>
+        <sub>Groq</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/xai.png" width="50" alt="xAI"/><br/>
+        <sub>xAI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/mistral.png" width="50" alt="Mistral"/><br/>
+        <sub>Mistral</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/perplexity.png" width="50" alt="Perplexity"/><br/>
+        <sub>Perplexity</sub>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/together.png" width="50" alt="Together"/><br/>
+        <sub>Together AI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/fireworks.png" width="50" alt="Fireworks"/><br/>
+        <sub>Fireworks</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/cerebras.png" width="50" alt="Cerebras"/><br/>
+        <sub>Cerebras</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/cohere.png" width="50" alt="Cohere"/><br/>
+        <sub>Cohere</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/nvidia.png" width="50" alt="NVIDIA"/><br/>
+        <sub>NVIDIA</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/siliconflow.png" width="50" alt="SiliconFlow"/><br/>
+        <sub>SiliconFlow</sub>
+      </td>
+    </tr>
+  </table>
+  <p><i>...dan 20+ provider lain seperti Nebius, Chutes, Hyperbolic, serta endpoint custom yang kompatibel dengan OpenAI/Anthropic</i></p>
+</div>
+
+---
+
+## 💡 Fitur Utama
+
+| Fitur | Ringkasan | Manfaat |
+|-------|-----------|---------|
+| 🎯 **Smart Fallback 3 Tingkat** | Routing otomatis: langganan → murah → gratis | Ngoding tanpa berhenti, zero downtime |
+| 📊 **Pelacakan Kuota Real-time** | Hitungan token live + hitung mundur reset | Nilai langganan termanfaatkan maksimal |
+| 🔄 **Konversi Format** | OpenAI ↔ Claude ↔ Gemini mulus | Bekerja dengan tool CLI apa pun |
+| 👥 **Dukungan Multi-akun** | Beberapa akun per provider | Load balancing + redundansi |
+| 🔄 **Auto Refresh Token** | Token OAuth diperbarui otomatis | Tidak perlu login ulang manual |
+| 🎨 **Combo Kustom** | Buat kombinasi model tanpa batas | Fallback sesuai kebutuhanmu |
+| 📝 **Log Request** | Log lengkap request/response | Troubleshooting jadi mudah |
+| 💾 **Cloud Sync** | Sinkronkan pengaturan antar perangkat | Setup sama di mana pun |
+| 📊 **Analitik Penggunaan** | Lacak token, biaya, dan tren | Optimalkan pengeluaran |
+| 🌐 **Deploy di Mana Saja** | Localhost, VPS, Docker, Cloudflare Workers | Opsi deployment fleksibel |
+
+<details>
+<summary><b>📖 Detail Fitur</b></summary>
+
+### 🎯 Smart Fallback 3 Tingkat
+
+Buat combo dengan fallback otomatis:
+
+```
+Combo: "my-coding-stack"
+  1. cc/claude-opus-4-6        (langganan)
+  2. glm/glm-4.7               (backup murah, $0.6/1M)
+  3. if/kimi-k2-thinking       (fallback gratis)
+
+→ Otomatis beralih saat kuota habis atau terjadi error
+```
+
+### 📊 Pelacakan Kuota Real-time
+
+- Konsumsi token per provider
+- Hitung mundur reset (5 jam, harian, mingguan)
+- Estimasi biaya untuk tier berbayar
+- Laporan pengeluaran bulanan
+
+### 🔄 Konversi Format
+
+Konversi mulus antar format:
+- **OpenAI** ↔ **Claude** ↔ **Gemini** ↔ **OpenAI Responses**
+- Tool CLI mengirim dalam format OpenAI → 9Router mengonversi → provider menerima dalam format nativenya
+- Bekerja dengan semua tool yang mendukung custom OpenAI endpoint
+
+### 👥 Dukungan Multi-akun
+
+- Tambahkan beberapa akun per provider
+- Round-robin otomatis atau routing berbasis prioritas
+- Saat satu akun mencapai kuota, fallback ke akun berikutnya
+
+### 🔄 Auto Refresh Token
+
+- Token OAuth di-refresh otomatis sebelum kedaluwarsa
+- Tidak perlu autentikasi ulang manual
+- Pengalaman mulus di semua provider
+
+### 🎨 Combo Kustom
+
+- Buat kombinasi model tanpa batas
+- Campur tier langganan, murah, dan gratis
+- Beri nama combo agar mudah diakses
+- Bagikan combo antar perangkat lewat cloud sync
+
+### 📝 Log Request
+
+- Log lengkap request/response dalam mode debug
+- Lacak API call, header, dan payload
+- Troubleshoot masalah integrasi
+- Ekspor log untuk dianalisis
+
+### 💾 Cloud Sync
+
+- Sinkronkan provider, combo, dan pengaturan antar perangkat
+- Sinkronisasi latar belakang otomatis
+- Penyimpanan terenkripsi yang aman
+- Akses setup dari mana saja
+
+#### Catatan tentang cloud runtime
+
+- Untuk produksi, disarankan memakai variabel cloud sisi server:
+  - `BASE_URL` (URL callback internal yang dipakai scheduler sinkronisasi)
+  - `CLOUD_URL` (base URL endpoint cloud sync)
+- `NEXT_PUBLIC_BASE_URL` dan `NEXT_PUBLIC_CLOUD_URL` masih didukung untuk kompatibilitas/UI, tetapi runtime server memprioritaskan `BASE_URL`/`CLOUD_URL`.
+- Request cloud sync memakai timeout + perilaku fail-fast untuk menghindari UI menggantung saat DNS/jaringan cloud tidak tersedia.
+
+### 📊 Analitik Penggunaan
+
+- Lacak pemakaian token per provider dan per model
+- Estimasi biaya dan tren pengeluaran
+- Laporan dan insight bulanan
+- Optimalkan pengeluaran AI
+
+> **💡 PENTING - tentang biaya di dashboard:**
+>
+> "Biaya" yang ditampilkan pada analitik penggunaan **hanya untuk pelacakan dan perbandingan**.
+> 9Router sendiri **tidak menagih apa pun**. Kamu hanya membayar langsung ke provider jika memakai layanan berbayar.
+>
+> **Contoh:** jika dashboard menampilkan "Total biaya $290" untuk pemakaian model iFlow,
+> itu adalah jumlah yang seharusnya kamu bayar bila memakai API berbayar secara langsung. Biaya sebenarnya = **$0** (iFlow gratis tanpa batas).
+>
+> Anggap saja ini "pelacak penghematan" yang menunjukkan berapa banyak yang kamu hemat lewat model gratis dan routing 9Router!
+
+### 🌐 Deploy di Mana Saja
+
+- 💻 **Localhost** - default, jalan offline
+- ☁️ **VPS/Cloud** - berbagi antar perangkat
+- 🐳 **Docker** - deploy satu perintah
+- 🚀 **Cloudflare Workers** - jaringan edge global
+
+</details>
+
+---
+
+## 💰 Ringkasan Harga
+
+| Tier | Provider | Biaya | Reset Kuota | Cocok Untuk |
+|------|----------|-------|-------------|-------------|
+| **💳 Langganan** | Claude Code (Pro) | $20/bulan | 5 jam + mingguan | Yang sudah punya langganan |
+| | Codex (Plus/Pro) | $20-200/bulan | 5 jam + mingguan | Pengguna OpenAI |
+| | Gemini CLI | **Gratis** | 180rb/bulan + 1rb/hari | Semua orang! |
+| | GitHub Copilot | $10-19/bulan | Bulanan | Pengguna GitHub |
+| **💰 Murah** | GLM-4.7 | $0.6/1M | Setiap hari jam 10.00 | Backup hemat |
+| | MiniMax M2.1 | $0.2/1M | Rolling 5 jam | Opsi paling murah |
+| | Kimi K2 | $9/bulan flat | 10 juta token/bulan | Biaya yang bisa diprediksi |
+| **🆓 Gratis** | iFlow | $0 | Unlimited | 8 model gratis |
+| | Qwen | $0 | Unlimited | 3 model gratis |
+| | Kiro | $0 | Unlimited | Claude gratis |
+
+**💡 Tips pro:** combo Gemini CLI (180rb request/bulan gratis) + iFlow (gratis unlimited) = biaya $0!
+
+---
+
+### 📊 Tentang Biaya dan Penagihan 9Router
+
+**Fakta soal penagihan 9Router:**
+
+✅ **Software 9Router = gratis selamanya** (open source, tanpa tagihan)
+✅ **"Biaya" di dashboard = tampilan/pelacakan saja** (bukan tagihan sungguhan)
+✅ **Pembayaran langsung ke provider** (langganan atau biaya API)
+✅ **Provider gratis tetap gratis** (iFlow, Kiro, Qwen = $0 unlimited)
+❌ **9Router tidak mengirim invoice** atau menagih kartumu
+
+**Cara kerja tampilan biaya:**
+
+Dashboard menampilkan **estimasi biaya** seandainya kamu memakai API berbayar secara langsung. Ini **bukan tagihan**, melainkan alat pembanding yang menunjukkan penghematanmu.
+
+**Contoh skenario:**
+```
+Tampilan dashboard:
+• Total request: 1.662
+• Total token: 47 juta
+• Biaya tertampil: $290
+
+Kenyataannya:
+• Provider: iFlow (gratis unlimited)
+• Yang benar-benar dibayar: $0.00
+• Arti $290: jumlah yang kamu hemat dengan memakai model gratis!
+```
+
+**Aturan pembayaran:**
+- **Provider langganan** (Claude Code, Codex): bayar langsung di website masing-masing
+- **Provider murah** (GLM, MiniMax): bayar langsung, 9Router hanya melakukan routing
+- **Provider gratis** (iFlow, Kiro, Qwen): benar-benar gratis selamanya, tanpa biaya tersembunyi
+- **9Router**: tidak menagih apa pun
+
+---
+
+## 🎯 Studi Kasus
+
+### Kasus 1: "Saya punya langganan Claude Pro"
+
+**Masalah:** kuota hangus tanpa terpakai, kena rate limit saat ngoding berat
+
+**Solusi:**
+```
+Combo: "maximize-claude"
+  1. cc/claude-opus-4-6        (manfaatkan langganan semaksimal mungkin)
+  2. glm/glm-4.7               (backup murah saat kuota habis)
+  3. if/kimi-k2-thinking       (fallback darurat gratis)
+
+Biaya bulanan: $20 (langganan) + ~$5 (backup) = total $25
+vs. $20 + kena limit = frustrasi
+```
+
+### Kasus 2: "Saya mau biaya nol"
+
+**Masalah:** tidak mampu bayar langganan, tapi butuh AI coding yang andal
+
+**Solusi:**
+```
+Combo: "free-forever"
+  1. gc/gemini-3-flash         (180rb request/bulan gratis)
+  2. if/kimi-k2-thinking       (gratis unlimited)
+  3. qw/qwen3-coder-plus       (gratis unlimited)
+
+Biaya bulanan: $0
+Kualitas: model siap produksi
+```
+
+### Kasus 3: "Ngoding 24/7 tanpa terputus"
+
+**Masalah:** deadline mepet, downtime tidak dapat ditoleransi
+
+**Solusi:**
+```
+Combo: "always-on"
+  1. cc/claude-opus-4-6        (kualitas terbaik)
+  2. cx/gpt-5.2-codex          (langganan kedua)
+  3. glm/glm-4.7               (murah, reset harian)
+  4. minimax/MiniMax-M2.1      (paling murah, reset 5 jam)
+  5. if/kimi-k2-thinking       (gratis unlimited)
+
+Hasil: 5 lapis fallback = zero downtime
+Biaya bulanan: $20-200 (langganan) + $10-20 (backup)
+```
+
+### Kasus 4: "Saya mau pakai AI gratis di OpenClaw"
+
+**Masalah:** butuh asisten AI di aplikasi pesan (WhatsApp, Telegram, Slack...), sepenuhnya gratis
+
+**Solusi:**
+```
+Combo: "openclaw-free"
+  1. if/glm-4.7                (gratis unlimited)
+  2. if/minimax-m2.1           (gratis unlimited)
+  3. if/kimi-k2-thinking       (gratis unlimited)
+
+Biaya bulanan: $0
+Cara akses: WhatsApp, Telegram, Slack, Discord, iMessage, Signal...
+```
+
+---
+
+## ❓ FAQ
+
+<details>
+<summary><b>📊 Kenapa dashboard menampilkan biaya yang besar?</b></summary>
+
+Dashboard melacak pemakaian token dan menampilkan **estimasi biaya** seandainya kamu memakai API berbayar secara langsung. Ini **bukan tagihan nyata**, melainkan acuan untuk melihat berapa banyak yang kamu hemat dengan memakai model gratis atau langganan yang sudah ada lewat 9Router.
+
+**Contoh:**
+- **Tampilan dashboard:** "Total biaya $290"
+- **Kenyataan:** sedang memakai iFlow (gratis unlimited)
+- **Biaya sebenarnya:** **$0.00**
+- **Arti $290:** jumlah yang **dihemat** karena memakai model gratis alih-alih API berbayar!
+
+Tampilan biaya adalah "pelacak penghematan" untuk memahami pola pemakaian dan peluang optimasi.
+
+</details>
+
+<details>
+<summary><b>💳 Apakah 9Router menagih saya?</b></summary>
+
+**Tidak.** 9Router adalah software open source gratis yang berjalan di komputermu sendiri. Tidak ada penagihan sama sekali.
+
+**Kamu membayar ke:**
+- ✅ **Provider langganan** (Claude Code $20/bulan, Codex $20-200/bulan) → bayar langsung di website masing-masing
+- ✅ **Provider murah** (GLM, MiniMax) → bayar langsung, 9Router hanya me-routing request
+- ❌ **9Router sendiri** → **tidak menagih apa pun**
+
+9Router adalah proxy/router lokal. Ia tidak menyimpan informasi kartu kredit, tidak bisa mengirim invoice, dan tidak punya sistem penagihan. Sepenuhnya software gratis.
+
+</details>
+
+<details>
+<summary><b>🆓 Apakah provider gratis benar-benar unlimited?</b></summary>
+
+**Ya!** Provider yang ditandai gratis (iFlow, Kiro, Qwen) benar-benar unlimited dan **tanpa biaya tersembunyi**.
+
+Ini adalah layanan gratis yang disediakan masing-masing perusahaan:
+- **iFlow**: akses gratis unlimited ke 8+ model via OAuth
+- **Kiro**: model Claude gratis unlimited via AWS Builder ID
+- **Qwen**: akses gratis unlimited ke model Qwen via device authentication
+
+9Router hanya me-routing request — tidak ada "jebakan" atau tagihan di kemudian hari. Layanannya memang gratis, dan 9Router membuatnya lebih mudah dipakai dengan dukungan fallback.
+
+**Catatan:** beberapa provider langganan (Antigravity, GitHub Copilot) punya masa preview gratis dan bisa jadi berbayar nanti, tetapi hal itu diumumkan secara jelas oleh provider tersebut, bukan oleh 9Router.
+
+</details>
+
+<details>
+<summary><b>💰 Bagaimana cara menekan biaya AI seminimal mungkin?</b></summary>
+
+**Strategi free-first:**
+
+1. **Mulai dari combo 100% gratis:**
+   ```
+   1. gc/gemini-3-flash (180rb/bulan gratis dari Google)
+   2. if/kimi-k2-thinking (gratis unlimited dari iFlow)
+   3. qw/qwen3-coder-plus (gratis unlimited dari Qwen)
+   ```
+   **Biaya: $0/bulan**
+
+2. **Tambahkan backup murah hanya bila perlu:**
+   ```
+   4. glm/glm-4.7 ($0.6 per 1 juta token)
+   ```
+   **Tambahan biaya: bayar sesuai pemakaian saja**
+
+3. **Gunakan provider langganan paling akhir:**
+   - Hanya jika kamu memang sudah punya
+   - 9Router memaksimalkan nilainya lewat pelacakan kuota
+
+**Hasil:** sebagian besar pengguna bisa jalan dengan $0/bulan hanya dengan tier gratis!
+
+</details>
+
+<details>
+<summary><b>📈 Bagaimana kalau pemakaian tiba-tiba melonjak?</b></summary>
+
+Smart fallback 9Router mencegah tagihan tak terduga:
+
+**Skenario:** kuota habis di tengah sprint coding
+
+**Tanpa 9Router:**
+- ❌ Kena rate limit → kerja berhenti → frustrasi
+- ❌ Atau: tagihan API mahal tanpa disengaja
+
+**Dengan 9Router:**
+- ✅ Langganan mencapai batas → otomatis fallback ke tier murah
+- ✅ Tier murah jadi mahal → otomatis fallback ke tier gratis
+- ✅ Ngoding tidak berhenti → biaya tetap terprediksi
+
+**Kamu yang pegang kendali:** atur batas pengeluaran per provider di dashboard, dan 9Router akan mematuhinya.
+
+</details>
+
+---
+
+## 📖 Panduan Setup
+
+<details>
+<summary><b>🔐 Provider Langganan (maksimalkan nilainya)</b></summary>
+
+### Claude Code (Pro/Max)
+
+```bash
+Dashboard → Providers → hubungkan Claude Code
+→ login OAuth → refresh token otomatis
+→ pelacakan kuota 5 jam + mingguan
+
+Model:
+  cc/claude-opus-4-6
+  cc/claude-sonnet-4-5-20250929
+  cc/claude-haiku-4-5-20251001
+```
+
+**Tips pro:** pakai Opus untuk tugas kompleks, Sonnet kalau mengutamakan kecepatan. 9Router melacak kuota per model!
+
+### OpenAI Codex (Plus/Pro)
+
+```bash
+Dashboard → Providers → hubungkan Codex
+→ login OAuth (port 1455)
+→ reset 5 jam + mingguan
+
+Model:
+  cx/gpt-5.2-codex
+  cx/gpt-5.1-codex-max
+```
+
+### Gemini CLI (180rb request/bulan gratis!)
+
+```bash
+Dashboard → Providers → hubungkan Gemini CLI
+→ Google OAuth
+→ 180rb/bulan + 1rb/hari
+
+Model:
+  gc/gemini-3-flash-preview
+  gc/gemini-2.5-pro
+```
+
+**Value terbaik:** free tier-nya besar sekali! Pakai ini sebelum tier berbayar.
+
+### GitHub Copilot
+
+```bash
+Dashboard → Providers → hubungkan GitHub
+→ OAuth via GitHub
+→ reset bulanan (tanggal 1 tiap bulan)
+
+Model:
+  gh/gpt-5
+  gh/claude-4.5-sonnet
+  gh/gemini-3-pro
+```
+
+</details>
+
+<details>
+<summary><b>💰 Provider Murah (backup)</b></summary>
+
+### GLM-4.7 (reset harian, $0.6/1M)
+
+1. Daftar: [Zhipu AI](https://open.bigmodel.cn/)
+2. Ambil API key dari Coding Plan
+3. Dashboard → tambahkan API key:
+   - Provider: `glm`
+   - API Key: `your-key`
+
+**Pemakaian:** `glm/glm-4.7`
+
+**Tips pro:** Coding Plan memberi kuota 3x lipat dengan biaya 1/7! Reset setiap hari jam 10.00.
+
+### MiniMax M2.1 (reset 5 jam, $0.20/1M)
+
+1. Daftar: [MiniMax](https://www.minimax.io/)
+2. Ambil API key
+3. Dashboard → tambahkan API key
+
+**Pemakaian:** `minimax/MiniMax-M2.1`
+
+**Tips pro:** opsi termurah dengan konteks panjang (1 juta token)!
+
+### Kimi K2 ($9/bulan flat)
+
+1. Berlangganan: [Moonshot AI](https://platform.moonshot.ai/)
+2. Ambil API key
+3. Dashboard → tambahkan API key
+
+**Pemakaian:** `kimi/kimi-latest`
+
+**Tips pro:** $9/bulan flat untuk 10 juta token = biaya efektif $0.90/1M!
+
+</details>
+
+<details>
+<summary><b>🆓 Provider Gratis (backup darurat)</b></summary>
+
+### iFlow (8 model gratis)
+
+```bash
+Dashboard → hubungkan iFlow
+→ login OAuth iFlow
+→ pemakaian unlimited
+
+Model:
+  if/kimi-k2-thinking
+  if/qwen3-coder-plus
+  if/glm-4.7
+  if/minimax-m2
+  if/deepseek-r1
+```
+
+### Qwen (3 model gratis)
+
+```bash
+Dashboard → hubungkan Qwen
+→ autentikasi device code
+→ pemakaian unlimited
+
+Model:
+  qw/qwen3-coder-plus
+  qw/qwen3-coder-flash
+```
+
+### Kiro (Claude gratis)
+
+```bash
+Dashboard → hubungkan Kiro
+→ AWS Builder ID atau Google/GitHub
+→ pemakaian unlimited
+
+Model:
+  kr/claude-sonnet-4.5
+  kr/claude-haiku-4.5
+```
+
+</details>
+
+<details>
+<summary><b>🎨 Membuat Combo</b></summary>
+
+### Contoh 1: maksimalkan langganan → backup murah
+
+```
+Dashboard → Combos → buat baru
+
+Nama: premium-coding
+Model:
+  1. cc/claude-opus-4-6 (langganan, utama)
+  2. glm/glm-4.7 (backup murah, $0.6/1M)
+  3. minimax/MiniMax-M2.1 (fallback termurah, $0.20/1M)
+
+Pemakaian di CLI: premium-coding
+
+Contoh biaya bulanan (100 juta token):
+  80 juta lewat Claude (langganan): tambahan $0
+  15 juta lewat GLM: $9
+  5 juta lewat MiniMax: $1
+  Total: $10
+```
+
+### Contoh 2: combo 100% gratis
+
+```
+Nama: free-forever
+Model:
+  1. gc/gemini-3-flash (180rb request/bulan gratis)
+  2. if/kimi-k2-thinking (gratis unlimited)
+  3. qw/qwen3-coder-plus (gratis unlimited)
+  4. kr/claude-sonnet-4.5 (gratis unlimited)
+
+Biaya bulanan: $0
+```
+
+### Tips membuat combo
+
+- Urutkan dari kualitas/prioritas tertinggi ke fallback paling murah
+- Selalu taruh minimal satu provider gratis di posisi terakhir
+- Pakai nama combo yang deskriptif agar mudah dipilih dari CLI
+- Aktifkan cloud sync agar combo ikut tersedia di perangkat lain
+
+</details>
+
+---
+
+## 🐳 Deployment
+
+<details>
+<summary><b>Docker</b></summary>
+
+```bash
+docker run -d \
+  --name 9router \
+  -p 20128:20128 \
+  -v 9router-data:/app/data \
+  -e PORT=20128 \
+  -e BASE_URL=http://localhost:20128 \
+  ghcr.io/decolua/9router:latest
+```
+
+Dashboard: `http://localhost:20128/dashboard`
+
+</details>
+
+<details>
+<summary><b>VPS / Cloud</b></summary>
+
+```bash
+npm install -g 9router
+PORT=20128 HOSTNAME=0.0.0.0 BASE_URL=https://your-domain.com 9router
+```
+
+Disarankan menaruhnya di belakang reverse proxy (Nginx/Caddy) dengan HTTPS, dan membatasi akses hanya untuk dirimu sendiri.
+
+</details>
+
+<details>
+<summary><b>Cloudflare Workers</b></summary>
+
+```bash
+npm run build
+npx wrangler deploy
+```
+
+Atur `BASE_URL` dan `CLOUD_URL` sebagai environment variable di dashboard Cloudflare.
+
+</details>
+
+---
+
+## 🧪 Troubleshooting
+
+| Masalah | Kemungkinan Penyebab | Solusi |
+|---------|----------------------|--------|
+| Tool CLI tidak bisa konek | Endpoint salah | Pastikan `http://localhost:20128/v1` |
+| 401 / Unauthorized | API key salah | Salin ulang key dari dashboard |
+| Model tidak ditemukan | Prefix provider salah | Pakai format `provider/model`, mis. `if/kimi-k2-thinking` |
+| Selalu fallback ke gratis | Kuota langganan habis | Cek hitung mundur reset di dashboard |
+| OAuth gagal | Port callback terpakai | Tutup proses lain (mis. port 1455 untuk Codex) |
+| UI menggantung saat sync | DNS/jaringan cloud bermasalah | Cek `CLOUD_URL`; sync memakai timeout fail-fast |
+
+Aktifkan mode debug di dashboard untuk melihat log lengkap request/response.
+
+---
+
+## 🤝 Kontribusi
+
+Kontribusi sangat diterima!
+
+1. Fork repo ini
+2. Buat branch fitur (`git checkout -b feature/nama-fitur`)
+3. Commit perubahanmu (`git commit -m 'feat: tambah fitur X'`)
+4. Push ke branch (`git push origin feature/nama-fitur`)
+5. Buka Pull Request
+
+---
+
+## 📄 Lisensi
+
+MIT License — lihat [LICENSE](https://github.com/decolua/9router/blob/main/LICENSE) untuk detailnya.
+
+---
+
+<div align="center">
+
+**Kalau 9Router membantumu, kasih ⭐ di [GitHub](https://github.com/decolua/9router)!**
+
+[🌐 Website](https://9router.com) • [📦 npm](https://www.npmjs.com/package/9router) • [🐛 Laporkan Bug](https://github.com/decolua/9router/issues)
+
+</div>
--- a/i18n/README.th.md
+++ b/i18n/README.th.md
@@ -0,0 +1,723 @@
+นี่คือเอกสารแปลภาษาไทยของไฟล์ Markdown ต้นฉบับ โดยรักษาโครงสร้างและซินแท็กซ์ทางเทคนิคทั้งหมดไว้เหมือนเดิม
+
+<div align="center">
+  <img src="../images/9router.png?1" alt="แดชบอร์ด 9Router" width="800"/>
+  
+  # 9Router - Free AI Router
+  
+  **ไม่ต้องหยุดเขียนโค้ด ประหยัดโทเค็น 20-40% ด้วย RTK + สลับอัตโนมัติไปยังโมเดล AI ฟรีและราคาถูก**
+  
+  **ผู้ให้บริการ AI ฟรีสำหรับ OpenClaw**
+  
+  <p align="center">
+    <img src="../public/providers/openclaw.png" alt="OpenClaw" width="80"/>
+  </p>
+  
+  [![npm](https://img.shields.io/npm/v/9router.svg)](https://www.npmjs.com/package/9router)
+  [![Downloads](https://img.shields.io/npm/dm/9router.svg)](https://www.npmjs.com/package/9router)
+  [![License](https://img.shields.io/npm/l/9router.svg)](https://github.com/decolua/9router/blob/main/LICENSE)
+  
+  [🚀 เริ่มต้นใช้งาน](#-quick-start) • [💡 ฟีเจอร์](#-key-features) • [📖 การตั้งค่า](#-setup-guide) • [🌐 เว็บไซต์](https://9router.com)
+</div>
+
+---
+
+## 🤔 ทำไมต้อง 9Router?
+
+**หยุดเสียเงินและเจอขีดจำกัด:**
+
+- ❌ โควตาสมาชิกหมดอายุโดยไม่ได้ใช้ทุกเดือน
+- ❌ Rate Limit หยุดคุณระหว่างเขียนโค้ด
+- ❌ ค่า API แพง ($20-50/เดือน ต่อผู้ให้บริการแต่ละราย)
+- ❌ ต้องสลับผู้ให้บริการด้วยตนเอง
+
+**9Router แก้ปัญหาเหล่านี้:**
+
+- ✅ **ประหยัดโทเค็น RTK** - บีบอัดผลลัพธ์จากเครื่องมือ (`git diff`, `grep`, `ls`...) ก่อนส่งให้ LLM
+- ✅ **เพิ่มประสิทธิภาพสมาชิก** - ติดตามโควตา ใช้ทุกบิตก่อนรีเซ็ต
+- ✅ **สลับอัตโนมัติ** - สมาชิก → ถูก → ฟรี, ไม่มีเวลาหยุดทำงาน
+- ✅ **รองรับหลายบัญชี** - Round-robin ระหว่างบัญชีของผู้ให้บริการแต่ละราย
+- ✅ **ใช้งานได้ทุกที่** - ใช้ได้กับ Claude Code, Codex, Cursor, Cline, เครื่องมือ CLI ใดก็ได้
+
+---
+
+## 🔄 วิธีการทำงาน
+
+```
+┌─────────────┐
+│  Your CLI   │  (Claude Code, Codex, OpenClaw, Cursor, Cline...)
+│   Tool      │
+└──────┬──────┘
+       │ http://localhost:20128/v1
+       ↓
+┌─────────────────────────────────────────────┐
+│           9Router (Smart Router)            │
+│  • RTK Token Saver (ตัดโทเค็น tool_result) │
+│  • แปลงรูปแบบ (OpenAI ↔ Claude)           │
+│  • ติดตามโควตา                              │
+│  • รีเฟรชโทเค็นอัตโนมัติ                     │
+└──────┬──────────────────────────────────────┘
+       │
+       ├─→ [Tier 1: สมาชิก] Claude Code, Codex, GitHub Copilot
+       │   ↓ โควตาหมด
+       ├─→ [Tier 2: ถูก] GLM ($0.6/1M), MiniMax ($0.2/1M)
+       │   ↓ งบหมด
+       └─→ [Tier 3: ฟรี] Kiro, OpenCode Free, Vertex ($300 เครดิตฟรี)
+
+ผลลัพธ์: ไม่ต้องหยุดเขียนโค้ด ค่าใช้จ่ายน้อยที่สุด + ประหยัดโทเค็น 20-40% ด้วย RTK
+```
+
+---
+
+## ⚡ เริ่มต้นใช้งาน
+
+**1. ติดตั้งแบบ Global:**
+
+```bash
+npm install -g 9router
+9router
+```
+
+🎉 เปิดแดชบอร์ดที่ `http://localhost:20128`
+
+**2. เชื่อมต่อผู้ให้บริการฟรี (ไม่ต้องสมัคร):**
+
+แดชบอร์ด → Providers → เชื่อมต่อ **Kiro AI** (Claude ฟรีไม่จำกัด) หรือ **OpenCode Free** (ไม่ต้องยืนยันตัวตน) → เสร็จ!
+
+**3. ใช้ในเครื่องมือ CLI ของคุณ:**
+
+```
+ตั้งค่า Claude Code/Codex/OpenClaw/Cursor/Cline:
+  Endpoint: http://localhost:20128/v1
+  API Key: [คัดลอกจากแดชบอร์ด]
+  Model: kr/claude-sonnet-4.5
+```
+
+**เสร็จแล้ว!** เริ่มเขียนโค้ดด้วยโมเดล AI ฟรี
+
+**วิธีอื่น: รันจากซอร์สโค้ด (เก็บรักษาไว้ใน repo นี้):**
+
+Repo นี้เป็น private package (`9router-app`) ดังนั้นการรันจากซอร์ส/Docker คือเส้นทางพัฒนาท้องถิ่นที่คาดไว้
+
+```bash
+cp .env.example .env
+npm install
+PORT=20128 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run dev
+```
+
+โหมด Production:
+
+```bash
+npm run build
+PORT=20128 HOSTNAME=0.0.0.0 NEXT_PUBLIC_BASE_URL=http://localhost:20128 npm run start
+```
+
+URL ค่าเริ่มต้น:
+- แดชบอร์ด: `http://localhost:20128/dashboard`
+- OpenAI-compatible API: `http://localhost:20128/v1`
+
+---
+
+## 🛠️ เครื่องมือ CLI ที่รองรับ
+
+9Router ทำงานได้อย่างราบรื่นกับเครื่องมือเขียนโค้ด AI ทุกประเภท:
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/claude.png" width="60" alt="Claude Code"/><br/>
+        <b>Claude-Code</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/openclaw.png" width="60" alt="OpenClaw"/><br/>
+        <b>OpenClaw</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/codex.png" width="60" alt="Codex"/><br/>
+        <b>Codex</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/opencode.png" width="60" alt="OpenCode"/><br/>
+        <b>OpenCode</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/cursor.png" width="60" alt="Cursor"/><br/>
+        <b>Cursor</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/antigravity.png" width="60" alt="Antigravity"/><br/>
+        <b>Antigravity</b>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/cline.png" width="60" alt="Cline"/><br/>
+        <b>Cline</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/continue.png" width="60" alt="Continue"/><br/>
+        <b>Continue</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/droid.png" width="60" alt="Droid"/><br/>
+        <b>Droid</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/roo.png" width="60" alt="Roo"/><br/>
+        <b>Roo</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/copilot.png" width="60" alt="Copilot"/><br/>
+        <b>Copilot</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/kilocode.png" width="60" alt="Kilo Code"/><br/>
+        <b>Kilo Code</b>
+      </td>
+    </tr>
+  </table>
+</div>
+
+---
+
+## ผู้ให้บริการที่รองรับ
+
+### 🔐 ผู้ให้บริการ OAuth
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="120">
+        <img src="../public/providers/claude.png" width="60" alt="Claude Code"/><br/>
+        <b>Claude-Code</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/antigravity.png" width="60" alt="Antigravity"/><br/>
+        <b>Antigravity</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/codex.png" width="60" alt="Codex"/><br/>
+        <b>Codex</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/github.png" width="60" alt="GitHub"/><br/>
+        <b>GitHub</b>
+      </td>
+      <td align="center" width="120">
+        <img src="../public/providers/cursor.png" width="60" alt="Cursor"/><br/>
+        <b>Cursor</b>
+      </td>
+    </tr>
+  </table>
+</div>
+
+### 🆓 ผู้ให้บริการฟรี
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="150">
+        <img src="../public/providers/kiro.png" width="70" alt="Kiro"/><br/>
+        <b>Kiro AI</b><br/>
+        <sub>Claude 4.5 + GLM-5 + MiniMax • ไม่จำกัด ฟรี</sub>
+      </td>
+      <td align="center" width="150">
+        <img src="../public/providers/opencode.png" width="70" alt="OpenCode"/><br/>
+        <b>OpenCode Free</b><br/>
+        <sub>ไม่ต้องยืนยันตัวตน • ดึงโมเดลอัตโนมัติ • ไม่จำกัด ฟรี</sub>
+      </td>
+      <td align="center" width="150">
+        <img src="../public/providers/gemini.png" width="70" alt="Vertex AI"/><br/>
+        <b>Vertex AI</b><br/>
+        <sub>Gemini 3 Pro + GLM-5 + DeepSeek • เครดิตฟรี $300</sub>
+      </td>
+    </tr>
+  </table>
+</div>
+
+> **หมายเหตุ:** iFlow, Qwen และ Gemini CLI หยุดให้บริการในปี 2026 แล้ว ใช้ Kiro / OpenCode Free / Vertex แทน
+
+### 🔑 ผู้ให้บริการ API Key (40+)
+
+<div align="center">
+  <table>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/openrouter.png" width="50" alt="OpenRouter"/><br/>
+        <sub>OpenRouter</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/glm.png" width="50" alt="GLM"/><br/>
+        <sub>GLM</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/kimi.png" width="50" alt="Kimi"/><br/>
+        <sub>Kimi</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/minimax.png" width="50" alt="MiniMax"/><br/>
+        <sub>MiniMax</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/openai.png" width="50" alt="OpenAI"/><br/>
+        <sub>OpenAI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/anthropic.png" width="50" alt="Anthropic"/><br/>
+        <sub>Anthropic</sub>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/gemini.png" width="50" alt="Gemini"/><br/>
+        <sub>Gemini</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/deepseek.png" width="50" alt="DeepSeek"/><br/>
+        <sub>DeepSeek</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/groq.png" width="50" alt="Groq"/><br/>
+        <sub>Groq</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/xai.png" width="50" alt="xAI"/><br/>
+        <sub>xAI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/mistral.png" width="50" alt="Mistral"/><br/>
+        <sub>Mistral</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/perplexity.png" width="50" alt="Perplexity"/><br/>
+        <sub>Perplexity</sub>
+      </td>
+    </tr>
+    <tr>
+      <td align="center" width="100">
+        <img src="../public/providers/together.png" width="50" alt="Together"/><br/>
+        <sub>Together AI</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/fireworks.png" width="50" alt="Fireworks"/><br/>
+        <sub>Fireworks</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/cerebras.png" width="50" alt="Cerebras"/><br/>
+        <sub>Cerebras</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/cohere.png" width="50" alt="Cohere"/><br/>
+        <sub>Cohere</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/nvidia.png" width="50" alt="NVIDIA"/><br/>
+        <sub>NVIDIA</sub>
+      </td>
+      <td align="center" width="100">
+        <img src="../public/providers/siliconflow.png" width="50" alt="SiliconFlow"/><br/>
+        <sub>SiliconFlow</sub>
+      </td>
+    </tr>
+  </table>
+  <p><i>...และผู้ให้บริการอีกกว่า 20 ราย รวมถึง Nebius, Chutes, Hyperbolic และ OpenAI/Anthropic compatible endpoints แบบกำหนดเอง</i></p>
+</div>
+
+---
+
+## 💡 ฟีเจอร์หลัก
+
+| ฟีเจอร์ | ทำอะไร | ทำไมถึงสำคัญ |
+|---------|--------------|----------------|
+| 🚀 **RTK Token Saver** ([RTK](https://github.com/rtk-ai/rtk) ⭐40K) | บีบอัดผลลัพธ์จากเครื่องมือ (`git diff`, `grep`, `ls`, `tree`...) ก่อนส่งให้ LLM | ประหยัด **โทเค็น input 20-40%** ต่อคำขอ |
+| 🧠 **Headroom Token Saver** ([Headroom](https://github.com/chopratejas/headroom)) | พร็อกซี `/v1/compress` ภายนอกก่อนเลือกผู้ให้บริการ | ประหยัดโทเค็นบริบทมากขึ้นโดยไม่ต้องเปลี่ยน client |
+| 🪨 **Caveman Mode** ([Caveman](https://github.com/JuliusBrussee/caveman) ⭐52K) | ฉีด caveman-speak prompt → LLM ตอบสั้นกระชับ เนื้อหาทางเทคนิคยังครบถ้วน | ประหยัด **โทเค็น output สูงสุด 65%** |
+| 🐴 **Ponytail** ([Ponytail](https://github.com/DietrichGebert/ponytail)) | ฉีด prompt "lazy senior dev" → LLM เขียนโค้ดน้อยที่สุด YAGNI-first (Lite/Full/Ultra) | **โทเค็น output น้อยลง, ไม่ต้อง refactor มาก** |
+| 🎯 **Smart 3-Tier Fallback** | เลือกเส้นทางอัตโนมัติ: สมาชิก → ถูก → ฟรี | ไม่ต้องหยุดเขียนโค้ด, ไม่มีเวลาหยุดทำงาน |
+| 📊 **ติดตามโควตาแบบ Real-Time** | นับโทเค็นแบบ live + นับถอยหลังรีเซ็ต | เพิ่มประสิทธิภาพมูลค่าสมาชิก |
+| 🔄 **แปลงรูปแบบ** | OpenAI ↔ Claude ↔ Gemini ↔ Cursor ↔ Kiro ↔ Vertex | ใช้ได้กับเครื่องมือ CLI ทุกประเภท |
+| 👥 **รองรับหลายบัญชี** | หลายบัญชีต่อผู้ให้บริการ | Load balancing + สำรองข้อมูล |
+| 🔄 **รีเฟรชโทเค็นอัตโนมัติ** | OAuth token รีเฟรชอัตโนมัติ | ไม่ต้องล็อกอินซ้ำด้วยตนเอง |
+| 🎨 **Combo กำหนดเอง** | สร้างการผสมผสานโมเดลไม่จำกัด | ปรับแต่ง fallback ตามความต้องการ |
+| 📝 **บันทึก Request** | โหมด debug พร้อม log request/response ครบถ้วน | แก้ไขปัญหาได้ง่าย |
+| 💾 **ซิงค์คลาวด์** | ซิงค์การตั้งค่าระหว่างอุปกรณ์ | การตั้งค่าเดียวกันทุกที่ |
+| 📊 **วิเคราะห์การใช้งาน** | ติดตามโทเค็น, ค่าใช้จ่าย, แนวโน้มตามเวลา | ปรับแต่งค่าใช้จ่าย |
+| 🌐 **Deploy ได้ทุกที่** | Localhost, VPS, Docker, Cloudflare Workers | ตัวเลือก deploy ที่ยืดหยุ่น |
+
+<details>
+<summary><b>📖 รายละเอียดฟีเจอร์</b></summary>
+
+### 🚀 RTK Token Saver
+
+ผลลัพธ์จากเครื่องมือ (`git diff`, `grep`, `find`, `ls`, `tree`, log dumps...) มักกินงบประมาณ prompt 30-50% RTK ตรวจสอบและบีบอัดอย่างชาญฉลาดแบบ lossless **ก่อน**คำขอถึง LLM:
+
+- **ตัวกรอง:** `git-diff`, `git-status`, `grep`, `find`, `ls`, `tree`, `dedup-log`, `smart-truncate`, `read-numbered`, `search-list`
+- **ตรวจจับอัตโนมัติ:** ไม่ต้องตั้งค่า — RTK .peek 1KB แรกของแต่ละ `tool_result` และเลือกตัวกรองที่ถูกต้อง
+- **ปลอดภัยโดยการออกแบบ:** ถ้าตัวกรองล้มเหลว, ขว้าง error, หรือทำให้ผลลัพธ์ใหญ่ขึ้น RTK จะเก็บข้อความต้นฉบับไว้โดยเงียบๆ ไม่มี error ทำให้คำขอของคุณล้มเหลว
+- **ใช้ได้ทุกที่:** ใช้ได้กับทุกรูปแบบ (OpenAI, Claude, Gemini, Cursor, Kiro, OpenAI Responses) เพราะทำงาน **ก่อน**การแปลงรูปแบบใดๆ
+- **เปิดใช้งานเป็นค่าเริ่มต้น:** ปิด/เปิดได้ตลอดเวลาใน แดชบอร์ด → ตั้งค่า Endpoint
+
+```
+ไม่ใช้ RTK: ส่ง 47K โทเค็นให้ LLM
+ใช้ RTK:    ส่ง 28K โทเค็นให้ LLM   (ประหยัด 40% · บริบทเดียวกัน · คำตอบเดียวกัน)
+```
+
+### 🧠 Headroom Token Saver
+
+Headroom เป็นตัวเลือกและทำงานแยกกัน 9Router เรียก endpoint `/v1/compress` ของ Headroom จากนั้นยังคงเลือกเส้นทาง, fallback, auth และติดตามการใช้งานตามปกติ:
+
+```
+Client → 9Router → Headroom /v1/compress → 9Router → provider
+```
+
+ตั้งค่าท้องถิ่น:
+
+```bash
+pip install "headroom-ai[proxy]"
+headroom proxy --port 8787
+```
+
+เปิดใช้งานใน แดชบอร์ด → Endpoint → Token Saver → Headroom URL ค่าเริ่มต้น: `http://localhost:8787`
+
+ตัวอย่าง Docker:
+
+```bash
+# Headroom service ใน Docker network เดียวกัน
+http://host.docker.internal:8787
+```
+
+ถ้า Headroom ดับหรือคืน error, 9Router จะ fail open และส่งคำขอต้นฉบับ
+
+### 🐴 Ponytail (Lazy Senior Dev)
+
+Ponytail ฉีด prompt *"lazy senior dev"* เข้าไปในทุกคำขอ ทำให้ LLM เขียนโค้ดน้อยที่สุดแบบ YAGNI-first — ลบมากกว่าเพิ่ม, stdlib มากกว่า dep ใหม่, one-liner มากกว่า abstraction
+
+- **Lite** — สร้างตามที่ขอ, บอกชื่อทางเลือกที่ lazy กว่า
+- **Full** — บังคับ YAGNI ladder: stdlib → native → existing deps → one-liner → minimal code
+- **Ultra** — YAGNI extremist: ลบก่อน, ส่ง one-liner, ตั้งคำถามกับ requirement ที่เหลือในคำตอบเดียวกัน
+
+```
+ไม่ใช้ Ponytail: โค้ดเยอะ, abstraction เยอะ, "เผื่อไว้" scaffolding
+ใช้ Ponytail:    diff สั้นที่สุดที่ทำงานได้, ไม่เพิ่ม abstraction ที่ไม่ได้ขอ, โทเค็นน้อยลง
+```
+
+ไม่มีวันแลก: input validation, error handling ที่ป้องกัน data loss, security, accessibility หรือสิ่งที่ขอมาอย่างชัดเจน เปิดใช้งานใน แดชบอร์ด → Endpoint → Ponytail ใช้คู่กับ Caveman (ความกระชับ output) และ RTK (การบีบอัด input) ได้
+
+### 🎯 Smart 3-Tier Fallback
+
+สร้าง combo พร้อม fallback อัตโนมัติ:
+
+```
+Combo: "my-coding-stack"
+  1. cc/claude-opus-4-6        (สมาชิกของคุณ)
+  2. glm/glm-4.7               (สำรองราคาถูก, $0.6/1M)
+  3. if/kimi-k2-thinking       (fallback ฟรี)
+
+→ สลับอัตโนมัติเมื่อโควตาหมดหรือเกิด error
+```
+
+### 📊 ติดตามโควตาแบบ Real-Time
+
+- การใช้โทเค็นต่อผู้ให้บริการ
+- นับถอยหลังรีเซ็ต (5 ชั่วโมง, รายวัน, รายสัปดาห์)
+- ประมาณการค่าใช้จ่ายสำหรับชั้นแบบเสียค่าใช้จ่าย
+- รายงานค่าใช้จ่ายรายเดือน
+
+### 🔄 แปลงรูปแบบ
+
+แปลงรูปแบบได้อย่างราบรื่น:
+- **OpenAI** ↔ **Claude** ↔ **Gemini** ↔ **Cursor** ↔ **Kiro** ↔ **Vertex** ↔ **Antigravity** ↔ **Ollama** ↔ **OpenAI Responses**
+- เครื่องมือ CLI ของคุณส่งรูปแบบ OpenAI → 9Router แปลง → ผู้ให้บริการได้รับรูปแบบต้นฉบับ
+- ใช้ได้กับเครื่องมือใดก็ได้ที่รองรับ custom OpenAI endpoints
+
+### 👥 รองรับหลายบัญชี
+
+- เพิ่มหลายบัญชีสำหรับผู้ให้บริการแต่ละราย
+- เลือกเส้นทาง round-robin หรือตามลำดับความสำคัญอัตโนมัติ
+- Fallback ไปยังบัญชีถัดไปเมื่อบัญชีหนึ่งชนโควตา
+
+### 🔄 รีเฟรชโทเค็นอัตโนมัติ
+
+- OAuth token รีเฟรชอัตโนมัติก่อนหมดอายุ
+- ไม่ต้องยืนยันตัวตนใหม่ด้วยตนเอง
+- ประสบการณ์ที่ราบรื่นบนผู้ให้บริการทุกราย
+
+### 🎨 Combo กำหนดเอง
+
+- สร้างการผสมผสานโมเดลไม่จำกัด
+- ผสมชั้นสมาชิก, ราคาถูกและฟรี
+- ตั้งชื่อ combo เพื่อเข้าถึงง่าย
+- แชร์ combo ระหว่างอุปกรณ์ด้วยการซิงค์คลาวด์
+
+### 📝 บันทึก Request
+
+- เปิดโหมด debug เพื่อดู log request/response ครบถ้วน
+- ติดตาม API calls, headers และ payloads
+- แก้ไขปัญหาการเชื่อมต่อ
+- Export log เพื่อวิเคราะห์
+
+### 💾 ซิงค์คลาวด์
+
+- ซิงค์ผู้ให้บริการ, combo และการตั้งค่าระหว่างอุปกรณ์
+- ซิงค์เบื้องหลังอัตโนมัติ
+- จัดเก็บข้อมูลแบบเข้ารหัสปลอดภัย
+- เข้าถึงการตั้งค่าของคุณจากทุกที่
+
+### 📊 วิเคราะห์การใช้งาน
+
+- ติดตามการใช้โทเค็นตามผู้ให้บริการและโมเดล
+- ประมาณการค่าใช้จ่ายและแนวโน้มค่าใช้จ่าย
+- รายงานและข้อมูลเชิงลึกรายเดือน
+- ปรับแต่งค่าใช้จ่าย AI ของคุณ
+
+### 🌐 Deploy ได้ทุกที่
+
+- 💻 **Localhost** - ค่าเริ่มต้น, ทำงานออฟไลน์
+- ☁️ **VPS/Cloud** - แชร์ระหว่างอุปกรณ์
+- 🐳 **Docker** - Deploy ด้วยคำสั่งเดียว
+- 🚀 **Cloudflare Workers** - เครือข่าย edge ทั่วโลก
+
+</details>
+
+---
+
+## 💰 สรุปราคา
+
+| ประเภท | ผู้ให้บริการ | ค่าใช้จ่าย | รีเซ็ตโควตา | ดีที่สุดสำหรับ |
+|------|----------|------|-------------|----------|
+| **💳 สมาชิก** | Claude Code (Pro) | $20/เดือน | 5 ชม. + รายสัปดาห์ | มีสมาชิกอยู่แล้ว |
+| | Codex (Plus/Pro) | $20-200/เดือน | 5 ชม. + รายสัปดาห์ | ผู้ใช้ OpenAI |
+| | GitHub Copilot | $10-19/เดือน | รายเดือน | ผู้ใช้ GitHub |
+| **💰 ราคาถูก** | GLM-4.7 | $0.6/1M | ทุกวัน 10:00 AM | สำรองงบ |
+| | MiniMax M2.1 | $0.2/1M | 5 ชั่วโมง | ถูกที่สุด |
+| | Kimi K2 | $9/เดือน คงที่ | 10M โทเค็น/เดือน | ค่าใช้จ่ายที่คาดเดาได้ |
+| **🆓 ฟรี** | Kiro | $0 | ไม่จำกัด | Claude ฟรี |
+| | OpenCode Free | $0 | ไม่จำกัด | ไม่ต้องยืนยันตัวตน |
+| | Vertex AI | $0 | $300 เครดิตฟรี | Gemini 3 Pro |
+
+**💡 เคล็ดลับ:** เริ่มจาก combo Kiro (Claude ฟรีไม่จำกัด) + OpenCode Free (ไม่ต้องยืนยันตัวตน) = ค่าใช้จ่าย $0!
+
+---
+
+## 🎯 กรณีการใช้งาน
+
+### กรณีที่ 1: "ฉันมีสมาชิก Claude Pro"
+
+**ปัญหา:** โควตาหมดอายุโดยไม่ได้ใช้, Rate Limit ตอนเขียนโค้ดหนัก
+
+**วิธีแก้:**
+```
+Combo: "maximize-claude"
+  1. cc/claude-opus-4-6        (ใช้สมาชิกเต็มที่)
+  2. glm/glm-4.7               (สำรองราคาถูกเมื่อโควตาหมด)
+  3. kr/claude-sonnet-4.5       (fallback ฉุกเฉินฟรี)
+
+ค่าใช้จ่ายรายเดือน: $20 (สมาชิก) + ~$5 (สำรอง) = $25 รวม
+เทียบกับ $20 + ชนโควตา = ผิดหวัง
+```
+
+### กรณีที่ 2: "ฉันต้องการค่าใช้จ่ายเป็นศูนย์"
+
+**ปัญหา:** ไม่มีงบจ่ายสมาชิก, ต้องการ AI เขียนโค้ดที่เชื่อถือได้
+
+**วิธีแก้:**
+```
+Combo: "free-forever"
+  1. kr/claude-sonnet-4.5       (Claude ฟรีไม่จำกัด)
+  2. oc/*                       (OpenCode Free ไม่ต้องยืนยันตัวตน)
+  3. vertex/gemini-3.1-pro-preview (Vertex $300 เครดิตฟรี)
+
+ค่าใช้จ่ายรายเดือน: $0
+คุณภาพ: โมเดลพร้อมใช้งาน production
+```
+
+### กรณีที่ 3: "ฉันต้องเขียนโค้ด 24/7 ไม่มีสะดุด"
+
+**ปัญหา:** Deadline, ไม่สามารถหยุดทำงานได้
+
+**วิธีแก้:**
+```
+Combo: "always-on"
+  1. cc/claude-opus-4-6        (คุณภาพดีที่สุด)
+  2. cx/gpt-5.5                (สมาชิกที่สอง)
+  3. glm/glm-5.1               (ราคาถูก, รีเซ็ตทุกวัน)
+  4. minimax/MiniMax-M2.7      (ถูกที่สุด, รีเซ็ต 5 ชม.)
+  5. kr/claude-sonnet-4.5       (ฟรีไม่จำกัด)
+
+ผลลัพธ์: 5 ชั้น fallback = ไม่มีเวลาหยุดทำงาน
+ค่าใช้จ่ายเดือน: $20-200 (สมาชิก) + $10-20 (สำรอง)
+```
+
+### กรณีที่ 4: "ฉันต้องการ AI ฟรีใน OpenClaw"
+
+**ปัญหา:** ต้องการ AI assistant ในแอปพลิเคชันแชท (WhatsApp, Telegram, Slack...), ฟรีทั้งหมด
+
+**วิธีแก้:**
+```
+Combo: "openclaw-free"
+  1. kr/claude-sonnet-4.5       (Claude ฟรีไม่จำกัด)
+  2. kr/glm-5                   (GLM ฟรีไม่จำกัด)
+  3. kr/MiniMax-M2.5            (MiniMax ฟรีไม่จำกัด)
+
+ค่าใช้จ่ายรายเดือน: $0
+เข้าถึงผ่าน: WhatsApp, Telegram, Slack, Discord, iMessage, Signal...
+```
+
+---
+
+## ❓ คำถามที่พบบ่อย
+
+<details>
+<summary><b>💳 9Router เก็บเงินฉันหรือไม่?</b></summary>
+
+**ไม่.** 9Router เป็นซอฟต์แวร์ฟรีแบบ open source ที่ทำงานบนเครื่องของคุณเอง มันไม่มีวันเรียกเก็บเงินจากคุณ
+
+**คุณจ่ายเงินเฉพาะ:**
+- ✅ **ผู้ให้บริการสมาชิก** (Claude Code $20/เดือน, Codex $20-200/เดือน) → จ่ายตรงให้พวกเขาบนเว็บไซต์ของพวกเขา
+- ✅ **ผู้ให้บริการราคาถูก** (GLM, MiniMax) → จ่ายตรงให้พวกเขา, 9Router แค่เลือกเส้นทางคำขอของคุณ
+- ❌ **ตัว 9Router เอง** → **ไม่มีวันเรียกเก็บเงินใดๆ ทั้งสิ้น**
+
+9Router เป็น proxy/router ท้องถิ่น มันไม่มีบัตรเครดิตของคุณ, ไม่สามารถส่งใบแจ้งหนี้ได้ และไม่มีระบบชำระเงิน เป็นซอฟต์แวร์ฟรีทั้งหมด
+
+</details>
+
+<details>
+<summary><b>🆓 ผู้ให้บริการฟรีไม่จำกัดจริงหรือ?</b></summary>
+
+**จริง!** ผู้ให้บริการที่ระบุว่าฟรี (Kiro, OpenCode Free, Vertex) ไม่จำกัดจริงๆ **ไม่มีค่าใช้จ่ายแอบแฝง**
+
+นี่คือบริการฟรีที่บริษัทต่างๆ ให้บริการ:
+- **Kiro**: Claude ฟรีไม่จำกัดผ่าน AWS Builder ID
+- **OpenCode Free**: ไม่ต้องยืนยันตัวตน, ดึงโมเดลอัตโนมัติ
+- **Vertex AI**: $300 เครดิตฟรีสำหรับ Gemini 3 Pro
+
+9Router แค่เลือกเส้นทางคำขอของคุณไปหาพวกเขา — ไม่มี "กับดัก" หรือการเรียกเก็บเงินในอนาคต เป็นบริการที่ฟรีจริงๆ และ 9Router ทำให้ใช้งานง่ายด้วยการรองรับ fallback
+
+</details>
+
+<details>
+<summary><b>💰 ทำอย่างไรเพื่อลดค่าใช้จ่าย AI จริงของฉัน?</b></summary>
+
+**กลยุทธ์ Free First:**
+
+1. **เริ่มจาก combo ฟรี 100%:**
+   ```
+   1. kr/claude-sonnet-4.5 (Claude ฟรีไม่จำกัด)
+   2. oc/* (OpenCode Free ไม่ต้องยืนยันตัวตน)
+   3. vertex/gemini-3.1-pro-preview ($300 เครดิตฟรี)
+   ```
+   **ค่าใช้จ่าย: $0/เดือน**
+
+2. **เพิ่มสำรองราคาถูก** เมื่อจำเป็นเท่านั้น:
+   ```
+   4. glm/glm-5.1 ($0.6/1M โทเค็น)
+   ```
+   **ค่าใช้จ่ายเพิ่มเติม:** จ่ายเฉพาะที่ใช้
+
+3. **ใช้ผู้ให้บริการสมาชิก** ก็ต่อเมื่อมีอยู่แล้ว:
+   - 9Router ช่วยเพิ่มประสิทธิภาพมูลค่าของพวกเขาผ่านการติดตามโควตา
+
+**ผลลัพธ์:** ผู้ใช้ส่วนใหญ่สามารถทำงานที่ $0/เดือน โดยใช้เฉพาะชั้นฟรี!
+
+</details>
+
+---
+
+## 🐛 การแก้ไขปัญหา
+
+**"Language model did not provide messages"**
+- โควตาผู้ให้บริการหมด → ตรวจสอบตัวติดตามโควตาในแดชบอร์ด
+- วิธีแก้: ใช้ combo fallback หรือสลับไปชั้นที่ถูกกว่า
+
+**Rate Limiting**
+- สมาชิกหมดโควตา → Fallback ไป GLM/MiniMax
+- เพิ่ม combo: `cc/claude-opus-4-6 → glm/glm-5.1 → kr/claude-sonnet-4.5`
+
+**OAuth Token หมดอายุ**
+- รีเฟรชอัตโนมัติโดย 9Router
+- ถ้าปัญหายังคงอยู่: แดชบอร์ด → ผู้ให้บริการ → เชื่อมต่อใหม่
+
+**ค่าใช้จ่ายสูง**
+- เปิดใช้ RTK ใน แดชบอร์ด → ตั้งค่า Endpoint (เปิดเป็นค่าเริ่มต้น, ประหยัด 20-40% โทเค็น)
+- ตรวจสอบสถิติการใช้งานในแดชบอร์ด
+- สลับโมเดลหลักไป GLM/MiniMax
+- ใช้ชั้นฟรี (Kiro, OpenCode Free, Vertex) สำหรับงานที่ไม่สำคัญ
+
+**แดชบอร์ดเปิดผิดพอร์ต**
+- ตั้ง `PORT=20128` และ `NEXT_PUBLIC_BASE_URL=http://localhost:20128`
+
+**ล็อกอินครั้งแรกไม่ทำงาน**
+- ตรวจสอบ `INITIAL_PASSWORD` ใน `.env`
+- ถ้ายังไม่ตั้งค่า รหัสผ่านสำรองคือ `123456`
+
+**ไม่มี request log ใต้ `logs/`**
+- ตั้ง `ENABLE_REQUEST_LOGS=true`
+
+---
+
+## 🛠️ Tech Stack
+
+- **Runtime**: Node.js 20+
+- **Framework**: Next.js 16
+- **UI**: React 19 + Tailwind CSS 4
+- **Database**: SQLite (better-sqlite3 / node:sqlite / sql.js fallback)
+- **Streaming**: Server-Sent Events (SSE)
+- **Auth**: OAuth 2.0 (PKCE) + JWT + API Keys
+
+---
+
+## 📝 API Reference
+
+### Chat Completions
+
+```bash
+POST http://localhost:20128/v1/chat/completions
+Authorization: Bearer your-api-key
+Content-Type: application/json
+
+{
+  "model": "cc/claude-opus-4-6",
+  "messages": [
+    {"role": "user", "content": "เขียนฟังก์ชันเพื่อ..."}
+  ],
+  "stream": true
+}
+```
+
+### List Models
+
+```bash
+GET http://localhost:20128/v1/models
+Authorization: Bearer your-api-key
+
+→ คืนค่าโมเดลทั้งหมด + combo ในรูปแบบ OpenAI
+```
+
+---
+
+## 📧 สนับสนุน
+
+- **เว็บไซต์**: [9router.com](https://9router.com)
+- **GitHub**: [github.com/decolua/9router](https://github.com/decolua/9router)
+- **Issues**: [github.com/decolua/9router/issues](https://github.com/decolua/9router/issues)
+
+---
+
+## 👥 ผู้มีส่วนร่วม
+
+ขอขอบคุณผู้มีส่วนร่วมทุกคนที่ช่วยทำให้ 9Router ดียิ่งขึ้น!
+
+[![Contributors](https://contrib.rocks/image?repo=decolua/9router&max=150&columns=15&anon=1)](https://github.com/decolua/9router/graphs/contributors)
+
+---
+
+## 📄 ลิขสิทธิ์
+
+MIT License - ดู [LICENSE](../LICENSE) สำหรับรายละเอียด
+
+---
+
+<div align="center">
+  <sub>สร้างด้วย ❤️ สำหรับนักพัฒนาที่เขียนโค้ด 24/7</sub>
+</div>
--- a/next.config.mjs
+++ b/next.config.mjs
@@ -30,6 +30,8 @@ const nextConfig = {
    proxyClientMaxBodySize,
    // Cache fetch responses across HMR refreshes for faster dev reloads.
    serverComponentsHmrCache: true,
+    // Tree-shake heavy barrel imports to cut compile + bundle size
+    optimizePackageImports: ["@xyflow/react", "@dnd-kit/core", "@dnd-kit/sortable", "material-symbols", "marked"],
  },
  webpack: (config, { isServer }) => {
    // Ignore fs/path modules in browser bundle
--- a/open-sse/AGENTS.md
+++ b/open-sse/AGENTS.md
@@ -37,3 +37,7 @@ Provider-agnostic SSE engine: one OpenAI-style request → any provider (LLM cha
 - `registry/index.js` is an auto-generated static import list; regenerate it (don't hand-edit) after adding a `registry/{id}.js`. REGISTRY_TEMPLATE is excluded by design.
 - Special binary/protobuf formats (kiro EventStream, cursor protobuf, commandcode NDJSON) don't round-trip through OpenAI — handle in their executor.
 - `rtk/` + `headroom.js` mutate the request body in-place and are **fail-open**: any error returns null and leaves the body untouched — never throw out of them. RTK skips `is_error`/`status:"error"` tool results to preserve traces.
+- **HTTP 200 in-stream errors**: some upstreams signal failure INSIDE a 200 stream (AI SDK v5 `{"type":"error"}` events, error text in content). HTTP-level success checks miss these → no fallback, `Status: success` in logs. Three hook points + one config escape hatch:
+  1. **Translator** — never map an error event to content. Emit an OpenAI-shaped `chunk.error = { message, type }` + terminal chunk (`translator/response/commandcode-to-openai.js` is the worked example). Downstream `parseSSEToOpenAIResponse` already detects `chunk?.error`.
+  2. **Executor early-peek** — for streaming fallback, read the first events BEFORE returning the response; an error → non-ok Response (`executors/commandcode.js` `peekForUpstreamError`).
+  3. **Config escape hatch (no code)** — per-provider `streamErrorPatterns` setting (UI: provider page → Stream Error Patterns). Patterns matched against the first ~8KB of the stream and the assembled non-streaming content; see `utils/streamErrorPeek.js` + `utils/streamErrorPatterns.js`.
--- a/open-sse/config/appConstants.js
+++ b/open-sse/config/appConstants.js
@@ -1,5 +1,7 @@
-import { platform, arch } from "os";
+import { platform, arch, hostname } from "os";
 import { PROVIDERS, PROVIDER_OAUTH } from "./providers.js";
+import { ANTIGRAVITY_IDE_USER_AGENT } from "../providers/shared.js";
+import { createRequire } from "module";

 // === Gemini CLI === derive từ registry gemini-cli.transport
 export const GEMINI_CLI_VERSION = PROVIDERS["gemini-cli"]?.cliVersion;
@@ -59,7 +61,7 @@ export function getPlatformEnum() {
 }

 export function getPlatformUserAgent() {
-  return `antigravity/1.104.0 ${platform()}/${arch()}`;
+  return ANTIGRAVITY_IDE_USER_AGENT;
 }

 export const CLIENT_METADATA = {
@@ -129,13 +131,22 @@ export const AG_DEFAULT_TOOLS = new Set([

 // Antigravity chat/stream headers
 export const ANTIGRAVITY_HEADERS = {
-  "User-Agent": `antigravity/1.107.0 ${platform()}/${arch()}`
+  "User-Agent": ANTIGRAVITY_IDE_USER_AGENT
 };

-// Cloud Code Assist API
+// Cloud Code Assist API endpoints differ by client ecosystem.
 export const CLOUD_CODE_API = {
-  loadCodeAssist: "https://cloudcode-pa.googleapis.com/v1internal:loadCodeAssist",
-  onboardUser: "https://cloudcode-pa.googleapis.com/v1internal:onboardUser",
+  "gemini-cli": {
+    loadCodeAssist: "https://cloudcode-pa.googleapis.com/v1internal:loadCodeAssist",
+    onboardUser: "https://cloudcode-pa.googleapis.com/v1internal:onboardUser",
+  },
+  // Project discovery (loadCodeAssist/onboardUser) stays on PROD — the daily host
+  // rejects these auth/onboarding calls. Only chat traffic uses the daily host
+  // (see transport.apiEndpoint in registry/antigravity.js, set to bypass prod 429).
+  antigravity: {
+    loadCodeAssist: "https://cloudcode-pa.googleapis.com/v1internal:loadCodeAssist",
+    onboardUser: "https://cloudcode-pa.googleapis.com/v1internal:onboardUser",
+  },
 };

 export const LOAD_CODE_ASSIST_HEADERS = {
@@ -170,12 +181,44 @@ export const OAUTH_ENDPOINTS = {
  github:    { token: PROVIDER_OAUTH["github"]?.tokenUrl, auth: PROVIDER_OAUTH["github"]?.authorizeUrl, deviceCode: PROVIDER_OAUTH["github"]?.deviceCodeUrl },
 };

-// Generate Kimi OAuth custom headers
-export function buildKimiHeaders() {
+let _appVersion;
+function getAppPackageVersion() {
+  if (_appVersion) return _appVersion;
+  try {
+    const require = createRequire(import.meta.url);
+    _appVersion = require("../../package.json").version || "0.0.0";
+  } catch {
+    _appVersion = process.env.npm_package_version || "0.0.0";
+  }
+  return _appVersion;
+}
+
+// Kimi Code OAuth / API headers (CLIProxyAPI internal/auth/kimi commonHeaders parity).
+// deviceId must stay stable per connection for the whole OAuth session.
+export function buildKimiHeaders(deviceId) {
+  const osName = platform();
+  const architecture = arch();
+  let deviceModel = `${osName} ${architecture}`;
+  if (osName === "darwin") deviceModel = `macOS ${architecture}`;
+  else if (osName === "win32") deviceModel = `Windows ${architecture}`;
+  else if (osName === "linux") deviceModel = `Linux ${architecture}`;
+
+  let deviceName = "unknown";
+  try {
+    deviceName = hostname() || "unknown";
+  } catch {
+    deviceName = "unknown";
+  }
+
+  const resolvedId = (typeof deviceId === "string" && deviceId.trim())
+    ? deviceId.trim()
+    : `kimi-${Date.now()}`;
+
  return {
    "X-Msh-Platform": "9router",
-    "X-Msh-Version": "2.1.2",
-    "X-Msh-Device-Model": typeof process !== "undefined" ? `${process.platform} ${process.arch}` : "unknown",
-    "X-Msh-Device-Id": `kimi-${Date.now()}`
+    "X-Msh-Version": getAppPackageVersion(),
+    "X-Msh-Device-Name": deviceName,
+    "X-Msh-Device-Model": deviceModel,
+    "X-Msh-Device-Id": resolvedId,
  };
 }
--- a/open-sse/config/grokCli.js
+++ b/open-sse/config/grokCli.js
@@ -0,0 +1,10 @@
+export const GROK_CLI_VERSION = "0.2.99";
+export const GROK_CLI_MODEL = "grok-build";
+export const GROK_CLI_BASE_URL = "https://cli-chat-proxy.grok.com/v1";
+export const GROK_CLI_CLIENT_IDENTIFIER = "grok-shell";
+export const GROK_CLI_USER_AGENT = `grok-shell/${GROK_CLI_VERSION} (linux; x86_64)`;
+
+export function supportsGrokCliReasoningEffort(model) {
+  // ponytail: unknown models omit effort until live metadata reaches dispatch.
+  return /^grok-4\.5(?:$|-)/.test(String(model || ""));
+}
--- a/open-sse/config/kiroConstants.js
+++ b/open-sse/config/kiroConstants.js
@@ -8,18 +8,24 @@
 *   - `-agentic` model suffix detection + chunked-write system prompt
 *   - reasoning / thinking trigger detection (Anthropic-Beta header,
 *     Claude `thinking`, OpenAI `reasoning_effort`, AMP/Cursor magic tag)
- *   - the `<thinking_mode>enabled</thinking_mode>` system-prompt injection
- *     that turns Kiro reasoning on
+ *   - schema-specific native effort fields for supported GPT and Claude models
+ *   - legacy `<thinking_mode>` system-prompt injection for other models
 *
 * Kiro upstream does not advertise `-agentic` model IDs; they are a 9router
 * fiction. The suffix is stripped before the request leaves this process.
 */

-import { extractThinking } from "../translator/concerns/thinkingUnified.js";
+import { extractThinking, parseSuffix } from "../translator/concerns/thinkingUnified.js";
 import { effortToBudget } from "../translator/concerns/thinking.js";

 export const KIRO_AGENTIC_SUFFIX = "-agentic";
 export const KIRO_THINKING_SUFFIX = "-thinking";
+export const KIRO_TOOL_NAME_MAX_LENGTH = 64;
+export const KIRO_TOOL_DESCRIPTION_MAX_LENGTH = 10237;
+export const KIRO_TOOL_ID_MAX_LENGTH = 64;
+export const KIRO_CODEWHISPERER_TARGET =
+  "AmazonCodeWhispererStreamingService.GenerateAssistantResponse";
+export const KIRO_ENDPOINT_FALLBACK_STATUSES = new Set([401, 403, 404]);

 // Public default CodeWhisperer profile ARNs (us-east-1), keyed by auth method.
 // Used when an account cannot resolve its own profileArn. Builder ID and social
@@ -40,6 +46,39 @@ export function resolveDefaultProfileArn(authMethod) {

 export const KIRO_THINKING_BUDGET_DEFAULT = 16000;

+/**
+ * Resolve a Kiro model after consuming the generic model(level) suffix.
+ * The suffix is a 9router request override, not part of Kiro's upstream model id.
+ */
+export function resolveKiroModelIntent(model) {
+  const { cleanModel, override } = parseSuffix(model);
+  return {
+    model: cleanModel,
+    ...resolveKiroModel(cleanModel),
+    thinkingOverride: override,
+  };
+}
+
+/** Apply a parsed model(level) override without mutating the caller's body. */
+export function applyKiroThinkingOverride(body, override) {
+  if (!override) return body;
+
+  const next = { ...body };
+  if (override.mode === "budget") {
+    delete next.output_config;
+    delete next.reasoning_effort;
+    delete next.reasoning;
+    next.thinking = { type: "enabled", budget_tokens: override.budget };
+    return next;
+  }
+
+  next.output_config = {
+    ...(body.output_config || {}),
+    effort: override.mode === "level" ? override.level : override.mode,
+  };
+  return next;
+}
+
 export const KIRO_AGENTIC_SYSTEM_PROMPT = `
 # CRITICAL: CHUNKED WRITE PROTOCOL (MANDATORY)

@@ -109,6 +148,7 @@ export function resolveKiroThinkingBudget(body, headers, model) {
  const cfg = extractThinking(body);
  if (cfg) {
    if (cfg.mode === "none") return null;
+    if (cfg.mode === "level" && cfg.level === "disabled") return null;
    if (cfg.mode === "budget") return cfg.budget;
    if (cfg.mode === "level") return effortToBudget(cfg.level) ?? KIRO_THINKING_BUDGET_DEFAULT;
    return KIRO_THINKING_BUDGET_DEFAULT;
@@ -131,6 +171,86 @@ export function resolveKiroThinkingBudget(body, headers, model) {
  return null;
 }

+export function extractKiroEffortLevel(body) {
+  const effort =
+    body?.output_config?.effort ??
+    body?.reasoning_effort ??
+    (typeof body?.reasoning === "object" ? body.reasoning?.effort : null);
+  if (typeof effort !== "string") return null;
+  const normalized = effort.toLowerCase();
+  if (normalized === "none" || normalized === "off" || normalized === "disabled") return null;
+  if (normalized === "xhigh" || normalized === "max") return "high";
+  if (["low", "medium", "high"].includes(normalized)) return normalized;
+  return null;
+}
+
+function extractKiroGptEffortLevel(body) {
+  const effort =
+    body?.output_config?.effort ??
+    body?.reasoning_effort ??
+    (typeof body?.reasoning === "object" ? body.reasoning?.effort : null);
+  if (typeof effort !== "string") return null;
+  const normalized = effort.toLowerCase();
+  if (normalized === "max") return "xhigh";
+  // Kiro CLI does not advertise an explicit GPT "none" wire value; omit it.
+  if (["low", "medium", "high", "xhigh"].includes(normalized)) {
+    return normalized;
+  }
+  return null;
+}
+
+export function buildKiroAdditionalModelRequestFields(body, effortPath = "output_config") {
+  const effort = effortPath === "reasoning"
+    ? extractKiroGptEffortLevel(body)
+    : extractKiroEffortLevel(body);
+  if (!effort) return undefined;
+  if (effortPath === "reasoning") {
+    // Mirrors Kiro CLI/KAS buildEffortRequestFields("reasoning") for GPT.
+    return { reasoning: { effort } };
+  }
+  // Mirrors Kiro CLI/KAS buildEffortRequestFields("output_config").
+  return {
+    thinking: { type: "adaptive", display: "summarized" },
+    output_config: { effort },
+  };
+}
+
+export function resolveKiroEffortPath(model) {
+  if (typeof model !== "string") return null;
+  const normalized = model.toLowerCase().replace(/-/g, ".");
+  if (/(?:^|[/.])gpt[/.]5[/.]6(?:[/.]|$)/.test(normalized)) {
+    return "reasoning";
+  }
+  if (!normalized.includes("claude")) return null;
+  const match = normalized.match(/(?:^|[/.])claude(?:[/.][a-z]+)*[/.](\d+)(?:[/.](\d+))?(?:[/.]|$)/);
+  if (!match) return null;
+  const [, majorText, minorText] = match;
+  const major = Number(majorText);
+  const minor = minorText === undefined ? null : Number(minorText);
+  const dateSuffixMinor = minor !== null && minor >= 1000;
+  // Kiro rejected additionalModelRequestFields on legacy 4.5 models in live smoke.
+  // Default future Claude/Kiro models to supported so new model releases do not
+  // need a code allowlist update.
+  return major < 4 || (major === 4 && (minor === null || minor <= 5 || dateSuffixMinor))
+    ? null
+    : "output_config";
+}
+
+export function supportsKiroAdditionalModelRequestFields(model) {
+  return resolveKiroEffortPath(model) !== null;
+}
+
+export function usesKiroNativeGptEffort(body, model) {
+  return resolveKiroEffortPath(model) === "reasoning"
+    && extractKiroGptEffortLevel(body) !== null;
+}
+
+export function buildKiroAdditionalModelRequestFieldsForModel(body, model) {
+  const effortPath = resolveKiroEffortPath(model);
+  if (!effortPath) return undefined;
+  return buildKiroAdditionalModelRequestFields(body, effortPath);
+}
+
 /**
 * Detect whether an inbound request is asking for reasoning / thinking output.
 * Thin wrapper over resolveKiroThinkingBudget (single source of truth).
--- a/open-sse/config/providerModels.js
+++ b/open-sse/config/providerModels.js
@@ -2,9 +2,8 @@ import { PROVIDERS } from "./providers.js";
 import REGISTRY from "../providers/registry/index.js";
 // PROVIDER_MODELS now built from providers/registry (transport + models co-located)
 import { PROVIDER_MODELS } from "../providers/index.js";
-import { modelQuotaFamily, modelStrip, modelTargetFormat } from "../providers/models/schema.js";
+import { modelQuotaFamily, modelStrip, modelTargetFormat, normalizeModelId } from "../providers/models/schema.js";
 import { CODEX_REVIEW_SUFFIX } from "../providers/models/helpers.js";
-
 export { PROVIDER_MODELS };


@@ -18,46 +17,74 @@ export function getDefaultModel(aliasOrId) {
  return models?.[0]?.id || null;
 }

+// Providers whose registry uses dots in version numbers (e.g. "claude-sonnet-4.5").
+// For these, we tolerate clients sending dashes ("claude-sonnet-4-5") by normalizing
+// digit-hyphen-digit to digit-dot-digit before lookup. Other providers are left untouched.
+const DOT_VERSION_PROVIDERS = new Set(["kr", "kiro"]);
+
+// Find a registry entry by id. For Kiro models, tolerates dash/dot version separators
+// ("claude-sonnet-4-5" ~= "claude-sonnet-4.5"). Other providers use exact match only.
+function findModel(models, modelId, aliasOrId) {
+  if (!models) return undefined;
+  const found = models.find(m => m.id === modelId);
+  if (found) return found;
+  if (!DOT_VERSION_PROVIDERS.has(aliasOrId)) return undefined;
+  const normalized = normalizeModelId(modelId);
+  if (normalized === modelId) return undefined;
+  return models.find(m => m.id === normalized);
+}
+
 export function isValidModel(aliasOrId, modelId, passthroughProviders = new Set()) {
  if (passthroughProviders.has(aliasOrId)) return true;
  const models = PROVIDER_MODELS[aliasOrId];
  if (!models) return false;
-  return models.some(m => m.id === modelId);
+  return !!findModel(models, modelId, aliasOrId);
 }

 export function findModelName(aliasOrId, modelId) {
  const models = PROVIDER_MODELS[aliasOrId];
  if (!models) return modelId;
-  const found = models.find(m => m.id === modelId);
+  const found = findModel(models, modelId, aliasOrId);
  return found?.name || modelId;
 }

 export function getModelTargetFormat(aliasOrId, modelId) {
  const models = PROVIDER_MODELS[aliasOrId];
  if (!models) return null;
-  return modelTargetFormat(models.find(m => m.id === modelId));
+  return modelTargetFormat(findModel(models, modelId, aliasOrId));
 }

 export function getModelType(aliasOrId, modelId) {
  const models = PROVIDER_MODELS[aliasOrId];
  if (!models) return null;
-  const found = models.find(m => m.id === modelId);
+  const found = findModel(models, modelId, aliasOrId);
  return found?.kind || found?.type || null;
 }

 export function getModelUpstreamId(aliasOrId, modelId) {
+  // Split off thinking suffix "(level)" so lookup hits the base id; re-append it to
+  // the result so downstream applyThinking still sees the suffix (body.model is stripped separately).
+  const sufMatch = typeof modelId === "string" ? modelId.match(/\([^()]+\)\s*$/) : null;
+  const suffix = sufMatch ? sufMatch[0] : "";
+  const baseId = suffix ? modelId.slice(0, sufMatch.index).trim() : modelId;
  const models = PROVIDER_MODELS[aliasOrId];
-  const found = models?.find(m => m.id === modelId);
-  if (found?.upstreamModelId) return found.upstreamModelId;
-  if (aliasOrId === "cx" && typeof modelId === "string" && modelId.endsWith(CODEX_REVIEW_SUFFIX)) {
-    return modelId.slice(0, -CODEX_REVIEW_SUFFIX.length);
+  const found = findModel(models, baseId, aliasOrId);
+  const resolvedId = found?.upstreamModelId || found?.id;
+  if (resolvedId) {
+    const presetMatch = resolvedId.match(/\([^()]+\)\s*$/);
+    const presetSuffix = presetMatch?.[0] || "";
+    const resolvedBase = presetSuffix ? resolvedId.slice(0, presetMatch.index).trim() : resolvedId;
+    return resolvedBase + (suffix || presetSuffix);
  }
-  return modelId;
+  if (aliasOrId === "cx" && typeof baseId === "string" && baseId.endsWith(CODEX_REVIEW_SUFFIX)) {
+    return baseId.slice(0, -CODEX_REVIEW_SUFFIX.length) + suffix;
+  }
+  return baseId + suffix;
 }

 export function getModelQuotaFamily(aliasOrId, modelId) {
  const models = PROVIDER_MODELS[aliasOrId];
-  return modelQuotaFamily(models?.find(m => m.id === modelId));
+  return modelQuotaFamily(findModel(models, modelId, aliasOrId));
 }

 // OAuth short aliases — derived from registry `alias` (single source). everything else: alias = id.
@@ -79,5 +106,5 @@ export function getModelsByProviderId(providerId) {
 // Get strip list for a model entry (explicit opt-in only)
 // Returns array of content types to strip, e.g. ["image", "audio"]
 export function getModelStrip(alias, modelId) {
-  return modelStrip(PROVIDER_MODELS[alias]?.find(m => m.id === modelId));
+  return modelStrip(findModel(PROVIDER_MODELS[alias], modelId, alias));
 }
--- a/open-sse/config/runtimeConfig.js
+++ b/open-sse/config/runtimeConfig.js
@@ -39,6 +39,15 @@ function envMs(name, def) {
  return Number.isFinite(n) && n > 0 ? n : def;
 }

+function envUrl(name, def) {
+  const raw = process.env[name]?.trim();
+  return raw || def;
+}
+
+// SearXNG endpoint used by the unauthenticated web-search provider.
+// Configure this for a separate Docker service or remote SearXNG instance.
+export const SEARXNG_URL = envUrl("SEARXNG_URL", "http://localhost:8888/search");
+
 // Inter-chunk stall timeout (once tokens are flowing). Generous headroom so
 // slow reasoning models aren't aborted mid-stream. Env: STREAM_STALL_TIMEOUT_MS.
 export const STREAM_STALL_TIMEOUT_MS = envMs("STREAM_STALL_TIMEOUT_MS", 360 * 1000);
@@ -56,6 +65,8 @@ export const GEMINI_NATIVE_TTS_FETCH_TIMEOUT_MS = envMs("GEMINI_NATIVE_TTS_FETCH
 export const DEFAULT_MAX_TOKENS = 64000;
 export const DEFAULT_MIN_TOKENS = 32000;

+export const TOKEN_SAVER_HEADER = "x-9router-token-saver";
+
 // Retry config for 429 responses (legacy - kept for backward compatibility)
 export const RETRY_CONFIG = {
  maxAttempts: 2,
--- a/open-sse/executors/antigravity.js
+++ b/open-sse/executors/antigravity.js
@@ -1,11 +1,12 @@
 import crypto from "crypto";
 import { BaseExecutor } from "./base.js";
 import { PROVIDERS } from "../config/providers.js";
-import { OAUTH_ENDPOINTS, ANTIGRAVITY_HEADERS, INTERNAL_REQUEST_HEADER, AG_DEFAULT_TOOLS, AG_TOOL_SUFFIX } from "../config/appConstants.js";
+import { OAUTH_ENDPOINTS, ANTIGRAVITY_HEADERS, AG_DEFAULT_TOOLS, AG_TOOL_SUFFIX } from "../config/appConstants.js";
 import { HTTP_STATUS } from "../config/runtimeConfig.js";
 import { resolveSessionId } from "../utils/sessionManager.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
 import { cleanJSONSchemaForAntigravity } from "../translator/formats/gemini.js";
+import { DEFAULT_THINKING_AG_SIGNATURE } from "../config/defaultThinkingSignature.js";

 // Sanitize function name: Gemini requires [a-zA-Z_][a-zA-Z0-9_.:\-]{0,63}
 function sanitizeFunctionName(name) {
@@ -17,7 +18,8 @@ function sanitizeFunctionName(name) {

 const MAX_RETRY_AFTER_MS = 10000;
 const ANTIGRAVITY_TRANSIENT_RETRY_MAX_MS = 15000;
-const MAX_ANTIGRAVITY_OUTPUT_TOKENS = 16384;
+const MAX_ANTIGRAVITY_OUTPUT_TOKENS = 64000;
+const ANTIGRAVITY_IDE_REQUEST_ID_RE = /^agent\/[^/]+\/\d+\/[^/]+\/\d+$/;

 const ANTIGRAVITY_TRANSIENT_ERROR_PATTERNS = [
  /high\s+traffic/i,
@@ -86,6 +88,27 @@ function parseImageConfig(model) {
  return config;
 }

+function uuidFromSeed(seed) {
+  const bytes = crypto.createHash("sha256").update(String(seed || "antigravity")).digest().subarray(0, 16);
+  bytes[6] = (bytes[6] & 0x0f) | 0x50;
+  bytes[8] = (bytes[8] & 0x3f) | 0x80;
+  const hex = bytes.toString("hex");
+  return `${hex.slice(0, 8)}-${hex.slice(8, 12)}-${hex.slice(12, 16)}-${hex.slice(16, 20)}-${hex.slice(20)}`;
+}
+
+function buildIdeRequestId({ body, request, credentials, model, requestType }) {
+  if (ANTIGRAVITY_IDE_REQUEST_ID_RE.test(body?.requestId || "")) {
+    return body.requestId;
+  }
+
+  const sessionId = request?.sessionId || body?.request?.sessionId || credentials?._clientSessionId || credentials?.connectionId || credentials?.email || "anonymous";
+  const conversationId = uuidFromSeed(`antigravity:conversation:${sessionId}`);
+  const trajectoryId = uuidFromSeed(`antigravity:trajectory:${sessionId}:${model}:${requestType}`);
+  const contentCount = Array.isArray(request?.contents) ? request.contents.length : 1;
+  const step = Math.max(1, contentCount * 2 - 1);
+  return `agent/${conversationId}/${Date.now()}/${trajectoryId}/${step}`;
+}
+
 export class AntigravityExecutor extends BaseExecutor {
  constructor() {
    super("antigravity", PROVIDERS.antigravity);
@@ -103,20 +126,20 @@ export class AntigravityExecutor extends BaseExecutor {
  // sessionId comes from transformRequest output; base.execute runs transformRequest before
  // buildHeaders, so we read it from instance state cached there (fallback: explicit arg).
  buildHeaders(credentials, stream = true, sessionId = null) {
-    const sid = sessionId || this._lastSessionId;
    return {
      "Content-Type": "application/json",
      "Authorization": `Bearer ${credentials.accessToken}`,
      "User-Agent": this.config.headers?.["User-Agent"] || ANTIGRAVITY_HEADERS["User-Agent"],
-      [INTERNAL_REQUEST_HEADER.name]: INTERNAL_REQUEST_HEADER.value,
-      ...(sid && { "X-Machine-Session-Id": sid }),
-      "Accept": stream ? "text/event-stream" : "application/json"
    };
  }

  transformRequest(model, body, stream, credentials) {
    const projectId = credentials?.projectId || this.generateProjectId();

+    // OpenAI clients may include stream_options even for non-streaming calls.
+    // Google generateContent rejects that combination before processing the request.
+    if (stream !== true) delete body.stream_options;
+
    // ─── Image generation: completely different request structure ───
    if (isImageModel(model)) {
      const imageConfig = parseImageConfig(model);
@@ -141,25 +164,26 @@ export class AntigravityExecutor extends BaseExecutor {
      });

      this._lastSessionId = sessionId;
+      const request = {
+        contents,
+        generationConfig: {
+          temperature: 1.0,
+          topP: 0.95,
+          topK: 40,
+          maxOutputTokens: 8192,
+          imageConfig,
+        },
+        sessionId,
+        // No tools, no systemInstruction, no safetySettings for image gen
+      };

      return {
        project: projectId,
        model: cleanModel,
        userAgent: "antigravity",
        requestType: "image_gen",
-        requestId: `agent-${crypto.randomUUID()}`,
-        request: {
-          contents,
-          generationConfig: {
-            temperature: 1.0,
-            topP: 0.95,
-            topK: 40,
-            maxOutputTokens: 8192,
-            imageConfig,
-          },
-          sessionId,
-          // No tools, no systemInstruction, no safetySettings for image gen
-        },
+        requestId: buildIdeRequestId({ body, request, credentials, model: cleanModel, requestType: "image_gen" }),
+        request,
      };
    }

@@ -177,8 +201,19 @@ export class AntigravityExecutor extends BaseExecutor {
        if (p.thoughtSignature && !p.functionCall && !p.text) return false;
        return true;
      });
-      if (role !== c.role || parts?.length !== c.parts?.length) {
-        return { ...c, role, parts };
+      // Gemini 3+ rejects functionCall parts without thoughtSignature. Clients (Claude Code, IDE)
+      // don't persist thoughtSignature in their history, so backfill the default signature on any
+      // functionCall part that arrives without one.
+      const needsBackfill = parts?.some(p => p.functionCall && !p.thoughtSignature) ?? false;
+      if (role !== c.role || parts?.length !== c.parts?.length || needsBackfill) {
+        return {
+          ...c, role,
+          parts: needsBackfill
+            ? parts.map(p => (p.functionCall && !p.thoughtSignature)
+                ? { ...p, thoughtSignature: DEFAULT_THINKING_AG_SIGNATURE }
+                : p)
+            : parts,
+        };
      }
      return c;
    });
@@ -233,10 +268,10 @@ export class AntigravityExecutor extends BaseExecutor {
    return {
      ...body,
      project: projectId,
-      model: model,
+      model: body.model || model,
      userAgent: "antigravity",
      requestType: "agent",
-      requestId: `agent-${crypto.randomUUID()}`,
+      requestId: buildIdeRequestId({ body, request: transformedRequest, credentials, model, requestType: "agent" }),
      request: transformedRequest
    };
  }
--- a/open-sse/executors/base.js
+++ b/open-sse/executors/base.js
@@ -2,6 +2,7 @@ import { HTTP_STATUS, RETRY_CONFIG, DEFAULT_RETRY_CONFIG, resolveRetryEntry, FET
 import { shouldRefreshCredentials } from "../services/oauthCredentialManager.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
 import { dbg } from "../utils/debugLog.js";
+import { resolveProviderTimeoutMs } from "../services/providerTimeout.js";
 import { ANTHROPIC_API_VERSION, OPENAI_COMPAT_BASE, ANTHROPIC_COMPAT_BASE } from "../providers/shared.js";

 /**
@@ -126,19 +127,20 @@ export class BaseExecutor {
    for (let urlIndex = 0; urlIndex < fallbackCount; urlIndex++) {
      const url = this.buildUrl(model, stream, urlIndex, credentials);
      const transformedBody = this.transformRequest(model, body, stream, credentials);
-      const headers = this.buildHeaders(credentials, stream);
+      const headers = this.buildHeaders(credentials, stream, url);

      if (!retryAttemptsByUrl[urlIndex]) retryAttemptsByUrl[urlIndex] = 0;

      // Abort if upstream doesn't return response headers within connection timeout
      const connectCtrl = new AbortController();
-      const timeoutMs = this.config?.timeoutMs || FETCH_CONNECT_TIMEOUT_MS;
+      const timeoutMs = await resolveProviderTimeoutMs(this.provider, this.config?.timeoutMs, FETCH_CONNECT_TIMEOUT_MS);
      const connectTimer = setTimeout(() => connectCtrl.abort(new Error("fetch connect timeout")), timeoutMs);
      const mergedSignal = signal ? AbortSignal.any([signal, connectCtrl.signal]) : connectCtrl.signal;
+      let fetchT0 = 0;

      try {
        const bodyStr = JSON.stringify(transformedBody);
-        const fetchT0 = Date.now();
+        fetchT0 = Date.now();
        dbg("FETCH", `${this.provider.toUpperCase()} → ${url} | body=${bodyStr.length}B | connectTimeout=${timeoutMs}ms`);
        const response = await proxyAwareFetch(url, {
          method: "POST",
@@ -164,6 +166,11 @@ export class BaseExecutor {
        clearTimeout(connectTimer);
        lastError = error;
        const isConnectTimeout = connectCtrl.signal.aborted && error.name === "AbortError";
+        // Error diagnostic — only logs on actual upstream failure. Distinguishes
+        // undici connect timeout (UND_ERR_CONNECT_TIMEOUT), DNS (ENOTFOUND),
+        // refused (ECONNREFUSED) vs our own connectCtrl abort (AbortError).
+        const cause = error?.cause || {};
+        console.log(`[FETCH-DIAG] ${this.provider} fetch error | name=${error.name} | code=${error.code ?? cause?.code ?? "none"} | msg=${String(error.message).slice(0, 120)} | connectTimeout=${timeoutMs}ms | elapsed=${Date.now() - fetchT0}ms`);
        dbg("FETCH", `${this.provider.toUpperCase()} ✖ ${error.name}: ${error.message}${isConnectTimeout ? " (connect timeout)" : ""}`);
        // Connect timeout is internal — convert to retryable network error, don't propagate AbortError
        if (error.name === "AbortError" && !isConnectTimeout) throw error;
--- a/open-sse/executors/codebuddy-intl.js
+++ b/open-sse/executors/codebuddy-intl.js
@@ -0,0 +1,30 @@
+import { DefaultExecutor } from "./default.js";
+
+/**
+ * CodeBuddyIntlExecutor — talks to https://www.codebuddy.ai/v2/chat/completions
+ *
+ * Same OpenAI-compatible-but-stream-only gateway behavior as codebuddy-cn:
+ * non-stream requests are rejected, and reasoning is surfaced only when the
+ * request carries the IDE's OpenAI-style reasoning params. Force stream and
+ * mirror reasoning_summary exactly like CodeBuddyExecutor.
+ */
+export class CodeBuddyIntlExecutor extends DefaultExecutor {
+  constructor() {
+    super("codebuddy-intl");
+  }
+
+  transformRequest(model, body, stream, credentials) {
+    const transformed = super.transformRequest(model, body, stream, credentials);
+    transformed.stream = true;
+
+    const eff = transformed.reasoning_effort;
+    if (eff === "none" || eff === "off") {
+      delete transformed.reasoning_effort;
+    } else if (eff) {
+      transformed.reasoning_summary = "auto";
+    }
+    return transformed;
+  }
+}
+
+export default CodeBuddyIntlExecutor;
--- a/open-sse/executors/codex.js
+++ b/open-sse/executors/codex.js
@@ -8,13 +8,21 @@ import {
 import { normalizeResponsesInput } from "../translator/formats/responsesApi.js";
 import { fetchImageAsBase64 } from "../translator/concerns/image.js";
 import { getModelUpstreamId } from "../config/providerModels.js";
-import { DEFAULT_RETRY_CONFIG, resolveRetryEntry } from "../config/runtimeConfig.js";
+import { DEFAULT_RETRY_CONFIG, HTTP_STATUS, resolveRetryEntry } from "../config/runtimeConfig.js";
 import { dbg } from "../utils/debugLog.js";
 import { resolveSessionId } from "../utils/sessionManager.js";

-// SSE error patterns inside 200-OK body that should trigger retry as if 503
-const CODEX_SSE_OVERLOADED_PATTERNS = ["server_is_overloaded", "service_unavailable_error"];
-const CODEX_SSE_PEEK_BYTES = 4096;
+// SSE error patterns inside 200-OK bodies. Some retry same account first; capacity rotates accounts.
+const CODEX_SSE_RETRY_PATTERNS = ["server_is_overloaded", "service_unavailable_error"];
+const CODEX_SSE_ACCOUNT_FALLBACK_PATTERNS = ["selected model is at capacity", "model_at_capacity"];
+const CODEX_SSE_USER_OUTPUT_PATTERNS = [
+  "event: response.output_text.delta",
+  "event: response.function_call_arguments.delta",
+  '"type":"response.output_text.delta"',
+  '"type":"response.function_call_arguments.delta"',
+];
+const CODEX_SSE_PEEK_BYTES = 256 * 1024;
+const CODEX_MODEL_CAPACITY_MESSAGE = "Selected model is at capacity. Please try a different model.";

 // Server-generated item id prefixes that Codex /responses cannot resolve when store=false
 const SERVER_ID_PATTERN = /^(rs|fc|resp|msg)_/;
@@ -116,6 +124,62 @@ function resolveCacheSessionId(body, credentials) {
  });
 }

+function normalizeReasoningEffort(value) {
+  return value === "max" ? "xhigh" : value;
+}
+
+function findNestedMessage(value, depth = 0) {
+  if (!value || depth > 6 || typeof value === "string") return null;
+  if (Array.isArray(value)) {
+    for (const item of value) {
+      const found = findNestedMessage(item, depth + 1);
+      if (found) return found;
+    }
+    return null;
+  }
+  if (typeof value !== "object") return null;
+  if (typeof value.message === "string" && value.message.trim()) return value.message;
+  if (typeof value.error?.message === "string" && value.error.message.trim()) return value.error.message;
+  if (typeof value.response?.error?.message === "string" && value.response.error.message.trim()) return value.response.error.message;
+  for (const child of Object.values(value)) {
+    const found = findNestedMessage(child, depth + 1);
+    if (found) return found;
+  }
+  return null;
+}
+
+function extractSseErrorMessage(text, fallback) {
+  const exact = text?.match(/Selected model is at capacity\. Please try a different model\./i)?.[0];
+  if (exact) return exact;
+
+  for (const line of String(text || "").split(/\r?\n/)) {
+    if (!line.startsWith("data:")) continue;
+    const data = line.slice(5).trim();
+    if (!data || data === "[DONE]") continue;
+    try {
+      const message = findNestedMessage(JSON.parse(data));
+      if (message) return message;
+    } catch {
+      // Ignore non-JSON SSE data lines.
+    }
+  }
+
+  return fallback || CODEX_MODEL_CAPACITY_MESSAGE;
+}
+
+function codexSseErrorResponse(status, message) {
+  return new Response(JSON.stringify({
+    error: {
+      message,
+      type: status >= 500 ? "server_error" : "invalid_request_error",
+      code: status === HTTP_STATUS.SERVICE_UNAVAILABLE ? "service_unavailable" : "upstream_error",
+    }
+  }), {
+    status,
+    headers: { "Content-Type": "application/json" },
+  });
+}
+
 /**
 * Codex Executor - handles OpenAI Codex API (Responses API format)
 * Automatically injects default instructions if missing
@@ -135,10 +199,17 @@ export class CodexExecutor extends BaseExecutor {
    headers["session_id"] = this._currentSessionId || credentials?.connectionId || "default";
    // Identify client type to Codex backend (matches official codex CLI)
    if (!headers["originator"]) headers["originator"] = "codex_cli_rs";
-    // Workspace binding header — improves account scope + cache affinity
-    const workspaceId = credentials?.providerSpecificData?.workspaceId;
-    if (typeof workspaceId === "string" && workspaceId && !headers["chatgpt-account-id"]) {
-      headers["chatgpt-account-id"] = workspaceId;
+    // Account/workspace binding header — required when multiple Codex accounts
+    // are configured. OAuth import stores ChatGPT account ID as chatgptAccountId;
+    // older/custom rows may use workspaceId/accountId. Prefer explicit workspaceId
+    // but fall back to chatgptAccountId so requests don't cross-bind to the wrong
+    // OpenAI account and surface as token_invalid after adding another account.
+    const accountId =
+      credentials?.providerSpecificData?.workspaceId ||
+      credentials?.providerSpecificData?.chatgptAccountId ||
+      credentials?.providerSpecificData?.accountId;
+    if (typeof accountId === "string" && accountId && !headers["ChatGPT-Account-ID"]) {
+      headers["ChatGPT-Account-ID"] = accountId;
    }
    return headers;
  }
@@ -198,7 +269,7 @@ export class CodexExecutor extends BaseExecutor {
    let attempt = 0;
    while (true) {
      const result = await super.execute(args);
-      const peek = await this._peekSseOverloaded(result.response);
+      const peek = await this._peekSseTransientError(result.response);
      if (!peek.matched) {
        // Replace body with re-assembled stream (prefix bytes already read + rest)
        if (peek.replacementBody) {
@@ -210,48 +281,57 @@ export class CodexExecutor extends BaseExecutor {
        }
        return result;
      }
+      if (peek.accountFallback) {
+        args.log?.warn?.("RETRY", `CODEX | SSE account fallback "${peek.message}"`);
+        result.response = codexSseErrorResponse(HTTP_STATUS.SERVICE_UNAVAILABLE, peek.message || CODEX_MODEL_CAPACITY_MESSAGE);
+        return result;
+      }
      if (attempt >= attempts) {
        args.log?.warn?.("RETRY", `CODEX | SSE overloaded "${peek.matched}" — retries exhausted (${attempt}/${attempts})`);
-        // Out of retries → return with replacement body so client gets the error
-        if (peek.replacementBody) {
-          result.response = new Response(peek.replacementBody, {
-            status: result.response.status,
-            statusText: result.response.statusText,
-            headers: result.response.headers,
-          });
-        }
+        result.response = codexSseErrorResponse(HTTP_STATUS.SERVICE_UNAVAILABLE, peek.message || peek.matched);
        return result;
      }
      attempt++;
      args.log?.debug?.("RETRY", `CODEX | SSE "${peek.matched}" retry ${attempt}/${attempts} after ${delayMs / 1000}s`);
      dbg("CODEX", `SSE overloaded "${peek.matched}" → retry ${attempt}/${attempts} in ${delayMs}ms`);
-      try { await result.response.body?.cancel?.(); } catch { /* noop */ }
      await new Promise(r => setTimeout(r, delayMs));
    }
  }

-  // Peek first N bytes of SSE body to detect upstream "overloaded" errors.
-  // Returns { matched: string|null, replacementBody: ReadableStream|null }.
-  // Caller MUST use replacementBody (original body has been read).
-  async _peekSseOverloaded(response) {
-    if (!response || !response.ok || !response.body) return { matched: null, replacementBody: null };
+  // Peek first N bytes of SSE body to detect upstream transient errors.
+  // Returns { matched: string|null, message: string|null, accountFallback: boolean, replacementBody: ReadableStream|null }.
+  // Caller must use replacementBody when no error matched (original body has been read).
+  async _peekSseTransientError(response) {
+    if (!response || !response.ok || !response.body) return { matched: null, message: null, accountFallback: false, replacementBody: null };
    const reader = response.body.getReader();
    const decoder = new TextDecoder();
    const chunks = [];
    let text = "";
    let matched = null;
+    let accountFallback = false;
    try {
      while (text.length < CODEX_SSE_PEEK_BYTES) {
        const { done, value } = await reader.read();
        if (done) break;
        chunks.push(value);
        text += decoder.decode(value, { stream: true });
-        const hit = CODEX_SSE_OVERLOADED_PATTERNS.find(p => text.includes(p));
-        if (hit) { matched = hit; break; }
+        const lowerText = text.toLowerCase();
+        const accountHit = CODEX_SSE_ACCOUNT_FALLBACK_PATTERNS.find(p => lowerText.includes(p));
+        if (accountHit) { matched = accountHit; accountFallback = true; break; }
+        const retryHit = CODEX_SSE_RETRY_PATTERNS.find(p => lowerText.includes(p));
+        if (retryHit) { matched = retryHit; break; }
+        if (CODEX_SSE_USER_OUTPUT_PATTERNS.some(p => lowerText.includes(p))) break;
      }
    } catch (e) {
      dbg("CODEX", `peek read error: ${e.message}`);
    }
+
+    if (matched) {
+      try { await reader.cancel(); } catch { /* noop */ }
+      try { reader.releaseLock(); } catch { /* noop */ }
+      return { matched, message: extractSseErrorMessage(text, matched), accountFallback, replacementBody: null };
+    }
+
    reader.releaseLock();

    // Re-assemble stream: prefix chunks + remaining upstream body
@@ -273,7 +353,7 @@ export class CodexExecutor extends BaseExecutor {
        try { upstreamReader?.cancel(reason); } catch { /* noop */ }
      },
    });
-    return { matched, replacementBody };
+    return { matched: null, message: null, accountFallback: false, replacementBody };
  }

  // Parse Codex usage_limit_reached to extract precise resetsAtMs; fallback to default otherwise
@@ -347,7 +427,7 @@ export class CodexExecutor extends BaseExecutor {

    // Extract thinking level from model name suffix
    // e.g., gpt-5.3-codex-high → high, gpt-5.3-codex → medium (default)
-    const effortLevels = ['none', 'low', 'medium', 'high', 'xhigh'];
+    const effortLevels = ['none', 'minimal', 'low', 'medium', 'high', 'xhigh'];
    let modelEffort = null;
    for (const level of effortLevels) {
      if (body.model.endsWith(`-${level}`)) {
@@ -360,10 +440,11 @@ export class CodexExecutor extends BaseExecutor {

    // Priority: explicit reasoning.effort > reasoning_effort param > model suffix > default (medium)
    if (!body.reasoning) {
-      const effort = body.reasoning_effort || modelEffort || 'low';
+      const effort = normalizeReasoningEffort(body.reasoning_effort || modelEffort || 'low');
      body.reasoning = { effort, summary: "auto" };
-    } else if (!body.reasoning.summary) {
-      body.reasoning.summary = "auto";
+    } else {
+      body.reasoning.effort = normalizeReasoningEffort(body.reasoning.effort);
+      if (!body.reasoning.summary) body.reasoning.summary = "auto";
    }
    delete body.reasoning_effort;

@@ -391,6 +472,9 @@ export class CodexExecutor extends BaseExecutor {
    delete body.safety_identifier; // Droid CLI sends this but Codex doesn't support it
    delete body.previous_response_id; // store=false → backend can't resolve previous resp; avoid 404

+    if (body.service_tier === "fast") body.service_tier = "priority";
+    if (body.service_tier && body.service_tier !== "priority") delete body.service_tier;
+
    // Final allowlist filter — strip any unknown field that could trigger upstream "routing_unsupported"
    for (const k of Object.keys(body)) {
      if (!RESPONSES_API_ALLOWLIST.has(k)) delete body[k];
--- a/open-sse/executors/commandcode.js
+++ b/open-sse/executors/commandcode.js
@@ -1,6 +1,7 @@
 import { randomUUID } from "crypto";
 import { BaseExecutor } from "./base.js";
 import { PROVIDERS } from "../config/providers.js";
+import { HTTP_STATUS } from "../config/runtimeConfig.js";
 import { commandCodeToOpenAIResponse } from "../translator/response/commandcode-to-openai.js";
 import { SSE_DONE } from "../utils/sseConstants.js";

@@ -14,81 +15,271 @@ import { SSE_DONE } from "../utils/sseConstants.js";
 * We translate each event to an OpenAI chat.completion.chunk and emit it as SSE so
 * both the streaming and non-streaming (forced SSE → JSON) downstream handlers in
 * 9router can consume it without further format translation.
+ *
+ * Terminal upstream failures arrive as `{"type":"error"}` events inside the HTTP
+ * 200 stream, so a plain `response.ok` check cannot see them. We peek the first
+ * events before committing the response (see peekForUpstreamError) so a stream
+ * that starts with an error fails fast — the normal `!response.ok` path then
+ * triggers account/model fallback instead of streaming fake success content.
 */
 export class CommandCodeExecutor extends BaseExecutor {
-  constructor() {
-    super("commandcode", PROVIDERS.commandcode);
-  }
+	constructor() {
+		super("commandcode", PROVIDERS.commandcode);
+	}

-  transformRequest(model, body, stream, credentials) {
-    body.stream = true;
-    return body;
-  }
+	transformRequest(_model, body, _stream, _credentials) {
+		body.stream = true;
+		return body;
+	}

-  buildHeaders(credentials, stream = true) {
-    const headers = {
-      "Content-Type": "application/json",
-      ...(this.config.headers || {}),
-      "x-session-id": randomUUID(),
-    };
+	buildHeaders(credentials, stream = true) {
+		const headers = {
+			"Content-Type": "application/json",
+			...(this.config.headers || {}),
+			"x-session-id": randomUUID(),
+		};

-    const token = credentials?.apiKey || credentials?.accessToken;
-    if (token) headers["Authorization"] = `Bearer ${token}`;
+		const token = credentials?.apiKey || credentials?.accessToken;
+		if (token) headers["Authorization"] = `Bearer ${token}`;

-    if (stream) headers["Accept"] = "text/event-stream";
-    return headers;
-  }
+		if (stream) headers["Accept"] = "text/event-stream";
+		return headers;
+	}

-  async execute(opts) {
-    const result = await super.execute(opts);
-    if (!result?.response?.ok || !result.response.body) return result;
-    result.response = wrapNdjsonAsOpenAISse(result.response, opts.model);
-    return result;
-  }
+	async execute(opts) {
+		const result = await super.execute(opts);
+		if (!result?.response?.ok || !result.response.body) return result;
+		result.response = await peekForUpstreamError(result.response, opts.model, {
+			signal: opts.signal,
+		});
+		return result;
+	}
+}
+
+// How long to hold the response open while peeking the first upstream events.
+// An upstream error event ("Network connection lost") is emitted at stream
+// start, so the peek is fast; the bound just prevents a slow-started stream
+// from being held hostage. Env: COMMANDCODE_PEEK_TIMEOUT_MS.
+const PEEK_TIMEOUT_MS = (() => {
+	const raw = process.env.COMMANDCODE_PEEK_TIMEOUT_MS;
+	const n = raw ? parseInt(raw, 10) : NaN;
+	return Number.isFinite(n) && n > 0 ? n : 10 * 1000;
+})();
+
+// Event types that count as "the stream has started producing". Everything
+// else (start, start-step, reasoning-start, text-start, ...) is metadata and
+// does not end the peek.
+const MEANINGFUL_EVENT_TYPES = new Set([
+	"text-delta",
+	"reasoning-delta",
+	"tool-input-start",
+	"tool-input-delta",
+	"tool-input-end",
+	"tool-call",
+	"finish-step",
+	"finish",
+]);
+
+function makeAbortError(reason) {
+	const error = new Error(reason?.message || reason || "Request aborted");
+	error.name = "AbortError";
+	return error;
+}
+
+function tryParseEvent(line) {
+	const trimmed = line.trim();
+	if (!trimmed) return null;
+	const json = trimmed.startsWith("data:") ? trimmed.slice(5).trim() : trimmed;
+	if (!json || json === "[DONE]") return null;
+	try {
+		return JSON.parse(json);
+	} catch {
+		return null;
+	}
+}
+
+function formatErrorValue(errVal) {
+	const errStr =
+		typeof errVal === "string"
+			? errVal
+			: typeof errVal?.message === "string"
+				? errVal.message
+				: JSON.stringify(errVal);
+	const errType =
+		typeof errVal === "string"
+			? "upstream_error"
+			: errVal?.type || "upstream_error";
+	return { message: errStr, type: errType };
+}
+
+/**
+ * Read the first upstream events before committing the response.
+ *
+ * - `{"type":"error"}` as the first meaningful event → return a 502 Response so
+ *   chatCore's `!response.ok` path parses the error and triggers fallback.
+ * - Otherwise → re-emit the buffered bytes + the rest of the stream through the
+ *   normal NDJSON → OpenAI SSE wrapper and return it untouched in spirit.
+ *
+ * Bounded by `timeoutMs` (default PEEK_TIMEOUT_MS): if no meaningful event
+ * arrives in time, or the request signal aborts, we commit whatever we have and
+ * let the regular stream pipeline (stall detection, abort handling) take over.
+ */
+export async function peekForUpstreamError(
+	originalResponse,
+	model,
+	{ signal = null, timeoutMs = PEEK_TIMEOUT_MS } = {},
+) {
+	const reader = originalResponse.body.getReader();
+	const decoder = new TextDecoder();
+	const abortController = new AbortController();
+	const forwardAbort = () => abortController.abort(signal?.reason);
+	if (signal?.aborted) abortController.abort(signal?.reason);
+	else if (signal)
+		signal.addEventListener("abort", forwardAbort, { once: true });
+
+	// Raw bytes for lossless re-emission; decoded text is only used for line
+	// parsing / error detection. Never re-encode decoded text: TextDecoder
+	// holds a split multi-byte char internally and flush() would replace it
+	// with U+FFFD, corrupting the stream.
+	const rawChunks = [];
+	let peeked = "";
+	let errorEvent = null;
+	let committed = false;
+
+	const readWithTimeout = (ms) => {
+		if (abortController.signal.aborted) {
+			return Promise.reject(makeAbortError(abortController.signal.reason));
+		}
+		const timeoutPromise = new Promise((_, reject) => {
+			const t = setTimeout(() => reject(new Error("peek timeout")), ms);
+			t.unref?.();
+		});
+		const abortPromise = new Promise((_, reject) => {
+			abortController.signal.addEventListener(
+				"abort",
+				() => reject(makeAbortError(abortController.signal.reason)),
+				{ once: true },
+			);
+		});
+		return Promise.race([reader.read(), timeoutPromise, abortPromise]);
+	};
+
+	try {
+		const deadline = Date.now() + timeoutMs;
+		while (!errorEvent && !committed && Date.now() < deadline) {
+			const { done, value } = await readWithTimeout(
+				Math.max(deadline - Date.now(), 1),
+			);
+			if (done) break;
+			rawChunks.push(value);
+			peeked += decoder.decode(value, { stream: true });
+			const lines = peeked.split("\n");
+			// The last segment may be a partial line — only parse complete ones.
+			for (const line of lines.slice(0, -1)) {
+				const event = tryParseEvent(line);
+				if (!event?.type) continue;
+				if (event.type === "error") {
+					errorEvent = event;
+					break;
+				}
+				if (MEANINGFUL_EVENT_TYPES.has(event.type)) {
+					committed = true;
+					break;
+				}
+			}
+		}
+	} catch {
+		// timeout / abort / read failure during the peek → commit whatever we have;
+		// the downstream stream pipeline (stall detection, abort handling) takes over.
+	}
+
+	if (signal) signal.removeEventListener("abort", forwardAbort);
+
+	if (errorEvent) {
+		await reader.cancel("commandcode early error detected").catch(() => {});
+		const { message, type } = formatErrorValue(
+			errorEvent.error ?? errorEvent.message ?? "unknown",
+		);
+		return new Response(JSON.stringify({ error: { message, type } }), {
+			status: HTTP_STATUS.BAD_GATEWAY,
+			statusText: message.slice(0, 200),
+			headers: { "Content-Type": "application/json" },
+		});
+	}
+
+	const remaining = new ReadableStream({
+		start(controller) {
+			(async () => {
+				try {
+					// Re-emit RAW bytes (never re-encoded decoded text) so split
+					// multi-byte UTF-8 sequences survive the peek untouched.
+					for (const c of rawChunks) controller.enqueue(c);
+					while (true) {
+						const { done, value } = await reader.read();
+						if (done) break;
+						controller.enqueue(value);
+					}
+					controller.close();
+				} catch (err) {
+					controller.error(err);
+				}
+			})();
+		},
+		cancel() {
+			reader.cancel("commandcode stream cancelled").catch(() => {});
+		},
+	});
+
+	const combined = new Response(remaining, {
+		status: originalResponse.status,
+		statusText: originalResponse.statusText,
+		headers: originalResponse.headers,
+	});
+	return wrapNdjsonAsOpenAISse(combined, model);
 }

 function wrapNdjsonAsOpenAISse(originalResponse, model) {
-  const decoder = new TextDecoder();
-  const encoder = new TextEncoder();
-  let buffer = "";
-  const state = { model };
+	const decoder = new TextDecoder();
+	const encoder = new TextEncoder();
+	let buffer = "";
+	const state = { model };

-  const emitChunks = (chunks, controller) => {
-    if (!chunks) return;
-    const list = Array.isArray(chunks) ? chunks : [chunks];
-    for (const c of list) {
-      if (c == null) continue;
-      controller.enqueue(encoder.encode(`data: ${JSON.stringify(c)}\n\n`));
-    }
-  };
+	const emitChunks = (chunks, controller) => {
+		if (!chunks) return;
+		const list = Array.isArray(chunks) ? chunks : [chunks];
+		for (const c of list) {
+			if (c == null) continue;
+			controller.enqueue(encoder.encode(`data: ${JSON.stringify(c)}\n\n`));
+		}
+	};

-  const transform = new TransformStream({
-    transform(chunk, controller) {
-      buffer += decoder.decode(chunk, { stream: true });
-      const lines = buffer.split("\n");
-      buffer = lines.pop() || "";
-      for (const line of lines) {
-        const trimmed = line.trim();
-        if (!trimmed) continue;
-        // Translate AI SDK v5 NDJSON line to one or more OpenAI chunks
-        emitChunks(commandCodeToOpenAIResponse(trimmed, state), controller);
-      }
-    },
-    flush(controller) {
-      const trimmed = buffer.trim();
-      if (trimmed) {
-        emitChunks(commandCodeToOpenAIResponse(trimmed, state), controller);
-      }
-      controller.enqueue(encoder.encode(SSE_DONE));
-    },
-  });
+	const transform = new TransformStream({
+		transform(chunk, controller) {
+			buffer += decoder.decode(chunk, { stream: true });
+			const lines = buffer.split("\n");
+			buffer = lines.pop() || "";
+			for (const line of lines) {
+				const trimmed = line.trim();
+				if (!trimmed) continue;
+				// Translate AI SDK v5 NDJSON line to one or more OpenAI chunks
+				emitChunks(commandCodeToOpenAIResponse(trimmed, state), controller);
+			}
+		},
+		flush(controller) {
+			const trimmed = buffer.trim();
+			if (trimmed) {
+				emitChunks(commandCodeToOpenAIResponse(trimmed, state), controller);
+			}
+			controller.enqueue(encoder.encode(SSE_DONE));
+		},
+	});

-  const newBody = originalResponse.body.pipeThrough(transform);
-  return new Response(newBody, {
-    status: originalResponse.status,
-    statusText: originalResponse.statusText,
-    headers: originalResponse.headers,
-  });
+	const newBody = originalResponse.body.pipeThrough(transform);
+	return new Response(newBody, {
+		status: originalResponse.status,
+		statusText: originalResponse.statusText,
+		headers: originalResponse.headers,
+	});
 }

 export default CommandCodeExecutor;
--- a/open-sse/executors/cursor.js
+++ b/open-sse/executors/cursor.js
@@ -1,18 +1,22 @@
 import { BaseExecutor } from "./base.js";
-import { PROVIDERS } from "../config/providers.js";
+import { PROVIDERS, PROVIDER_OAUTH } from "../config/providers.js";
 import { HTTP_STATUS } from "../config/runtimeConfig.js";
 import {
  generateCursorBody,
+  encodeField,
+  wrapConnectRPCFrame,
+  decodeMessage,
  parseConnectRPCFrame,
  extractTextFromResponse
 } from "../utils/cursorProtobuf.js";
 import { buildCursorHeaders } from "../utils/cursorChecksum.js";
 import { estimateUsage } from "../utils/usageTracking.js";
 import { SSE_DONE, SSE_HEADERS } from "../utils/sseConstants.js";
-import { chatChunkSse } from "../utils/sse.js";
+import { chatChunkSse, sseChunk } from "../utils/sse.js";
 import { FORMATS } from "../translator/formats.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
 import zlib from "zlib";
+import crypto from "crypto";

 // Detect cloud environment
 const isCloudEnv = () => {
@@ -38,6 +42,130 @@ const COMPRESS_FLAG = {
  GZIP_TRAILER: 0x03
 };

+const AGENT_RUN_PATH = "/agent.v1.AgentService/Run";
+const PROTOBUF_LEN = 2;
+const PROTOBUF_VARINT = 0;
+
+function concatBuffers(...parts) {
+  const length = parts.reduce((total, part) => total + part.length, 0);
+  const result = new Uint8Array(length);
+  let offset = 0;
+  for (const part of parts) {
+    result.set(part, offset);
+    offset += part.length;
+  }
+  return result;
+}
+
+const agentString = (field, value) => encodeField(field, PROTOBUF_LEN, value);
+const agentMessage = (field, value) => encodeField(field, PROTOBUF_LEN, value);
+const agentBool = (field, value) => encodeField(field, PROTOBUF_VARINT, value ? 1 : 0);
+
+function textFromContent(content) {
+  if (typeof content === "string") return content;
+  if (!Array.isArray(content)) return "";
+  return content
+    .filter((part) => part?.type === "text" && typeof part.text === "string")
+    .map((part) => part.text)
+    .join("\n");
+}
+
+function isAgentTextRequest(body) {
+  // Many compatible clients always attach their built-in tool schemas, even
+  // for a normal text turn. Cursor's retired ChatService rejects those
+  // requests; AgentService can still answer the text turn, so ignore schemas
+  // here. A real tool-call/result conversation is kept on the legacy path
+  // until its AgentService tool protocol is implemented.
+  return Array.isArray(body?.messages) && body.messages.every((message) => {
+    if (message?.tool_calls?.length || message?.role === "tool") return false;
+    return typeof message?.content === "string"
+      || Array.isArray(message?.content) && message.content.every((part) => part?.type === "text");
+  });
+}
+
+function encodeHistoryMessage(message) {
+  const content = textFromContent(message?.content);
+  if (!content) return null;
+
+  // ConversationHistoryMessage.user / .assistant -> repeated content -> text.
+  const text = agentString(1, content);
+  if (message.role === "assistant") {
+    return agentMessage(2, agentMessage(1, agentMessage(1, text)));
+  }
+  return agentMessage(1, agentMessage(1, agentMessage(1, text)));
+}
+
+function buildAgentRunFrame(messages, model) {
+  const system = messages
+    .filter((message) => message?.role === "system")
+    .map((message) => textFromContent(message.content))
+    .filter(Boolean)
+    .join("\n\n");
+  const chatMessages = messages.filter((message) => message?.role !== "system");
+  const currentIndex = [...chatMessages].map((message) => message?.role).lastIndexOf("user");
+  const current = currentIndex >= 0 ? chatMessages[currentIndex] : chatMessages.at(-1);
+  const history = chatMessages
+    .slice(0, currentIndex >= 0 ? currentIndex : -1)
+    .map(encodeHistoryMessage)
+    .filter(Boolean);
+  const userText = textFromContent(current?.content) || "Continue.";
+
+  // agent.v1.UserMessageAction.user_message and its optional history.
+  const userMessage = concatBuffers(
+    agentString(1, userText),
+    agentString(2, crypto.randomUUID()),
+  );
+  const conversationHistory = history.length
+    ? concatBuffers(...history.map((entry) => agentMessage(1, entry)))
+    : null;
+  const userAction = concatBuffers(
+    agentMessage(1, userMessage),
+    ...(conversationHistory ? [agentMessage(7, conversationHistory)] : []),
+  );
+  const conversationAction = agentMessage(1, userAction);
+  const requestedModel = concatBuffers(agentString(1, model), agentBool(7, true));
+  const runRequest = concatBuffers(
+    // An empty ConversationStateStructure starts a fresh local agent session.
+    agentMessage(1, new Uint8Array()),
+    agentMessage(2, conversationAction),
+    ...(system ? [agentString(8, system)] : []),
+    agentMessage(9, requestedModel),
+  );
+
+  // agent.v1.AgentClientMessage.run_request.
+  return wrapConnectRPCFrame(agentMessage(1, runRequest));
+}
+
+function extractAgentString(message, field) {
+  const value = message?.get(field)?.[0]?.value;
+  return value ? Buffer.from(value).toString("utf8") : "";
+}
+
+function decodeAgentFrames(buffer, onFrame) {
+  let pending = Buffer.from(buffer || []);
+  while (pending.length >= 5) {
+    const flags = pending[0];
+    const length = pending.readUInt32BE(1);
+    if (pending.length < 5 + length) break;
+    let payload = pending.subarray(5, 5 + length);
+    pending = pending.subarray(5 + length);
+    if (flags & COMPRESS_FLAG.GZIP) {
+      payload = zlib.gunzipSync(payload);
+    }
+    if (!(flags & COMPRESS_FLAG.TRAILER)) onFrame(payload);
+  }
+  return pending;
+}
+
+function createRequestContextResponse() {
+  // AgentService asks every run for client context. 9router has no IDE file
+  // context, so acknowledge with an empty RequestContext.
+  const requestContextSuccess = agentMessage(1, new Uint8Array());
+  const requestContextResult = agentMessage(1, requestContextSuccess);
+  const execClientMessage = agentMessage(10, requestContextResult);
+  return wrapConnectRPCFrame(agentMessage(2, execClientMessage));
+}
+
 const CURSOR_STREAM_DEBUG = process.env.CURSOR_STREAM_DEBUG === "1";
 const debugLog = (...args) => {
  if (CURSOR_STREAM_DEBUG) console.log(...args);
@@ -253,7 +381,304 @@ export class CursorExecutor extends BaseExecutor {
    });
  }

+  /**
+   * AgentService (agent.api5.cursor.sh) is HTTP/2-only. Node's fetch/undici speaks
+   * HTTP/1.1 and fails with HTTPParserError on the h2 preface — use http2 duplex.
+   */
+  openAgentHttp2Stream(url, headers, signal) {
+    if (!http2) {
+      throw new Error("HTTP/2 is required for Cursor AgentService (endpoint is h2-only)");
+    }
+
+    const urlObj = new URL(url);
+    const client = http2.connect(`https://${urlObj.host}`);
+    const chunkQueue = [];
+    let waiting = null;
+    let ended = false;
+    let streamError = null;
+    let req = null;
+
+    const wake = (result) => {
+      if (!waiting) return;
+      const resolve = waiting;
+      waiting = null;
+      resolve(result);
+    };
+
+    const fail = (error) => {
+      if (streamError) return;
+      streamError = error;
+      ended = true;
+      wake(null);
+    };
+
+    const close = () => {
+      try { req?.destroy(); } catch {}
+      try { client.close(); } catch {}
+    };
+
+    client.on("error", fail);
+
+    req = client.request({
+      ":method": "POST",
+      ":path": urlObj.pathname,
+      ":authority": urlObj.host,
+      ":scheme": "https",
+      ...headers,
+    });
+
+    req.on("error", fail);
+    req.on("data", (chunk) => {
+      if (waiting) wake({ value: chunk, done: false });
+      else chunkQueue.push(chunk);
+    });
+    req.on("end", () => {
+      ended = true;
+      wake({ value: undefined, done: true });
+    });
+
+    if (signal) {
+      const onAbort = () => {
+        fail(new Error("Request aborted"));
+        close();
+      };
+      if (signal.aborted) onAbort();
+      else signal.addEventListener("abort", onAbort, { once: true });
+    }
+
+    const responseHeaders = new Promise((resolve, reject) => {
+      const onEarlyError = (error) => reject(error);
+      client.once("error", onEarlyError);
+      req.once("error", onEarlyError);
+      req.once("response", (hdrs) => {
+        client.off("error", onEarlyError);
+        req.off("error", onEarlyError);
+        resolve(hdrs);
+      });
+    });
+
+    return {
+      responseHeaders,
+      write(frame) {
+        if (req && !req.destroyed) req.write(Buffer.from(frame));
+      },
+      end() {
+        try { if (req && !req.destroyed) req.end(); } catch {}
+      },
+      close,
+      async read() {
+        if (chunkQueue.length) return { value: chunkQueue.shift(), done: false };
+        if (ended) {
+          if (streamError) throw streamError;
+          return { value: undefined, done: true };
+        }
+        const result = await new Promise((resolve) => { waiting = resolve; });
+        if (streamError) throw streamError;
+        return result || { value: undefined, done: true };
+      },
+    };
+  }
+
+  async executeAgent({ model, body, stream, credentials, signal }) {
+    const agentEndpoint = PROVIDER_OAUTH.cursor?.agentEndpoint;
+    if (!agentEndpoint) throw new Error("Cursor AgentService endpoint is not configured");
+
+    const url = `${agentEndpoint}${AGENT_RUN_PATH}`;
+    const headers = this.buildHeaders(credentials);
+    const requestController = new AbortController();
+    if (signal?.addEventListener) {
+      signal.addEventListener("abort", () => requestController.abort(signal.reason), { once: true });
+    }
+
+    let session;
+    try {
+      session = this.openAgentHttp2Stream(url, headers, requestController.signal);
+      session.write(buildAgentRunFrame(body.messages || [], model));
+    } catch (error) {
+      throw new Error(`Cursor AgentService request failed: ${error.message}`);
+    }
+
+    let responseHeaders;
+    try {
+      responseHeaders = await session.responseHeaders;
+    } catch (error) {
+      session.close();
+      throw new Error(`Cursor AgentService request failed: ${error.message}`);
+    }
+
+    const status = Number(responseHeaders[":status"] || 0);
+    if (status !== 200) {
+      let errorText = "";
+      try {
+        while (true) {
+          const { done, value } = await session.read();
+          if (done) break;
+          errorText += Buffer.from(value).toString("utf8");
+        }
+      } catch {}
+      session.close();
+      return {
+        response: new Response(JSON.stringify({
+          error: { message: `Cursor AgentService ${status}: ${errorText || "request failed"}`, type: "api_error" },
+        }), { status: status || HTTP_STATUS.SERVER_ERROR, headers: { "Content-Type": "application/json" } }),
+        url,
+        headers,
+        transformedBody: body,
+        responseFormat: FORMATS.OPENAI,
+      };
+    }
+
+    // The Claude SSE translator derives Anthropic's message ID by stripping
+    // `chatcmpl-`. Keep the remaining ID in Anthropic's required `msg_` form
+    // so strict clients such as Claude Code accept the completed stream.
+    const responseId = `chatcmpl-msg_${Date.now()}`;
+    const created = Math.floor(Date.now() / 1000);
+    let pending = Buffer.alloc(0);
+    let finished = false;
+
+    const consume = async (onEvent) => {
+      try {
+        while (!finished) {
+          const { done, value } = await session.read();
+          if (done) break;
+          pending = Buffer.concat([pending, Buffer.from(value)]);
+          pending = decodeAgentFrames(pending, (payload) => {
+            // A single read can carry several frames; once the turn is over the
+            // rest of the batch must not reach the already-closed controller.
+            if (finished) return;
+            const serverMessage = decodeMessage(payload);
+
+            // agent.v1.AgentServerMessage.interaction_update
+            if (serverMessage.has(1)) {
+              const update = decodeMessage(serverMessage.get(1)[0].value);
+              if (update.has(1)) {
+                const textDelta = extractAgentString(decodeMessage(update.get(1)[0].value), 1);
+                if (textDelta) onEvent({ type: "text", value: textDelta });
+              }
+              // Cursor's AgentService emits internal reasoning without the
+              // cryptographic signature required by Anthropic thinking blocks.
+              // Forwarding it makes strict Anthropic clients (Claude Code)
+              // discard or wait on an otherwise complete response. Keep the
+              // reasoning upstream-only and emit the normal answer text.
+              if (update.has(14)) {
+                finished = true;
+                onEvent({ type: "done" });
+              }
+            }
+
+            // AgentService requests IDE context before producing a response.
+            // Return an empty context; 9router is not coupled to an editor.
+            if (serverMessage.has(2)) {
+              const execRequest = decodeMessage(serverMessage.get(2)[0].value);
+              if (execRequest.has(10)) {
+                session.write(createRequestContextResponse());
+              } else {
+                // Every other ExecServerMessage variant is an editor-backed tool
+                // (shell, read, write, …) that 9router cannot service. Fail the
+                // turn rather than narrating protocol state as assistant text.
+                debugLog(`[CURSOR AGENT] Unsupported exec request fields: ${[...execRequest.keys()].join(",")}`);
+                finished = true;
+                onEvent({ type: "error", value: "Cursor AgentService requested an unsupported IDE tool" });
+              }
+            }
+          });
+        }
+      } finally {
+        try { session.end(); } catch {}
+        try { session.close(); } catch {}
+        if (!finished) onEvent({ type: "done" });
+      }
+    };
+
+    if (stream === false) {
+      let content = "";
+      let reasoning = "";
+      let agentError = null;
+      await consume((event) => {
+        if (event.type === "text") content += event.value;
+        else if (event.type === "thinking") reasoning += event.value;
+        else if (event.type === "error") agentError = event.value;
+      });
+      if (agentError) {
+        return {
+          response: new Response(JSON.stringify({ error: { message: agentError, type: "api_error" } }), {
+            status: HTTP_STATUS.BAD_REQUEST,
+            headers: { "Content-Type": "application/json" },
+          }),
+          url,
+          headers,
+          transformedBody: body,
+          responseFormat: FORMATS.OPENAI,
+        };
+      }
+      return {
+        response: new Response(JSON.stringify({
+          id: responseId,
+          object: "chat.completion",
+          created,
+          model,
+          choices: [{ index: 0, message: { role: "assistant", content: content || null, ...(reasoning ? { reasoning_content: reasoning } : {}) }, finish_reason: "stop" }],
+          usage: estimateUsage(body, content.length, FORMATS.OPENAI),
+        }), { headers: { "Content-Type": "application/json" } }),
+        url,
+        headers,
+        transformedBody: body,
+        responseFormat: FORMATS.OPENAI,
+      };
+    }
+
+    const encoder = new TextEncoder();
+    const responseStream = new ReadableStream({
+      start(controller) {
+        consume((event) => {
+          if (event.type === "text") {
+            controller.enqueue(encoder.encode(chatChunkSse({ id: responseId, created, model, delta: { content: event.value } })));
+          } else if (event.type === "thinking") {
+            controller.enqueue(encoder.encode(chatChunkSse({ id: responseId, created, model, delta: { reasoning_content: event.value } })));
+          } else if (event.type === "error") {
+            // An SSE error frame, not a content delta: a protocol failure must not
+            // be rendered to the user as the assistant's reply, and downstream
+            // usage tracking must not record the turn as a success.
+            controller.enqueue(encoder.encode(sseChunk({ error: { message: event.value, type: "api_error" } })));
+            controller.enqueue(encoder.encode(SSE_DONE));
+            controller.close();
+          } else if (event.type === "done") {
+            controller.enqueue(encoder.encode(chatChunkSse({ id: responseId, created, model, delta: {}, finishReason: "stop" })));
+            controller.enqueue(encoder.encode(SSE_DONE));
+            controller.close();
+          }
+        }).catch((error) => controller.error(error));
+      },
+      cancel() {
+        requestController.abort();
+      },
+    });
+
+    return {
+      response: new Response(responseStream, { headers: SSE_HEADERS }),
+      url,
+      headers,
+      transformedBody: body,
+      responseFormat: FORMATS.OPENAI,
+    };
+  }
+
  async execute({ model, body, stream, credentials, signal, log, proxyOptions = null }) {
+    if (isAgentTextRequest(body)) {
+      try {
+        return await this.executeAgent({ model, body, stream, credentials, signal });
+      } catch (error) {
+        return {
+          response: new Response(JSON.stringify({
+            error: { message: error.message, type: "connection_error", code: "" },
+          }), { status: HTTP_STATUS.SERVER_ERROR, headers: { "Content-Type": "application/json" } }),
+          url: `${PROVIDER_OAUTH.cursor?.agentEndpoint || ""}${AGENT_RUN_PATH}`,
+          headers: {},
+          transformedBody: body,
+        };
+      }
+    }
+
    const url = this.buildUrl();
    const headers = this.buildHeaders(credentials);
    const transformedBody = this.transformRequest(model, body, stream, credentials);
--- a/open-sse/executors/default.js
+++ b/open-sse/executors/default.js
@@ -38,7 +38,8 @@ function applyAuth(headers, desc, credentials) {

 // Provider-specific header quirks kept as small hooks (not pure auth).
 const HEADER_HOOKS = {
-  kimiHeaders: (h) => Object.assign(h, buildKimiHeaders()),
+  // Stable device_id from OAuth connection (CLIProxyAPI KimiTokenStorage.DeviceID)
+  kimiHeaders: (h, c) => Object.assign(h, buildKimiHeaders(c?.providerSpecificData?.deviceId)),
  clineHeaders: (h, c) => Object.assign(h, buildClineHeaders(c.apiKey || c.accessToken)),
  kilocodeOrg: (h, c) => { if (c.providerSpecificData?.orgId) h["X-Kilocode-OrganizationID"] = c.providerSpecificData.orgId; },
  claudeOverlay: (h) => {
@@ -226,7 +227,9 @@ export class DefaultExecutor extends BaseExecutor {
      gemini: () => this.refreshFromGrant(credentials, proxyOptions),
      kiro: () => this.refreshKiro(credentials.refreshToken, proxyOptions),
      cline: () => this.refreshCline(credentials.refreshToken, proxyOptions),
-      "kimi-coding": () => this.refreshKimiCoding(credentials.refreshToken, proxyOptions),
+      clinepass: () => this.refreshCline(credentials.refreshToken, proxyOptions),
+      kimi: () => this.refreshKimi(credentials, proxyOptions),
+      "kimi-coding": () => this.refreshKimi(credentials, proxyOptions),
      kilocode: () => this.refreshKilocode(credentials.refreshToken, proxyOptions)
    };

@@ -299,19 +302,27 @@ export class DefaultExecutor extends BaseExecutor {
    const data = payload?.data || payload;
    const expiresAtIso = data?.expiresAt;
    const expiresIn = expiresAtIso ? Math.max(1, Math.floor((new Date(expiresAtIso).getTime() - Date.now()) / 1000)) : undefined;
-    return { accessToken: data?.accessToken, refreshToken: data?.refreshToken || refreshToken, expiresIn };
+    let accessToken = data?.accessToken;
+    if (accessToken && !accessToken.startsWith("workos:")) {
+      accessToken = `workos:${accessToken}`;
+    }
+    return { accessToken, refreshToken: data?.refreshToken || refreshToken, expiresIn };
  }

-  async refreshKimiCoding(refreshToken, proxyOptions = null) {
-    const kimiHeaders = buildKimiHeaders();
-    const response = await proxyAwareFetch(PROVIDERS["kimi-coding"].refreshUrl, {
+  // CLIProxyAPI DeviceFlowClient.RefreshToken — form body + X-Msh-* headers + stable device_id
+  async refreshKimi(credentials, proxyOptions = null) {
+    const refreshToken = credentials.refreshToken;
+    const cfg = PROVIDERS.kimi || PROVIDERS["kimi-coding"];
+    if (!cfg?.refreshUrl || !cfg?.clientId) return null;
+    const kimiHeaders = buildKimiHeaders(credentials?.providerSpecificData?.deviceId);
+    const response = await proxyAwareFetch(cfg.refreshUrl, {
      method: "POST",
      headers: {
        "Content-Type": "application/x-www-form-urlencoded",
        "Accept": "application/json",
        ...kimiHeaders
      },
-      body: new URLSearchParams({ grant_type: "refresh_token", refresh_token: refreshToken, client_id: PROVIDERS["kimi-coding"].clientId })
+      body: new URLSearchParams({ grant_type: "refresh_token", refresh_token: refreshToken, client_id: cfg.clientId })
    }, proxyOptions);
    if (!response.ok) return null;
    const tokens = await response.json();
--- a/open-sse/executors/devin-cli.js
+++ b/open-sse/executors/devin-cli.js
@@ -0,0 +1,847 @@
+/**
+ * DevinCliExecutor — routes completions through the official Devin CLI binary
+ * via the Agent Client Protocol (ACP) JSON-RPC 2.0 over stdio.
+ *
+ * Protocol flow:
+ *   1. Spawn `devin acp` (default agent = full built-in tools: fs/shell/search).
+ *      Set CLI_DEVIN_AGENT_TYPE=summarizer for a tool-less, text-only mode.
+ *   2. Send: initialize → session/new (with model + cwd + mcpServers) → session/prompt.
+ *   3. Receive: session/update notifications (agent_message_chunk = reply text,
+ *      tool_call/tool_call_update = built-in tool invocations, surfaced as text).
+ *      When devin calls a client-tool from the exposed MCP ("Calling mcp_X from
+ *      clientTools"), it is bridged to an OpenAI tool_use and the turn ends.
+ *   4. Emit deltas as OpenAI-compatible SSE chunks.
+ *   5. Kill subprocess on _cognition.ai/agent_stopped or error.
+ *
+ * Auth: noAuth — the subprocess inherits the parent env and uses credentials
+ * stored by `devin auth login` (~/.local/share/devin/credentials.toml).
+ *
+ * Binary discovery: CLI_DEVIN_BIN env → PATH lookup → platform installer paths.
+ */
+
+import { spawn } from "node:child_process";
+import path from "node:path";
+import os from "node:os";
+import fs from "node:fs";
+import { BaseExecutor } from "./base.js";
+
+// ─── Binary discovery ────────────────────────────────────────────────────────
+
+function resolveDevinBin() {
+  // 1. Explicit override
+  const envBin = process.env.CLI_DEVIN_BIN?.trim();
+  if (envBin) return envBin;
+
+  const isWin = process.platform === "win32";
+  const home = os.homedir();
+
+  // 2. Known installer / package-manager locations. spawn uses shell:false on
+  //    macOS/Linux, so process.env.PATH alone may miss ~/.local/bin, Homebrew,
+  //    Scoop, etc. when the server runs detached (tray/daemon/launchd) without
+  //    a login shell — probe these explicitly before falling back to PATH.
+  const candidates = isWin
+    ? [
+      // Official installer: %LOCALAPPDATA%\devin\cli\bin\devin.exe
+      path.join(process.env.LOCALAPPDATA || path.join(home, "AppData", "Local"), "devin", "cli", "bin", "devin.exe"),
+      path.join(home, ".local", "bin", "devin.exe"),
+      path.join(home, "scoop", "shims", "devin.exe"),
+      path.join(process.env.LOCALAPPDATA || path.join(home, "AppData", "Local"), "Programs", "devin", "devin.exe"),
+    ]
+    : [
+      path.join(home, ".local", "share", "devin", "bin", "devin"),
+      path.join(home, ".devin", "bin", "devin"),
+      path.join(home, ".local", "bin", "devin"), // pipx / user install
+      "/opt/homebrew/bin/devin",                  // Homebrew (Apple Silicon)
+      "/usr/local/bin/devin",                     // Homebrew (Intel) / manual
+      "/usr/bin/devin",
+    ];
+  for (const candidate of candidates) {
+    if (fs.existsSync(candidate)) return candidate;
+  }
+
+  // 3. Fallback — rely on process.env.PATH
+  return isWin ? "devin.exe" : "devin";
+}
+
+// ─── ACP JSON-RPC helper ────────────────────────────────────────────────────
+
+function rpc(method, params, id) {
+  const msg = { jsonrpc: "2.0", method, params };
+  if (id !== undefined) msg.id = id;
+  return JSON.stringify(msg) + "\n";
+}
+
+// ─── Client-tools → MCP bridge ───────────────────────────────────────────────
+// devin only invokes built-in + MCP tools, not OpenAI function-calling schemas.
+// body.tools are exposed as a stdio MCP server "clientTools" so devin can call
+// them. When devin calls one, we emit OpenAI tool_use and end the turn; the
+// client executes and returns tool_result on the next request. That next request
+// re-spawns with the full history (including tool_calls + tool results) and
+// seeds the MCP server with those results so a re-call gets the real data.
+// Tool schemas via DEVIN_MCP_TOOLS; prior results via DEVIN_MCP_RESULTS.
+
+const CLIENT_TOOLS_MCP_SCRIPT = `
+import readline from "node:readline";
+const TOOLS = JSON.parse(process.env.DEVIN_MCP_TOOLS || "[]");
+const RESULTS = JSON.parse(process.env.DEVIN_MCP_RESULTS || "{}");
+const rl = readline.createInterface({ input: process.stdin });
+function send(o){ process.stdout.write(JSON.stringify(o) + "\\n"); }
+rl.on("line", (line) => {
+  let m; try { m = JSON.parse(line); } catch { return; }
+  if (m.method === "initialize") {
+    send({ jsonrpc: "2.0", id: m.id, result: { protocolVersion: "2024-11-05", capabilities: { tools: {} }, serverInfo: { name: "clientTools", version: "1.0" } } });
+  } else if (m.method === "tools/list") {
+    send({ jsonrpc: "2.0", id: m.id, result: { tools: TOOLS } });
+  } else if (m.method === "tools/call") {
+    const name = m.params?.name || "";
+    const seeded = RESULTS[name];
+    const text = seeded !== undefined
+      ? String(seeded)
+      : "(awaiting client tool_result)";
+    process.stderr.write("[client-tools] tool_call name=" + name + " seeded=" + (seeded !== undefined) + "\\n");
+    send({ jsonrpc: "2.0", id: m.id, result: { content: [{ type: "text", text }] } });
+  }
+});
+`.trimStart();
+
+function ensureClientToolsScript() {
+  const scriptPath = path.join(os.tmpdir(), "9router-devin-client-tools.mjs");
+  // Always rewrite so script upgrades land without a process restart.
+  fs.writeFileSync(scriptPath, CLIENT_TOOLS_MCP_SCRIPT);
+  return scriptPath;
+}
+
+// Map OpenAI tools ([{type:"function",function:{name,description,parameters}}])
+// to MCP tool declarations ([{name,description,inputSchema}]).
+// devin only discovers MCP tools whose name carries the `mcp_` prefix, so we
+// add it here and strip it back when bridging the call to the client.
+const MCP_TOOL_PREFIX = "mcp_";
+function toMcpToolName(name) {
+  return name.startsWith(MCP_TOOL_PREFIX) ? name : MCP_TOOL_PREFIX + name;
+}
+function fromMcpToolName(name) {
+  return name.startsWith(MCP_TOOL_PREFIX) ? name.slice(MCP_TOOL_PREFIX.length) : name;
+}
+
+function buildClientToolsMcp(tools, resultMap) {
+  const mcpTools = [];
+  for (const t of tools) {
+    if (!t) continue;
+    const f = t.function || t;
+    if (!f?.name) continue;
+    mcpTools.push({
+      name: toMcpToolName(f.name),
+      description: f.description || "",
+      inputSchema: f.parameters || f.input_schema || { type: "object", properties: {} },
+    });
+  }
+  if (!mcpTools.length) return null;
+  const env = { DEVIN_MCP_TOOLS: JSON.stringify(mcpTools) };
+  if (resultMap && Object.keys(resultMap).length) {
+    env.DEVIN_MCP_RESULTS = JSON.stringify(resultMap);
+  }
+  return {
+    command: process.execPath,
+    args: [ensureClientToolsScript()],
+    env,
+  };
+}
+
+// Extract tool_result content keyed by MCP tool name (mcp_<original>).
+// Walks messages: assistant.tool_calls id→name, role=tool tool_call_id→content.
+function extractClientToolResults(messages) {
+  const idToMcpName = new Map();
+  const results = {};
+  for (const m of messages) {
+    if (m?.role === "assistant" && Array.isArray(m.tool_calls)) {
+      for (const tc of m.tool_calls) {
+        const name = tc?.function?.name || tc?.name;
+        if (tc?.id && name) idToMcpName.set(tc.id, toMcpToolName(name));
+      }
+    }
+    // Claude-style tool_use blocks in content
+    if (m?.role === "assistant" && Array.isArray(m.content)) {
+      for (const b of m.content) {
+        if (b?.type === "tool_use" && b.id && b.name) {
+          idToMcpName.set(b.id, toMcpToolName(b.name));
+        }
+      }
+    }
+    if (m?.role === "tool" && m.tool_call_id) {
+      const mcpName = idToMcpName.get(m.tool_call_id);
+      if (mcpName) {
+        results[mcpName] =
+          typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? "");
+      }
+    }
+    // Claude-style tool_result blocks in user content
+    if (m?.role === "user" && Array.isArray(m.content)) {
+      for (const b of m.content) {
+        if (b?.type === "tool_result" && b.tool_use_id) {
+          const mcpName = idToMcpName.get(b.tool_use_id);
+          if (mcpName) {
+            const c = b.content;
+            results[mcpName] =
+              typeof c === "string" ? c : JSON.stringify(c ?? "");
+          }
+        }
+      }
+    }
+  }
+  return results;
+}
+
+// Resolve workspace cwd from client request (Codex/CLI env context, body fields).
+// Prefer an absolute existing path so agent file tools hit the user's project
+// instead of os.tmpdir() (which made relative create/delete inconsistent).
+function resolveWorkspaceCwd(body) {
+  const candidates = [];
+  const push = (v) => {
+    if (typeof v === "string" && v.trim()) candidates.push(v.trim());
+  };
+  push(body?.cwd);
+  push(body?.working_directory);
+  push(body?.workdir);
+  push(body?.workspace);
+  push(body?.metadata?.cwd);
+  push(body?.metadata?.working_directory);
+
+  const scanText = (text) => {
+    if (typeof text !== "string") return;
+    for (const m of text.matchAll(/<cwd>\s*([^<]+?)\s*<\/cwd>/gi)) push(m[1]);
+  };
+  const scanMessages = (msgs) => {
+    if (!Array.isArray(msgs)) return;
+    for (const msg of msgs) {
+      if (!msg) continue;
+      if (typeof msg.content === "string") scanText(msg.content);
+      else if (Array.isArray(msg.content)) {
+        for (const p of msg.content) {
+          if (typeof p === "string") scanText(p);
+          else if (p && typeof p === "object") {
+            scanText(p.text);
+            scanText(p.input_text);
+            scanText(p.content);
+          }
+        }
+      }
+      // Responses API input items
+      if (typeof msg === "string") scanText(msg);
+      if (msg.type === "message" && Array.isArray(msg.content)) {
+        for (const p of msg.content) scanText(p?.text || p?.input_text);
+      }
+    }
+  };
+  scanMessages(body?.messages);
+  scanMessages(body?.input);
+
+  for (const c of candidates) {
+    try {
+      if (path.isAbsolute(c) && fs.existsSync(c) && fs.statSync(c).isDirectory()) {
+        return c;
+      }
+    } catch {
+      /* ignore */
+    }
+  }
+  return os.tmpdir();
+}
+
+// ─── Multi-turn message → single prompt builder ─────────────────────────────
+
+function buildPromptText(messages) {
+  // Inline the whole conversation so the model has full context, including
+  // prior tool_calls / tool_results so it can continue after a client round-trip.
+  const lines = [];
+  for (const m of messages) {
+    const role = String(m.role || "user");
+    let text = "";
+    if (typeof m.content === "string") {
+      text = m.content;
+    } else if (Array.isArray(m.content)) {
+      for (const p of m.content) {
+        if (!p || typeof p !== "object") continue;
+        if (p.type === "text") text += String(p.text || "");
+        else if (p.type === "tool_use") {
+          text += `\n[Tool call ${p.name} id=${p.id}]\n${JSON.stringify(p.input ?? {})}\n`;
+        } else if (p.type === "tool_result") {
+          const c =
+            typeof p.content === "string" ? p.content : JSON.stringify(p.content ?? "");
+          text += `\n[Tool result id=${p.tool_use_id}]\n${c}\n`;
+        }
+      }
+    }
+    // OpenAI tool_calls on assistant messages
+    if (role === "assistant" && Array.isArray(m.tool_calls) && m.tool_calls.length) {
+      const parts = m.tool_calls.map((tc) => {
+        const name = tc.function?.name || tc.name || "tool";
+        const args = tc.function?.arguments ?? tc.arguments ?? {};
+        const argStr = typeof args === "string" ? args : JSON.stringify(args);
+        return `[Tool call ${name} id=${tc.id}]\n${argStr}`;
+      });
+      text = [text, ...parts].filter(Boolean).join("\n\n");
+    }
+    // OpenAI role=tool messages
+    if (role === "tool") {
+      const c = typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? "");
+      text = `[Tool result id=${m.tool_call_id || ""}]\n${c}`;
+    }
+    if (!text.trim()) continue;
+    if (role === "system") {
+      lines.push(`[System]\n${text}`);
+    } else if (role === "assistant") {
+      lines.push(`[Assistant]\n${text}`);
+    } else if (role === "tool") {
+      lines.push(`[Tool]\n${text}`);
+    } else {
+      lines.push(`[User]\n${text}`);
+    }
+  }
+  return lines.join("\n\n") || "(empty)";
+}
+
+// ─── DevinCliExecutor ─────────────────────────────────────────────────────────
+
+export class DevinCliExecutor extends BaseExecutor {
+  constructor() {
+    super("devin-cli", { id: "devin-cli", baseUrl: "devin://acp/stdio" });
+  }
+
+  buildUrl() {
+    return "devin://acp/stdio";
+  }
+
+  buildHeaders() {
+    return {};
+  }
+
+  transformRequest() {
+    return null;
+  }
+
+  async execute({ model, body, credentials, signal, log }) {
+    const b = body ?? {};
+    const messages = Array.isArray(b.messages)
+      ? b.messages
+      : Array.isArray(b.input)
+        ? b.input
+        : [];
+    const promptText = buildPromptText(messages);
+    const workspaceCwd = resolveWorkspaceCwd(b);
+    const devinBin = resolveDevinBin();
+
+    log?.info?.(
+      "DEVIN",
+      `devin acp → model=${model}, bin=${devinBin}, cwd=${workspaceCwd}`
+    );
+
+    // Optional MCP servers via DEVIN_MCP_SERVERS (JSON object, devin config format):
+    //   {"echo":{"command":"/abs/node","args":["/srv/echo.js"],"env":{"K":"V"}}}
+    // Plus body.tools (OpenAI schema) → exposed as a "clientTools" MCP
+    // server so devin can invoke client-defined tools (bridged back in Phase 2).
+    // When any are present, a throwaway XDG_CONFIG_HOME holds devin/config.json so
+    // the agent auto-connects them (session/new mcpServers alone doesn't spawn
+    // them — see ACP mcp/connect, still unstable). Cleaned up on finish.
+    // NOTE: this replaces the user's global devin MCP config for the subprocess.
+    let mcpConfigDir = null;
+    const mcpServers = {};
+    const mcpJson = process.env.DEVIN_MCP_SERVERS?.trim();
+    if (mcpJson) {
+      try {
+        Object.assign(mcpServers, JSON.parse(mcpJson));
+      } catch (e) {
+        log?.info?.("DEVIN", `DEVIN_MCP_SERVERS parse failed: ${e.message}`);
+      }
+    }
+    const clientTools = Array.isArray(b.tools) ? b.tools.filter(Boolean) : [];
+    const clientToolResults = extractClientToolResults(messages);
+    const clientToolsMcp = buildClientToolsMcp(clientTools, clientToolResults);
+    const hasClientTools = !!clientToolsMcp;
+    if (clientToolsMcp) {
+      mcpServers["clientTools"] = clientToolsMcp;
+      const seeded = Object.keys(clientToolResults).length;
+      log?.info?.(
+        "DEVIN",
+        `exposing ${clientTools.length} client tool(s) as MCP` +
+          (seeded ? ` (seeded ${seeded} result(s))` : "")
+      );
+    }
+    if (Object.keys(mcpServers).length) {
+      try {
+        mcpConfigDir = fs.mkdtempSync(path.join(os.tmpdir(), "devin-mcp-"));
+        const cfgDev = path.join(mcpConfigDir, "devin");
+        fs.mkdirSync(cfgDev, { recursive: true });
+        fs.writeFileSync(
+          path.join(cfgDev, "config.json"),
+          JSON.stringify({ mcpServers })
+        );
+        log?.info?.("DEVIN", `mcp config written → ${mcpConfigDir}`);
+      } catch (e) {
+        log?.info?.("DEVIN", `mcp config write failed: ${e.message}`);
+        mcpConfigDir = null;
+      }
+    }
+    const cleanupMcp = () => {
+      if (!mcpConfigDir) return;
+      try {
+        fs.rmSync(mcpConfigDir, { recursive: true, force: true });
+      } catch {
+        /* ignore */
+      }
+      mcpConfigDir = null;
+    };
+
+    const sseStream = new ReadableStream({
+      start(controller) {
+        const enc = new TextEncoder();
+        const emit = (data) => controller.enqueue(enc.encode(data));
+
+        // Inherit the parent environment so devin resolves stored CLI credentials
+        // (~/.local/share/devin/credentials.toml from `devin auth login`). Do NOT
+        // inject WINDSURF_API_KEY: this provider is noAuth, and a bogus/leaked key
+        // overrides stored creds and makes devin return -32000 "invalid api key".
+        const env = { ...process.env };
+        // Auto-approve tool execution so the agent doesn't block waiting for a
+        // session/request_permission response we never send (default mode would
+        // hang the stream on the first shell/exec tool call). Override via env.
+        // WARNING: bypass lets the agent run shell/modify FS unattended — local only.
+        env.DEVIN_PERMISSION_MODE = process.env.DEVIN_PERMISSION_MODE || "bypass";
+        if (mcpConfigDir) env.XDG_CONFIG_HOME = mcpConfigDir;
+
+        // Agent type: default (omitted) = full agent with built-in tools
+        // (fs/shell/search) so the model can actually perform tasks. Override to
+        // `summarizer` (no tools, text-only) via CLI_DEVIN_AGENT_TYPE for a safer,
+        // tool-less mode. WARNING: the default agent can run shell commands and
+        // modify the filesystem on the host running 9router — only expose locally.
+        const agentType = process.env.CLI_DEVIN_AGENT_TYPE?.trim();
+        const acpArgs = ["acp"];
+        if (agentType) acpArgs.push("--agent-type", agentType);
+
+        // Spawn in the client workspace cwd (from <cwd> env context) so built-in
+        // file tools create/delete relative paths in the user's project.
+        // MCP config still comes from XDG_CONFIG_HOME (throwaway), not project .devin/.
+        const child = spawn(devinBin, acpArgs, {
+          env,
+          cwd: workspaceCwd,
+          stdio: ["pipe", "pipe", "pipe"],
+          // On Windows, devin.exe may need shell resolution
+          shell: process.platform === "win32",
+        });
+
+        let spawnError = null;
+        let stdinClosed = false;
+
+        child.on("error", (err) => {
+          spawnError = err;
+          const msg =
+            err.message.includes("ENOENT") || err.message.includes("not found")
+              ? `Devin CLI not found: ${devinBin}. Install via https://cli.devin.ai or set CLI_DEVIN_BIN env var.`
+              : `Devin CLI spawn error: ${err.message}`;
+          emit(
+            `data: ${JSON.stringify({ error: { message: msg, type: "devin_cli_error", code: "spawn_failed" } })}\n\n`
+          );
+          emit("data: [DONE]\n\n");
+          controller.close();
+        });
+
+        if (signal) {
+          signal.addEventListener("abort", () => {
+            if (!child.killed) child.kill("SIGTERM");
+          });
+        }
+
+        // ── JSON-RPC state machine ──────────────────────────────────────────
+        let idCounter = 1;
+        let sessionId = null;
+        let initDone = false;
+        let sessionCreated = false;
+        let promptSent = false;
+        const responseId = `chatcmpl-devin-${Date.now()}`;
+        const created = Math.floor(Date.now() / 1000);
+        let roleEmitted = false;
+        let totalText = "";
+        let finished = false;
+
+        const sendRpc = (method, params) => {
+          if (stdinClosed || child.stdin.destroyed) return;
+          const id = idCounter++;
+          try {
+            child.stdin.write(rpc(method, params, id));
+          } catch {
+            /* ignore write errors after close */
+          }
+          return id;
+        };
+
+        // Emit a content delta as an OpenAI-compatible SSE chunk (handles the
+        // leading role chunk once).
+        const emitDelta = (delta) => {
+          if (!roleEmitted) {
+            emit(
+              `data: ${JSON.stringify({
+                id: responseId,
+                object: "chat.completion.chunk",
+                created,
+                model,
+                choices: [{ index: 0, delta: { role: "assistant", content: "" }, finish_reason: null }],
+              })}\n\n`
+            );
+            roleEmitted = true;
+          }
+          totalText += delta;
+          emit(
+            `data: ${JSON.stringify({
+              id: responseId,
+              object: "chat.completion.chunk",
+              created,
+              model,
+              choices: [{ index: 0, delta: { content: delta }, finish_reason: null }],
+            })}\n\n`
+          );
+        };
+
+        // Emit an OpenAI tool_call delta (function calling). Ends the turn with
+        // finish_reason "tool_calls" so the client executes and returns tool_result.
+        let toolUseEmitted = false;
+        // ACP tool_call is upsert-by-id: the first event has title, a later update
+        // may only carry rawInput (title omitted). Track pending client-tool calls.
+        const pendingClientTools = new Map(); // toolCallId → original tool name
+        const emitToolUse = (toolName, args, toolCallId) => {
+          const argsStr = typeof args === "string" ? args : JSON.stringify(args ?? {});
+          if (!roleEmitted) {
+            emit(
+              `data: ${JSON.stringify({
+                id: responseId,
+                object: "chat.completion.chunk",
+                created,
+                model,
+                choices: [{ index: 0, delta: { role: "assistant", content: null }, finish_reason: null }],
+              })}\n\n`
+            );
+            roleEmitted = true;
+          }
+          emit(
+            `data: ${JSON.stringify({
+              id: responseId,
+              object: "chat.completion.chunk",
+              created,
+              model,
+              choices: [
+                {
+                  index: 0,
+                  delta: {
+                    tool_calls: [
+                      {
+                        index: 0,
+                        id: toolCallId,
+                        type: "function",
+                        function: { name: toolName, arguments: argsStr },
+                      },
+                    ],
+                  },
+                  finish_reason: null,
+                },
+              ],
+            })}\n\n`
+          );
+        };
+
+        const finish = (error, finishReason = "stop") => {
+          if (finished) return;
+          finished = true;
+
+          if (error) {
+            emit(
+              `data: ${JSON.stringify({ error: { message: error, type: "devin_cli_error" } })}\n\n`
+            );
+          } else {
+            // Emit finish chunk
+            emit(
+              `data: ${JSON.stringify({
+                id: responseId,
+                object: "chat.completion.chunk",
+                created,
+                model,
+                choices: [{ index: 0, delta: {}, finish_reason: finishReason }],
+                usage: {
+                  prompt_tokens: Math.ceil(promptText.length / 4),
+                  completion_tokens: Math.ceil(totalText.length / 4),
+                  total_tokens: Math.ceil((promptText.length + totalText.length) / 4),
+                  estimated: true,
+                },
+              })}\n\n`
+            );
+          }
+          emit("data: [DONE]\n\n");
+
+          // Gracefully close stdin → devin will exit
+          try {
+            if (!stdinClosed) {
+              stdinClosed = true;
+              child.stdin.end();
+            }
+          } catch {
+            /* ignore */
+          }
+
+          // Give it 2s to exit cleanly, then SIGKILL
+          const killTimer = setTimeout(() => {
+            if (!child.killed) child.kill("SIGKILL");
+          }, 2000);
+          killTimer.unref?.();
+
+          controller.close();
+          cleanupMcp();
+        };
+
+        // ── stdout reader (NDJSON) ──────────────────────────────────────────
+        let buffer = "";
+
+        child.stdout.on("data", (chunk) => {
+          buffer += chunk.toString("utf8");
+          let nl;
+          // Each ACP message is a newline-terminated JSON line
+          while ((nl = buffer.indexOf("\n")) !== -1) {
+            const line = buffer.slice(0, nl).trim();
+            buffer = buffer.slice(nl + 1);
+            if (!line) continue;
+
+            let msg;
+            try {
+              msg = JSON.parse(line);
+            } catch {
+              continue; // ignore non-JSON lines (banner text, etc.)
+            }
+
+            // ── Initialize response ───────────────────────────────────────
+            if (!initDone && msg.result !== undefined && !msg.method) {
+              initDone = true;
+              // Create session with the client workspace cwd so agent file tools
+              // resolve relative paths against the project (not /tmp).
+              // `mcpServers` is required by devin 3000.2.x (must be a sequence);
+              // omitting it returns -32602 "Invalid params: missing field mcpServers".
+              sendRpc("session/new", {
+                cwd: workspaceCwd,
+                mcpServers: [],
+                model: model || undefined,
+              });
+              continue;
+            }
+
+            // ── session/new response → get sessionId ──────────────────────
+            if (initDone && !sessionCreated && msg.result !== undefined && !msg.method) {
+              const res = msg.result || {};
+              sessionId = res.sessionId || null;
+              if (!sessionId) {
+                finish("Devin ACP: session/new returned no sessionId");
+                return;
+              }
+              sessionCreated = true;
+              // Send the prompt. devin 3000.2.x expects `prompt` (a sequence),
+              // not `content` — using `content` returns -32602 "missing field prompt".
+              promptSent = true;
+              sendRpc("session/prompt", {
+                sessionId,
+                prompt: [{ type: "text", text: promptText }],
+              });
+              continue;
+            }
+
+            // ── session/prompt response (ack / final result) ────────────
+            if (sessionCreated && promptSent && msg.result !== undefined && !msg.method) {
+              // Devin 3000.2.x only resolves session/prompt with the final result
+              // (stopReason) after streaming completes. Streaming notifications are
+              // handled below; nothing to do here unless we never streamed.
+              if (!roleEmitted) {
+                const res = msg.result || undefined;
+                const content = extractResultText(res);
+                if (content) {
+                  totalText = content;
+                  emitDelta(content);
+                }
+                const stopReason = (res && res.stopReason) || "";
+                if (stopReason && stopReason !== "cancelled") {
+                  finish();
+                  return;
+                }
+              }
+              continue;
+            }
+
+            // ── Permission requests → auto-approve the first allow option ──
+            // Devi asks before running shell/exec tools; as a headless proxy we
+            // grant once. (DEVIN_PERMISSION_MODE=bypass usually prevents these,
+            // but some tool kinds still prompt, so handle them here too.)
+            if (msg.method === "session/request_permission" && msg.id !== undefined) {
+              const options = msg.params?.options || [];
+              const allow =
+                options.find((o) => /allow/i.test(String(o.kind || ""))) || options[0];
+              if (allow) {
+                child.stdin.write(
+                  JSON.stringify({
+                    jsonrpc: "2.0",
+                    id: msg.id,
+                    result: { outcome: { outcome: "selected", optionId: allow.optionId } },
+                  }) + "\n"
+                );
+              }
+              continue;
+            }
+
+            // ── Agent stopped notification (devin 3000.2.x stop signal) ───
+            if (msg.method === "_cognition.ai/agent_stopped" || msg.method === "$/agent_stopped") {
+              const cause = msg.params?.cause;
+              if (cause === "error") {
+                // devin uses errorMessage on this notification (not message/error).
+                const errText =
+                  msg.params?.errorMessage ||
+                  msg.params?.message ||
+                  msg.params?.error ||
+                  "Devin agent error";
+                finish(String(errText));
+              } else {
+                finish();
+              }
+              return;
+            }
+
+            // ── Streaming notifications (session/update) ──────────────────
+            if (msg.method === "session/update" || msg.method === "$/update") {
+              const params = msg.params;
+              if (!params) continue;
+
+              // devin 3000.2.x nests the payload under params.update.sessionUpdate;
+              // older devin used a flat params.type.
+              const update = params.update || {};
+              const type = update.sessionUpdate || params.type;
+              const contentField = update.content !== undefined ? update.content : params.content;
+              const deltaText =
+                typeof contentField === "string"
+                  ? contentField
+                  : contentField?.text ?? params.delta ?? params.text ?? "";
+
+              // ── Client-tool bridge: devin calling a tool from our exposed MCP ──
+              // ACP title shape: "Calling mcp_<name> from clientTools".
+              // tool_call is upsert-by-id: title may only appear on the first event,
+              // rawInput on a later tool_call_update. Track pending ids so we don't
+              // require both fields on the same notification.
+              if (
+                hasClientTools &&
+                !toolUseEmitted &&
+                (type === "tool_call" || type === "tool_call_update")
+              ) {
+                const tcId = update.toolCallId;
+                if (typeof update.title === "string" && update.title.startsWith("Calling mcp_") && /from clientTools\b/.test(update.title)) {
+                  const nameMatch = update.title.match(/^Calling (mcp_\S+)\b/);
+                  const mcpName = nameMatch ? nameMatch[1] : "";
+                  const origName = fromMcpToolName(mcpName);
+                  if (tcId && origName) pendingClientTools.set(tcId, origName);
+                }
+                const origName = tcId ? pendingClientTools.get(tcId) : null;
+                if (origName && update.rawInput) {
+                  toolUseEmitted = true;
+                  pendingClientTools.delete(tcId);
+                  emitToolUse(origName, update.rawInput, tcId || `call_${Date.now()}`);
+                  finish(null, "tool_calls");
+                  return;
+                }
+                continue;
+              }
+
+              if (type === "agent_message_chunk" || type === "message_delta" || type === "text_delta" || type === "content_delta") {
+                if (deltaText) emitDelta(deltaText);
+              } else if (type === "agent_thought_chunk") {
+                // Internal reasoning — not surfaced to the client.
+              } else if (type === "message_stop" || type === "stop" || type === "done") {
+                finish();
+                return;
+              } else if (type === "error") {
+                finish(String(params.message || params.error || "Devin ACP error"));
+                return;
+              }
+              continue;
+            }
+
+            // ── Error responses ───────────────────────────────────────────
+            if (msg.error) {
+              finish(`Devin ACP error ${msg.error.code}: ${msg.error.message}`);
+              return;
+            }
+          }
+        });
+
+        child.stderr.on("data", (chunk) => {
+          log?.debug?.("DEVIN", `stderr: ${chunk.toString("utf8").slice(0, 200)}`);
+        });
+
+        child.on("close", (code) => {
+          if (!finished) {
+            if (code !== 0 && !spawnError) {
+              finish(roleEmitted ? undefined : `Devin CLI exited with code ${code}`);
+            } else {
+              finish();
+            }
+          } else {
+            cleanupMcp();
+          }
+        });
+
+        // ── Send initialize ───────────────────────────────────────────────
+        sendRpc("initialize", {
+          protocolVersion: "0.3",
+          clientInfo: { name: "9router", version: "1.0" },
+          capabilities: {},
+        });
+      },
+    });
+
+    return {
+      response: new Response(sseStream, {
+        status: 200,
+        headers: {
+          "Content-Type": "text/event-stream",
+          "Cache-Control": "no-cache",
+          Connection: "keep-alive",
+        },
+      }),
+      url: "devin://acp/stdio",
+      headers: {},
+      transformedBody: {
+        model,
+        cwd: workspaceCwd,
+        clientTools: clientTools.map((t) => t?.function?.name || t?.name).filter(Boolean),
+        clientToolResults: Object.keys(clientToolResults),
+        mcpServers: Object.keys(mcpServers),
+        promptLength: Array.isArray(body?.messages)
+          ? body.messages.length
+          : Array.isArray(body?.input)
+            ? body.input.length
+            : 0,
+      },
+    };
+  }
+}
+
+// ─── Helpers ─────────────────────────────────────────────────────────────────
+
+// Extract text from a final ACP session/prompt result object across common shapes.
+function extractResultText(result) {
+  // { message: { content: "..." } }
+  // { messages: [{ content: "..." }] }
+  // { content: "..." }
+  // { text: "..." }
+  if (typeof result.content === "string") return result.content;
+  if (typeof result.text === "string") return result.text;
+  const msg = result.message;
+  if (msg && typeof msg.content === "string") return msg.content;
+  const msgs = result.messages;
+  if (Array.isArray(msgs)) {
+    return msgs
+      .filter((m) => m.role === "assistant")
+      .map((m) => String(m.content || ""))
+      .join("\n");
+  }
+  return "";
+}
+
+export default DevinCliExecutor;
--- a/open-sse/executors/github.js
+++ b/open-sse/executors/github.js
@@ -4,11 +4,13 @@ import { OAUTH_ENDPOINTS, GITHUB_COPILOT } from "../config/appConstants.js";
 import { HTTP_STATUS } from "../config/runtimeConfig.js";
 import { openaiToOpenAIResponsesRequest } from "../translator/request/openai-responses.js";
 import { openaiResponsesToOpenAIResponse } from "../translator/response/openai-responses.js";
-import { initState } from "../translator/index.js";
+import { initState, translateRequest, translateResponse } from "../translator/index.js";
+import { FORMATS } from "../translator/formats.js";
 import { parseSSELine, formatSSE } from "../utils/streamHelpers.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
 import { stripUnsupportedParams } from "../translator/concerns/paramSupport.js";
 import { SSE_DONE } from "../utils/sseConstants.js";
+import { ANTHROPIC_API_VERSION } from "../providers/shared.js";
 import crypto from "crypto";

 export class GithubExecutor extends BaseExecutor {
@@ -17,6 +19,16 @@ export class GithubExecutor extends BaseExecutor {
    this.knownCodexModels = new Set();
  }

+  // Claude models get routed to Copilot's Anthropic-native /v1/messages shim (see
+  // executeWithMessagesEndpoint below) — the only Copilot endpoint that surfaces
+  // prompt-cache token counts. gpt/gemini/grok models stay on /chat/completions
+  // (or /responses). Name-pattern check, not a registry field: Copilot's live model
+  // catalog (services/copilotModels.js) regularly exposes claude-* variants ahead
+  // of the static registry (registry/github.js).
+  isClaudeModel(model) {
+    return /claude/i.test(model || "");
+  }
+
  buildUrl(model, stream, urlIndex = 0) {
    return this.config.baseUrl;
  }
@@ -35,47 +47,20 @@ export class GithubExecutor extends BaseExecutor {
      "x-request-id": crypto.randomUUID?.() || `${Date.now()}-${Math.random().toString(36).slice(2)}`,
      "x-vscode-user-agent-library-version": "electron-fetch",
      "X-Initiator": "user",
+      // Harmless no-op on /chat/completions and /responses; required by /v1/messages.
+      "anthropic-version": ANTHROPIC_API_VERSION,
      "Accept": stream ? "text/event-stream" : "application/json"
    };
  }

-  // Sanitize messages for GitHub Copilot /chat/completions endpoint.
+  // Sanitize messages for GitHub Copilot /chat/completions endpoint (gpt/gemini/grok models —
+  // claude models never reach this, see execute() below).
  // The endpoint only accepts 'text' and 'image_url' content part types.
  // Tool-related content (tool_use, tool_result, thinking) must be serialized as text.
  sanitizeMessagesForChatCompletions(body) {
    if (!body?.messages) return body;

    const sanitized = { ...body };
-    
-    // Handle response_format for Claude models via GitHub
-    // GitHub's internal translation doesn't respect response_format, so we inject it as a system prompt
-    // AND prepend a reminder to the last user message for maximum effectiveness
-    if (body.response_format && body.model?.includes('claude')) {
-      const responseFormat = body.response_format;
-      let systemInstruction = '';
-      if (responseFormat.type === 'json_schema' && responseFormat.json_schema?.schema) {
-        systemInstruction = 'CRITICAL: You must ONLY output raw JSON. Never use markdown code blocks. Never use backticks. Never wrap JSON in triple backticks. Output ONLY the raw JSON object.';
-      } else if (responseFormat.type === 'json_object') {
-        systemInstruction = 'CRITICAL: You must ONLY output raw JSON. Never use markdown code blocks. Never use backticks.';
-      }
-      if (systemInstruction) {
-        // Add to system message
-        const systemIdx = body.messages.findIndex(m => m.role === 'system');
-        if (systemIdx >= 0) {
-          body.messages[systemIdx].content = systemInstruction + '\n\n' + body.messages[systemIdx].content;
-        } else {
-          body.messages.unshift({ role: 'system', content: systemInstruction });
-        }
-        
-        // Also prepend to the last user message as a reminder
-        const lastUserIdx = body.messages.map((m, i) => m.role === 'user' ? i : -1).filter(i => i >= 0).pop();
-        if (lastUserIdx >= 0) {
-          const userMsg = body.messages[lastUserIdx];
-          const userContent = typeof userMsg.content === 'string' ? userMsg.content : JSON.stringify(userMsg.content);
-          userMsg.content = 'Respond with ONLY raw JSON (no markdown, no backticks, no code blocks): ' + userContent;
-        }
-      }
-    }
    sanitized.messages = body.messages.map(msg => {
      // assistant messages with only tool_calls have content: null — leave as-is
      if (!msg.content) return msg;
@@ -138,6 +123,15 @@ export class GithubExecutor extends BaseExecutor {
  async execute(options) {
    const { model, log } = options;

+    // Claude models: route to Copilot's Anthropic-native /v1/messages shim — the only
+    // Copilot endpoint that surfaces prompt-cache token counts for Claude. Detected by
+    // model NAME (not a registry field): Copilot's live model catalog regularly exposes
+    // claude-* variants the static registry hasn't caught up with yet (see registry/github.js).
+    if (this.isClaudeModel(model)) {
+      log?.debug("GITHUB", `Using /v1/messages route for ${model}`);
+      return this.executeWithMessagesEndpoint(options);
+    }
+
    // Only use /responses for models that are explicitly known to need it (e.g. gpt codex models)
    // and that the /responses endpoint actually serves (excludes Gemini/Claude, see #1062).
    if (this.knownCodexModels.has(model) && this.supportsResponsesEndpoint(model)) {
@@ -145,8 +139,8 @@ export class GithubExecutor extends BaseExecutor {
      return this.executeWithResponsesEndpoint(options);
    }

-    // Sanitize messages before sending to /chat/completions
-    // This handles Claude models on GitHub Copilot which reject non-text/image_url content types
+    // Sanitize messages before sending to /chat/completions (gpt/gemini/grok — the
+    // endpoint rejects non-text/image_url content parts).
    const sanitizedOptions = {
      ...options,
      body: this.sanitizeMessagesForChatCompletions(options.body)
@@ -251,6 +245,101 @@ export class GithubExecutor extends BaseExecutor {
    };
  }

+  // Claude models arrive here OpenAI-shape (chatCore.js targets "openai" for github —
+  // see the note in execute() above), so we translate to Anthropic-native ourselves.
+  // This is what makes prepareClaudeRequest() (translator/formats/claude.js) inject
+  // cache_control — /chat/completions never gets there, so it never sees cache tokens.
+  async executeWithMessagesEndpoint({ model, body, stream, credentials, signal, log, proxyOptions = null }) {
+    const url = this.config.messagesUrl;
+    const headers = this.buildHeaders(credentials, stream);
+
+    // Force stream:true upstream regardless of client preference, same as
+    // executeWithResponsesEndpoint below — chatCore.js's non-streaming handler already
+    // knows how to buffer an SSE response into a single JSON reply when the client
+    // asked for stream:false.
+    const transformedBody = translateRequest(FORMATS.OPENAI, FORMATS.CLAUDE, model, body, true, credentials, "github");
+    // _toolNameMap is internal bookkeeping (see openai-to-claude.js) — chatCore.js
+    // normally strips it before dispatch and threads it into the response state to
+    // restore original tool names; we must do the same here, or Anthropic's strict
+    // schema rejects the extra field with a 400.
+    const toolNameMap = transformedBody._toolNameMap;
+    delete transformedBody._toolNameMap;
+
+    log?.debug("GITHUB", "Sending translated request to /v1/messages");
+
+    const response = await proxyAwareFetch(url, {
+      method: "POST",
+      headers,
+      body: JSON.stringify(transformedBody),
+      signal
+    }, proxyOptions);
+
+    if (!response.ok) {
+      return { response, url, headers, transformedBody };
+    }
+
+    const state = initState(FORMATS.CLAUDE);
+    state.model = model;
+    if (toolNameMap) state.toolNameMap = toolNameMap;
+
+    const decoder = new TextDecoder();
+    let buffer = "";
+
+    const emitAll = (controller, chunks) => {
+      for (const c of chunks) {
+        controller.enqueue(new TextEncoder().encode(formatSSE(c, "openai")));
+      }
+    };
+
+    const transformStream = new TransformStream({
+      async transform(chunk, controller) {
+        buffer += decoder.decode(chunk, { stream: true });
+        const lines = buffer.split("\n");
+
+        buffer = lines.pop() || "";
+
+        for (const line of lines) {
+          const trimmed = line.trim();
+          if (!trimmed) continue;
+
+          const parsed = parseSSELine(trimmed);
+          if (!parsed) continue;
+
+          if (parsed.done && stream === true) {
+            controller.enqueue(new TextEncoder().encode(SSE_DONE));
+            continue;
+          }
+
+          emitAll(controller, translateResponse(FORMATS.CLAUDE, FORMATS.OPENAI, parsed, state));
+        }
+      },
+      flush(controller) {
+        if (buffer.trim()) {
+          const parsed = parseSSELine(buffer.trim());
+          if (parsed && !parsed.done) {
+            emitAll(controller, translateResponse(FORMATS.CLAUDE, FORMATS.OPENAI, parsed, state));
+          }
+        }
+      }
+    });
+
+    if (!response.body) {
+      return { response: new Response("", { status: response.status, headers: response.headers }), url, headers, transformedBody };
+    }
+    const convertedStream = response.body.pipeThrough(transformStream);
+
+    return {
+      response: new Response(convertedStream, {
+        status: response.status,
+        statusText: response.statusText,
+        headers: response.headers
+      }),
+      url,
+      headers,
+      transformedBody
+    };
+  }
+
  async refreshCopilotToken(githubAccessToken, log, proxyOptions = null) {
    try {
      const response = await proxyAwareFetch("https://api.github.com/copilot_internal/v2/token", {
--- a/open-sse/executors/grok-cli.js
+++ b/open-sse/executors/grok-cli.js
@@ -0,0 +1,552 @@
+import crypto from "node:crypto";
+import { BaseExecutor } from "./base.js";
+import { PROVIDERS } from "../config/providers.js";
+import {
+  refreshProviderCredentials,
+  shouldRefreshCredentials,
+} from "../services/oauthCredentialManager.js";
+import { normalizeResponsesInput } from "../translator/formats/responsesApi.js";
+import { getModelUpstreamId } from "../config/providerModels.js";
+import {
+  GROK_CLI_CLIENT_IDENTIFIER,
+  GROK_CLI_VERSION,
+  supportsGrokCliReasoningEffort,
+} from "../config/grokCli.js";
+import { MEMORY_CONFIG } from "../config/runtimeConfig.js";
+import { resolveSessionId } from "../utils/sessionManager.js";
+import { getConsistentMachineId } from "../shared/machineId.js";
+
+// Server-generated item id prefixes that /responses cannot resolve when store=false
+const SERVER_ID_PATTERN = /^(rs|fc|resp|msg)_/;
+
+// Hosted tool types executed server-side by Grok CLI backend
+const HOSTED_TOOL_TYPES = new Set([
+  "web_search",
+  "x_search",
+  "web_search_preview",
+  "file_search",
+  "image_generation",
+  "code_interpreter",
+  "mcp",
+  "local_shell",
+]);
+
+// Fields accepted by cli-chat-proxy Responses API (mirrors Codex allowlist + Grok extras)
+const RESPONSES_API_ALLOWLIST = new Set([
+  "model",
+  "input",
+  "instructions",
+  "tools",
+  "tool_choice",
+  "stream",
+  "store",
+  "reasoning",
+  "include",
+  "temperature",
+  "top_p",
+  "max_output_tokens",
+  "parallel_tool_calls",
+  "text",
+  "metadata",
+  "prompt_cache_key",
+]);
+
+const EFFORT_LEVELS = ["low", "medium", "high", "xhigh"];
+const GROK_CLI_TURN_STORE_MAX = 5000;
+const GROK_CLI_NATIVE_ITEM_ID = /^(?:rs|msg|fc)_[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+const GROK_CLI_FREEFORM_TOOL_PARAMETERS = {
+  type: "object",
+  properties: { input: { type: "string" } },
+  required: ["input"],
+};
+
+// Per-session last turn index so multi-turn headers never go backwards within this process
+const sessionTurnStore = new Map();
+let requestTurnStore = new WeakMap();
+
+/**
+ * Count user turns in a Responses `input` array.
+ * Official CLI sets x-grok-turn-idx to the 1-based conversation turn (≈ user messages).
+ * HAR: first chat turn → "1".
+ */
+export function countGrokCliUserTurns(input) {
+  if (!Array.isArray(input)) return 1;
+  let n = 0;
+  for (const item of input) {
+    if (!item || typeof item !== "object" || Array.isArray(item)) continue;
+    const type = typeof item.type === "string" ? item.type : "";
+    // Responses message items (type omitted or "message") with role user
+    if (item.role === "user" && (!type || type === "message")) n += 1;
+  }
+  return Math.max(1, n);
+}
+
+/**
+ * Resolve monotonic turn index for a session.
+ * Prefers user-message count from the payload (full history clients), but never
+ * decreases vs the last index observed for the same sessionId in this process.
+ */
+export function resolveGrokCliTurnIdx(sessionId, input, requestKey = null) {
+  const fromInput = countGrokCliUserTurns(input);
+  if (!sessionId) return fromInput;
+
+  if (requestKey && requestTurnStore.has(requestKey)) {
+    return requestTurnStore.get(requestKey);
+  }
+
+  const now = Date.now();
+  const existing = sessionTurnStore.get(sessionId);
+  const prev = existing && now - existing.lastUsed <= MEMORY_CONFIG.sessionTtlMs
+    ? existing.turn
+    : 0;
+  if (existing) sessionTurnStore.delete(sessionId);
+
+  // A new delta-style request advances the turn; retries reuse requestKey.
+  const turn = prev > 0 ? Math.max(fromInput, prev + (requestKey ? 1 : 0)) : fromInput;
+  while (sessionTurnStore.size >= GROK_CLI_TURN_STORE_MAX) {
+    sessionTurnStore.delete(sessionTurnStore.keys().next().value);
+  }
+  sessionTurnStore.set(sessionId, { turn, lastUsed: now });
+  if (requestKey) requestTurnStore.set(requestKey, turn);
+  return turn;
+}
+
+/** Test helper — clear in-memory turn counters */
+export function _resetGrokCliTurnStore() {
+  sessionTurnStore.clear();
+  requestTurnStore = new WeakMap();
+}
+
+export function _getGrokCliTurnStoreSize() {
+  return sessionTurnStore.size;
+}
+
+export function normalizeGrokCliEffort(value) {
+  const effort = typeof value === "string" ? value.trim().toLowerCase() : "";
+  if (effort === "max") return "xhigh";
+  if (EFFORT_LEVELS.includes(effort)) return effort;
+  return "high";
+}
+
+export { supportsGrokCliReasoningEffort } from "../config/grokCli.js";
+
+export function resolveGrokCliSessionId(credentials, body) {
+  // ponytail: clients without stable thread metadata share one connection session;
+  // split further when their wire format exposes a durable conversation id.
+  const explicitSessionBody = {
+    prompt_cache_key: body?.prompt_cache_key,
+    session_id: body?.session_id,
+    conversation_id: body?.conversation_id,
+    metadata: body?.metadata,
+  };
+  return resolveSessionId({
+    headers: credentials?.rawHeaders,
+    body: explicitSessionBody,
+    connectionId: credentials?.connectionId || credentials?.id,
+    workspaceId: credentials?.providerSpecificData?.workspaceId,
+    scope: "grok-cli",
+  });
+}
+
+function stringifyGrokCliToolOutput(output) {
+  if (typeof output === "string") return output;
+  if (output === undefined) return "";
+  return JSON.stringify(output);
+}
+
+function isNativeGrokCliItemId(id) {
+  return typeof id === "string" && GROK_CLI_NATIVE_ITEM_ID.test(id);
+}
+
+function normalizeGrokCliInputItem(item) {
+  if (!item || typeof item !== "object" || Array.isArray(item)) return item;
+  const { internal_chat_message_metadata_passthrough: _metadata, ...clean } = item;
+
+  if (item.type === "reasoning") {
+    if (!isNativeGrokCliItemId(item.id) || typeof item.encrypted_content !== "string") return null;
+    return clean;
+  }
+
+  if (item.type === "custom_tool_call") {
+    const callId = item.call_id || item.id;
+    const name = typeof item.name === "string" ? item.name.trim() : "";
+    if (!callId || !name) return null;
+    return {
+      type: "function_call",
+      call_id: callId,
+      name,
+      arguments: JSON.stringify({ input: stringifyGrokCliToolOutput(item.input ?? item.arguments) }),
+    };
+  }
+
+  if (item.type === "custom_tool_call_output" || item.type === "function_call_output") {
+    const callId = item.call_id || item.id;
+    if (!callId) return null;
+    return {
+      type: "function_call_output",
+      call_id: callId,
+      output: stringifyGrokCliToolOutput(item.output),
+    };
+  }
+
+  if (item.type === "function_call") {
+    const callId = item.call_id || item.id;
+    const name = typeof item.name === "string" ? item.name.trim() : "";
+    if (!callId || !name) return null;
+    return {
+      type: "function_call",
+      ...(isNativeGrokCliItemId(item.id) ? { id: item.id } : {}),
+      call_id: callId,
+      name,
+      arguments: typeof item.arguments === "string" ? item.arguments : JSON.stringify(item.arguments ?? {}),
+      ...(typeof item.status === "string" ? { status: item.status } : {}),
+    };
+  }
+
+  return clean;
+}
+
+export function normalizeGrokCliInput(body) {
+  if (!Array.isArray(body?.input)) return body;
+  const normalized = body.input.map(normalizeGrokCliInputItem).filter(Boolean);
+  const callIds = new Set(
+    normalized
+      .filter((item) => item?.type === "function_call" && item.call_id)
+      .map((item) => item.call_id)
+  );
+  body.input = normalized.filter(
+    (item) => item?.type !== "function_call_output" || callIds.has(item.call_id)
+  );
+  return body;
+}
+
+function stripStoredItemReferences(body) {
+  if (!Array.isArray(body.input)) return;
+  body.input = body.input.filter((item) => {
+    if (typeof item === "string" && SERVER_ID_PATTERN.test(item)) return false;
+    if (item && typeof item === "object" && !Array.isArray(item)) {
+      if (item.type === "item_reference") return false;
+      if (
+        typeof item.id === "string" &&
+        SERVER_ID_PATTERN.test(item.id) &&
+        !isNativeGrokCliItemId(item.id)
+      ) delete item.id;
+    }
+    return true;
+  });
+}
+
+/**
+ * Flatten Chat Completions tool shape → Responses flat format.
+ * Keep hosted tools (web_search / x_search) passthrough.
+ */
+function normalizeGrokCliTools(body) {
+  if (!Array.isArray(body.tools) || body.tools.length === 0) {
+    delete body.tools;
+    delete body.tool_choice;
+    return;
+  }
+  const validNames = new Set();
+  const hostedTypes = new Set();
+  body.tools = body.tools.filter((tool) => {
+    if (!tool || typeof tool !== "object" || Array.isArray(tool)) return false;
+    const type = typeof tool.type === "string" ? tool.type : "";
+
+    if (type !== "function") {
+      // Hosted tools: { type: "web_search" } / { type: "x_search" }
+      if (HOSTED_TOOL_TYPES.has(type)) {
+        hostedTypes.add(type);
+        return true;
+      }
+      // Nested function shape without type
+      if (!type && tool.function) {
+        // fall through to function flatten below
+      } else if (!type || typeof tool.name === "string") {
+        // treat as bare function if name present
+      } else {
+        return false;
+      }
+    }
+
+    const isFunction =
+      type === "function" || type === "" || tool.function || typeof tool.name === "string";
+    if (!isFunction || HOSTED_TOOL_TYPES.has(type)) {
+      return HOSTED_TOOL_TYPES.has(type);
+    }
+
+    const fn =
+      tool.function && typeof tool.function === "object" && !Array.isArray(tool.function)
+        ? tool.function
+        : null;
+    const rawName =
+      typeof tool.name === "string" ? tool.name : typeof fn?.name === "string" ? fn.name : "";
+    const name = rawName.trim();
+    if (!name) return false;
+
+    const description =
+      typeof tool.description === "string"
+        ? tool.description
+        : typeof fn?.description === "string"
+          ? fn.description
+          : "";
+    const parameters = type === "custom"
+      ? GROK_CLI_FREEFORM_TOOL_PARAMETERS
+      : tool.parameters && typeof tool.parameters === "object" && !Array.isArray(tool.parameters)
+        ? tool.parameters
+        : fn?.parameters && typeof fn.parameters === "object" && !Array.isArray(fn.parameters)
+          ? fn.parameters
+          : { type: "object", properties: {} };
+
+    for (const k of Object.keys(tool)) delete tool[k];
+    tool.type = "function";
+    tool.name = name.slice(0, 128);
+    if (description) tool.description = description;
+    tool.parameters = parameters;
+    validNames.add(tool.name);
+    return true;
+  });
+
+  if (body.tools.length === 0) {
+    delete body.tools;
+    delete body.tool_choice;
+    return;
+  }
+
+  if (body.tool_choice && typeof body.tool_choice === "object" && !Array.isArray(body.tool_choice)) {
+    const choiceType = typeof body.tool_choice.type === "string" ? body.tool_choice.type : "";
+    if (choiceType === "function" || choiceType === "custom") {
+      const rawName = body.tool_choice.name ?? body.tool_choice.function?.name;
+      const name = typeof rawName === "string" ? rawName.trim().slice(0, 128) : "";
+      if (!name || !validNames.has(name)) delete body.tool_choice;
+      else body.tool_choice = { type: "function", name };
+    } else if (!hostedTypes.has(choiceType)) {
+      delete body.tool_choice;
+    }
+  }
+}
+
+function resolveEffortFromModel(modelId) {
+  if (!modelId || typeof modelId !== "string") return null;
+  for (const level of EFFORT_LEVELS) {
+    if (modelId.endsWith(`-${level}`)) return level;
+  }
+  return null;
+}
+
+/**
+ * Grok CLI Executor — OpenAI Responses API on cli-chat-proxy.grok.com
+ * Auth: OAuth device-code access token (xai-grok-cli).
+ */
+export class GrokCliExecutor extends BaseExecutor {
+  constructor() {
+    super("grok-cli", PROVIDERS["grok-cli"]);
+    this._currentSessionId = null;
+    this._currentReqId = null;
+    this._currentTurnIdx = 1;
+    this._agentId = null;
+  }
+
+  buildUrl() {
+    return this.config.baseUrl;
+  }
+
+  async refreshCredentials(credentials, log, proxyOptions = null) {
+    if (!credentials?.refreshToken) return null;
+    return refreshProviderCredentials("grok-cli", credentials, log, proxyOptions);
+  }
+
+  needsRefresh(credentials) {
+    return shouldRefreshCredentials("grok-cli", credentials);
+  }
+
+  buildHeaders(credentials, stream = true) {
+    const headers = super.buildHeaders(credentials, stream);
+
+    // Static fingerprint from registry
+    const staticHeaders = this.config.headers || {};
+    for (const [k, v] of Object.entries(staticHeaders)) {
+      if (v != null && headers[k] === undefined) headers[k] = v;
+    }
+
+    headers["x-grok-client-identifier"] =
+      this.config.clientIdentifier || headers["x-grok-client-identifier"] || GROK_CLI_CLIENT_IDENTIFIER;
+    headers["x-grok-client-version"] =
+      this.config.clientVersion || headers["x-grok-client-version"] || GROK_CLI_VERSION;
+
+    const sessionId = this._currentSessionId || credentials?.connectionId || crypto.randomUUID();
+    const reqId = this._currentReqId || crypto.randomUUID();
+    headers["x-grok-session-id"] = sessionId;
+    // CLI uses the same id for conv + session on chat turns
+    headers["x-grok-conv-id"] = sessionId;
+    headers["x-grok-req-id"] = reqId;
+    headers["x-grok-turn-idx"] = String(this._currentTurnIdx || 1);
+
+    if (this._agentId) headers["x-grok-agent-id"] = this._agentId;
+
+    // Surface model override (CLI always sets this)
+    if (this._currentModel) headers["x-grok-model-override"] = this._currentModel;
+
+    // Identity: mapTokens stores email top-level AND in providerSpecificData;
+    // fall back either way so OAuth connections always fingerprint like the CLI.
+    const psd = credentials?.providerSpecificData || {};
+    const email = psd.email || credentials?.email;
+    const userId = psd.userId || credentials?.userId || credentials?.providerUserId;
+    if (email) headers["x-email"] = email;
+    if (userId) headers["x-userid"] = userId;
+
+    return headers;
+  }
+
+  parseError(response, bodyText) {
+    // 402 personal-team-blocked:spending-limit → surface as payment/quota for fallback
+    if (response.status === 402 && bodyText) {
+      try {
+        const json = JSON.parse(bodyText);
+        const code = json?.code || "";
+        const msg = json?.error || json?.message || bodyText;
+        return {
+          status: 402,
+          message: typeof msg === "string" ? msg : bodyText,
+          code: typeof code === "string" ? code : undefined,
+        };
+      } catch {
+        /* fall through */
+      }
+    }
+    return super.parseError(response, bodyText);
+  }
+
+  transformRequest(model, body, stream, credentials) {
+    // Session / request ids for headers — stable per client conversation when possible
+    const requestKey = body;
+    this._currentSessionId = resolveGrokCliSessionId(credentials, body);
+    this._currentReqId = crypto.randomUUID();
+    this._agentId =
+      credentials?.providerSpecificData?.deviceId ||
+      credentials?.providerSpecificData?.agentId ||
+      null;
+
+    // Normalize Responses input
+    const normalized = normalizeResponsesInput(body.input);
+    if (normalized) body.input = normalized;
+
+    // Chat Completions clients arrive with messages[] — translator should have
+    // converted already, but guard empty input.
+    if (!body.input || (Array.isArray(body.input) && body.input.length === 0)) {
+      if (Array.isArray(body.messages) && body.messages.length > 0) {
+        // Soft fallback: map messages → input messages (string content only)
+        body.input = body.messages.map((m) => ({
+          type: "message",
+          role: m.role || "user",
+          content: typeof m.content === "string" ? m.content : JSON.stringify(m.content ?? ""),
+        }));
+        delete body.messages;
+      } else {
+        body.input = [{ type: "message", role: "user", content: "..." }];
+      }
+    }
+
+    // Keep role:"system" as-is — official grok-pager HAR sends system, not developer
+    // (Codex converts system→developer; Grok CLI does not).
+    normalizeGrokCliInput(body);
+    stripStoredItemReferences(body);
+    normalizeGrokCliTools(body);
+
+    // Turn index after input is finalized (user-message count, monotonic per session)
+    this._currentTurnIdx = resolveGrokCliTurnIdx(this._currentSessionId, body.input, requestKey);
+
+    body.stream = true;
+    body.store = false;
+
+    // Resolve upstream model id (strip effort suffix virtual models)
+    let modelEffort = resolveEffortFromModel(body.model || model);
+    let resolvedModel = body.model || model;
+    if (modelEffort) {
+      resolvedModel = resolvedModel.replace(new RegExp(`-${modelEffort}$`), "");
+    }
+    resolvedModel = getModelUpstreamId("gcli", resolvedModel) || resolvedModel;
+    // Also try provider id key
+    if (resolvedModel === (body.model || model)) {
+      resolvedModel = getModelUpstreamId("grok-cli", resolvedModel) || resolvedModel;
+    }
+    body.model = resolvedModel;
+    this._currentModel = resolvedModel;
+
+    // Reasoning effort priority: explicit > reasoning_effort > model suffix > default high.
+    // grok-build and Composer reject reasoningEffort but still accept summary/encrypted continuity.
+    const supportsReasoningEffort = supportsGrokCliReasoningEffort(resolvedModel);
+    if (!body.reasoning || typeof body.reasoning !== "object") {
+      body.reasoning = { summary: "concise" };
+      if (supportsReasoningEffort) {
+        body.reasoning.effort = normalizeGrokCliEffort(body.reasoning_effort || modelEffort);
+      }
+    } else {
+      if (supportsReasoningEffort) {
+        body.reasoning.effort = normalizeGrokCliEffort(
+          body.reasoning.effort || body.reasoning_effort || modelEffort,
+        );
+      } else {
+        delete body.reasoning.effort;
+      }
+      if (!body.reasoning.summary) body.reasoning.summary = "concise";
+    }
+    delete body.reasoning_effort;
+
+    // Encrypted reasoning for multi-turn continuity (CLI always requests this)
+    if (body.reasoning && body.reasoning.effort !== "none") {
+      const include = Array.isArray(body.include) ? body.include : [];
+      if (!include.includes("reasoning.encrypted_content")) {
+        include.push("reasoning.encrypted_content");
+      }
+      body.include = include;
+    }
+
+    // Drop Chat Completions leftovers that Responses rejects
+    delete body.messages;
+    delete body.max_tokens;
+    delete body.max_completion_tokens;
+    delete body.n;
+    delete body.seed;
+    delete body.logprobs;
+    delete body.top_logprobs;
+    delete body.frequency_penalty;
+    delete body.presence_penalty;
+    delete body.logit_bias;
+    delete body.user;
+    delete body.stream_options;
+    delete body.prompt_cache_retention;
+    delete body.safety_identifier;
+    delete body.previous_response_id; // store=false → cannot resolve
+
+    for (const k of Object.keys(body)) {
+      if (!RESPONSES_API_ALLOWLIST.has(k)) delete body[k];
+    }
+
+    return body;
+  }
+
+  async execute(args) {
+    // Lazy-resolve stable agent id once per process if connection has none
+    if (!this._agentId && !args.credentials?.providerSpecificData?.deviceId) {
+      try {
+        const mid = await getConsistentMachineId("grok-cli-agent");
+        // Format as UUID-ish for header aesthetics
+        this._agentId = [
+          mid.slice(0, 8),
+          mid.slice(8, 12),
+          "5" + mid.slice(13, 16),
+          "a" + mid.slice(17, 20),
+          mid.slice(0, 12).padEnd(12, "0"),
+        ].join("-");
+      } catch {
+        this._agentId = crypto.randomUUID();
+      }
+    } else if (args.credentials?.providerSpecificData?.deviceId) {
+      this._agentId = args.credentials.providerSpecificData.deviceId;
+    }
+
+    return super.execute(args);
+  }
+}
+
+export default GrokCliExecutor;
--- a/open-sse/executors/index.js
+++ b/open-sse/executors/index.js
@@ -5,6 +5,7 @@ import { GithubExecutor } from "./github.js";
 import { IFlowExecutor } from "./iflow.js";
 import { QoderExecutor } from "./qoder.js";
 import { KiroExecutor } from "./kiro.js";
+import { KimchiExecutor } from "./kimchi.js";
 import { CodexExecutor } from "./codex.js";
 import { CursorExecutor } from "./cursor.js";
 import { VertexExecutor } from "./vertex.js";
@@ -12,13 +13,19 @@ import { QwenExecutor } from "./qwen.js";
 import { OpenCodeExecutor } from "./opencode.js";
 import { OpenCodeGoExecutor } from "./opencode-go.js";
 import { GrokWebExecutor } from "./grok-web.js";
+import { GrokCliExecutor } from "./grok-cli.js";
 import { PerplexityWebExecutor } from "./perplexity-web.js";
 import { OllamaLocalExecutor } from "./ollama-local.js";
 import { CommandCodeExecutor } from "./commandcode.js";
 import { XiaomiTokenplanExecutor } from "./xiaomi-tokenplan.js";
 import { MimoFreeExecutor } from "./mimo-free.js";
 import { CodeBuddyExecutor } from "./codebuddy-cn.js";
+import { CodeBuddyIntlExecutor } from "./codebuddy-intl.js";
+import TraeExecutor from "./trae.js";
+import ZedExecutor from "./zed.js";
+import WindsurfExecutor from "./windsurf.js";
 import { DefaultExecutor } from "./default.js";
+import { DevinCliExecutor } from "./devin-cli.js";

 const executors = {
  antigravity: new AntigravityExecutor(),
@@ -28,6 +35,7 @@ const executors = {
  iflow: new IFlowExecutor(),
  qoder: new QoderExecutor(),
  kiro: new KiroExecutor(),
+  kimchi: new KimchiExecutor(),
  codex: new CodexExecutor(),
  cursor: new CursorExecutor(),
  cu: new CursorExecutor(), // Alias for cursor
@@ -37,6 +45,9 @@ const executors = {
  opencode: new OpenCodeExecutor(),
  "opencode-go": new OpenCodeGoExecutor(),
  "grok-web": new GrokWebExecutor(),
+  "grok-cli": new GrokCliExecutor(),
+  gcli: new GrokCliExecutor(), // Alias
+  gb: new GrokCliExecutor(), // Alias (Grok Build)
  "perplexity-web": new PerplexityWebExecutor(),
  "ollama-local": new OllamaLocalExecutor(),
  commandcode: new CommandCodeExecutor(),
@@ -44,6 +55,11 @@ const executors = {
  "mimo-free": new MimoFreeExecutor(),
  mmf: new MimoFreeExecutor(), // Alias for mimo-free
  "codebuddy-cn": new CodeBuddyExecutor(),
+  "codebuddy-intl": new CodeBuddyIntlExecutor(),
+  trae: new TraeExecutor(),
+  zed: new ZedExecutor(),
+  windsurf: new WindsurfExecutor(),
+  "devin-cli": new DevinCliExecutor(),
 };

 const defaultCache = new Map();
@@ -66,6 +82,7 @@ export { GithubExecutor } from "./github.js";
 export { IFlowExecutor } from "./iflow.js";
 export { QoderExecutor } from "./qoder.js";
 export { KiroExecutor } from "./kiro.js";
+export { KimchiExecutor } from "./kimchi.js";
 export { CodexExecutor } from "./codex.js";
 export { CursorExecutor } from "./cursor.js";
 export { VertexExecutor } from "./vertex.js";
@@ -74,9 +91,15 @@ export { QwenExecutor } from "./qwen.js";
 export { OpenCodeExecutor } from "./opencode.js";
 export { OpenCodeGoExecutor } from "./opencode-go.js";
 export { GrokWebExecutor } from "./grok-web.js";
+export { GrokCliExecutor } from "./grok-cli.js";
 export { PerplexityWebExecutor } from "./perplexity-web.js";
 export { OllamaLocalExecutor } from "./ollama-local.js";
 export { CommandCodeExecutor } from "./commandcode.js";
 export { XiaomiTokenplanExecutor } from "./xiaomi-tokenplan.js";
 export { MimoFreeExecutor } from "./mimo-free.js";
 export { CodeBuddyExecutor } from "./codebuddy-cn.js";
+export { CodeBuddyIntlExecutor } from "./codebuddy-intl.js";
+export { default as TraeExecutor } from "./trae.js";
+export { default as ZedExecutor } from "./zed.js";
+export { default as WindsurfExecutor } from "./windsurf.js";
+export { DevinCliExecutor } from "./devin-cli.js";
--- a/open-sse/executors/kimchi.js
+++ b/open-sse/executors/kimchi.js
@@ -0,0 +1,123 @@
+import { DefaultExecutor } from "./default.js";
+import { getCachedKimchiModelMetadata } from "../services/kimchiModels.js";
+
+const TOP_LEVEL_OPENAI_GATEWAY_DROPS = [
+  "anthropic_version",
+  "anthropic_beta",
+  "client_metadata",
+  "mcp_servers",
+  "stop_sequences",
+  "thinking",
+  "top_k",
+];
+
+function systemToText(system) {
+  if (typeof system === "string") return system;
+  if (Array.isArray(system)) {
+    return system
+      .map((part) => {
+        if (typeof part === "string") return part;
+        if (typeof part?.text === "string") return part.text;
+        return "";
+      })
+      .filter(Boolean)
+      .join("\n");
+  }
+  return "";
+}
+
+function mergeTopLevelSystem(body) {
+  if (!body?.system || !Array.isArray(body.messages)) return;
+  const text = systemToText(body.system).trim();
+  if (!text) return;
+
+  const existing = body.messages.find((msg) => msg?.role === "system");
+  if (!existing) {
+    body.messages.unshift({ role: "system", content: text });
+    return;
+  }
+
+  if (typeof existing.content === "string") {
+    existing.content = `${text}\n\n${existing.content}`;
+  } else if (Array.isArray(existing.content)) {
+    existing.content.unshift({ type: "text", text });
+  }
+}
+
+function stripMessageArtifacts(body) {
+  if (!Array.isArray(body?.messages)) return;
+  for (const msg of body.messages) {
+    if (!msg || typeof msg !== "object") continue;
+    delete msg.cache_control;
+    if (!Array.isArray(msg.content)) continue;
+    msg.content = msg.content.map((part) => {
+      if (!part || typeof part !== "object") return part;
+      const { cache_control, signature, ...clean } = part;
+      return clean;
+    });
+  }
+}
+
+function stripToolArtifacts(body) {
+  if (!Array.isArray(body?.tools)) return;
+  body.tools = body.tools.map((tool) => {
+    if (!tool || typeof tool !== "object") return tool;
+    const { cache_control, ...clean } = tool;
+    return clean;
+  });
+}
+
+// Strip `reasoning_content` echoed by clients on assistant messages — but
+// only when it's a real thinking block. `DefaultExecutor.transformRequest`
+// runs `injectReasoningContent` first and may inject a 1-char placeholder
+// (" ") for upstream validation; the placeholder is small (no token cost
+// worth stripping) and stripping it would re-trigger upstream to complain
+// about missing reasoning on the next turn. Threshold matches the
+// placeholder length with a safety margin.
+const REASONING_PLACEHOLDER_MAX_LEN = 8;
+
+export function stripReasoningContent(body) {
+  if (!Array.isArray(body?.messages)) return;
+  for (const msg of body.messages) {
+    if (msg && msg.role === "assistant" && typeof msg.reasoning_content === "string"
+        && msg.reasoning_content.length > REASONING_PLACEHOLDER_MAX_LEN) {
+      delete msg.reasoning_content;
+    }
+  }
+}
+
+function isAnthropicBackedKimchiModel(model) {
+  const meta = getCachedKimchiModelMetadata(model);
+  if (meta?.provider === "anthropic" || meta?.upstreamProvider === "anthropic") return true;
+  return /(^|[-_/])(?:claude|anthropic)(?:[-_/]|$)/i.test(String(model || ""));
+}
+
+export class KimchiExecutor extends DefaultExecutor {
+  constructor() {
+    super("kimchi");
+  }
+
+  transformRequest(model, body, stream, credentials) {
+    const transformed = super.transformRequest(model, body, stream, credentials);
+    if (!transformed || typeof transformed !== "object") return transformed;
+
+    mergeTopLevelSystem(transformed);
+    for (const key of TOP_LEVEL_OPENAI_GATEWAY_DROPS) {
+      if (transformed[key] !== undefined) delete transformed[key];
+    }
+    delete transformed.system;
+
+    if (isAnthropicBackedKimchiModel(model)) {
+      delete transformed.reasoning_effort;
+      delete transformed.reasoning;
+      delete transformed.thinking;
+    }
+
+    stripMessageArtifacts(transformed);
+    stripToolArtifacts(transformed);
+    stripReasoningContent(transformed);
+    return transformed;
+  }
+}
+
+export default KimchiExecutor;
--- a/open-sse/executors/kiro.js
+++ b/open-sse/executors/kiro.js
--- a/open-sse/executors/qoder.js
+++ b/open-sse/executors/qoder.js
@@ -30,9 +30,14 @@ import { PROVIDERS } from "../config/providers.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
 import { SSE_DONE } from "../utils/sseConstants.js";
 import { FETCH_CONNECT_TIMEOUT_MS } from "../config/runtimeConfig.js";
+import { resolveProviderTimeoutMs } from "../services/providerTimeout.js";
 import {
  QODER_CHAT_URL_ENCODED,
+  QODER_JOB_TOKEN_EXCHANGE_URL,
+  QODER_USERINFO_URL,
  QODER_MODEL_MAP,
+  QODER_IDE_VERSION,
+  QODER_CLIENT_TYPE,
 } from "../shared/qoder/constants.js";
 import { getQoderModelConfig, resolveQoderModels } from "../services/qoderModels.js";

@@ -220,8 +225,13 @@ async function buildQoderRequestBody({ model, body, credentials, log, proxyOptio
 * Each upstream line looks like:
 *   data: {"statusCodeValue":200,"body":"{\"choices\":[{\"delta\":{...}}]}"}
 * The inner body is an OpenAI streaming chunk (or "[DONE]"). We unwrap it
- * and re-emit as `data: <inner>\n\n`. Errors become `data: [DONE]\n\n` plus
- * a synthetic OpenAI error chunk.
+ * and re-emit as `data: <inner>\n\n`. Errors become a synthetic OpenAI error
+ * chunk + [DONE].
+ *
+ * Critical: Qoder's SSE often keeps the socket open after the terminal
+ * [DONE]/error frame (agent keepalive). Non-streaming clients drain via
+ * response.text() which hangs until the socket closes — so on terminal
+ * events we cancel the upstream reader and close our stream immediately.
 */
 function wrapQoderSSE(response, model) {
  if (!response.ok || !response.body) return response;
@@ -230,15 +240,14 @@ function wrapQoderSSE(response, model) {
  const encoder = new TextEncoder();
  let buffer = "";
  let doneEmitted = false;
+  const reader = response.body.getReader();

-  // Process one already-extracted SSE line (no trailing newline). Returns
-  // false when the line indicated end-of-stream so the caller can stop
-  // forwarding any remaining chunks after [DONE].
+  // Process one already-extracted SSE line (no trailing newline).
  const processLine = (line, controller) => {
    const trimmed = line.replace(/\r$/, "").trim();
    if (!trimmed) return;
    if (!trimmed.startsWith("data:")) return;
-    if (doneEmitted) return; // never forward chunks past stream end
+    if (doneEmitted) return;

    const data = trimmed.slice(5).trimStart();
    if (data === "[DONE]") {
@@ -271,47 +280,60 @@ function wrapQoderSSE(response, model) {
      doneEmitted = true;
      return;
    }
-    // Inner is an OpenAI-shaped chunk. Strip any embedded newlines so the
-    // SSE frame stays a single event (a literal "\n" inside `inner` would
-    // otherwise split the frame across multiple data: lines and downstream
-    // parsers would reassemble them as separate events).
+    // Strip embedded newlines so the SSE frame stays a single event.
    const sanitized = inner.replace(/\r?\n/g, "");
    controller.enqueue(encoder.encode(`data: ${sanitized}\n\n`));
  };

-  const transform = new TransformStream({
-    transform(chunk, controller) {
-      buffer += decoder.decode(chunk, { stream: true });
-      let nl;
-      while ((nl = buffer.indexOf("\n")) !== -1) {
-        const line = buffer.slice(0, nl);
-        buffer = buffer.slice(nl + 1);
-        processLine(line, controller);
+  const stream = new ReadableStream({
+    // Use start()+loop (not pull): a pull that buffers a partial line without
+    // enqueueing would never be re-invoked, hanging consumers like .text().
+    async start(controller) {
+      try {
+        while (!doneEmitted) {
+          const { done, value } = await reader.read();
+          if (done) {
+            buffer += decoder.decode();
+            if (buffer.length > 0) {
+              processLine(buffer, controller);
+              buffer = "";
+            }
+            break;
+          }
+
+          buffer += decoder.decode(value, { stream: true });
+          let nl;
+          while ((nl = buffer.indexOf("\n")) !== -1) {
+            const line = buffer.slice(0, nl);
+            buffer = buffer.slice(nl + 1);
+            processLine(line, controller);
+            if (doneEmitted) {
+              // Terminal frame received — drop upstream keepalive and end.
+              await reader.cancel().catch(() => {});
+              controller.close();
+              return;
+            }
+          }
+        }
+      } catch {
+        // fall through to terminal [DONE] + close
+      } finally {
+        if (!doneEmitted) {
+          try {
+            controller.enqueue(encoder.encode(SSE_DONE));
+            doneEmitted = true;
+          } catch { /* already closed */ }
+        }
+        try { controller.close(); } catch { /* already closed */ }
+        await reader.cancel().catch(() => {});
      }
    },
-    flush(controller) {
-      // Finalize the decoder so any pending multi-byte sequence is
-      // released into `buffer` instead of being silently dropped.
-      buffer += decoder.decode();
-      // Drain any trailing line that arrived without a terminating newline
-      // (e.g. upstream closed the socket immediately after the last write,
-      // or a CDN stripped the final CRLF). Without this, the chunk that
-      // carries finish_reason is silently lost.
-      if (buffer.length > 0) {
-        processLine(buffer, controller);
-        buffer = "";
-      }
-      if (!doneEmitted) {
-        controller.enqueue(encoder.encode(SSE_DONE));
-        doneEmitted = true;
-      }
+    cancel() {
+      return reader.cancel().catch(() => {});
    },
  });

-  const transformed = response.body.pipeThrough(transform);
-  // Build a Response with passable headers; the streaming handler reads
-  // `.body` as a ReadableStream regardless of Content-Type.
-  return new Response(transformed, {
+  return new Response(stream, {
    status: response.status,
    statusText: response.statusText,
    headers: {
@@ -321,6 +343,92 @@ function wrapQoderSSE(response, model) {
  });
 }

+// ── PAT (Personal Access Token) → job-token exchange ───────────────────────
+// PATs (pt-...) cannot sign COSY requests directly. Exchange them for a
+// short-lived job token (jt-...) via /api/v1/jobToken/exchange (plain JSON,
+// not COSY-signed), then resolve the userId from userinfo. Mirrors the
+// official qodercli flow. Cached per-PAT until near-expiry.
+const PAT_PREFIX = "pt-";
+const PAT_REFRESH_BUFFER_MS = 5 * 60 * 1000;
+const patJobCache = new Map();
+
+export function isQoderPat(token) {
+  return typeof token === "string" && token.startsWith(PAT_PREFIX);
+}
+
+async function exchangeJobToken(pat, proxyOptions = null, signal = null) {
+  const res = await proxyAwareFetch(
+    QODER_JOB_TOKEN_EXCHANGE_URL,
+    {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        Accept: "application/json",
+        "User-Agent": "qodercli/1.0.0",
+        "Cosy-Version": QODER_IDE_VERSION,
+        "Cosy-ClientType": QODER_CLIENT_TYPE,
+      },
+      body: JSON.stringify({ personal_token: pat }),
+      signal,
+    },
+    proxyOptions,
+  );
+  if (!res.ok) {
+    const text = await res.text().catch(() => "");
+    throw new Error(`qoder PAT exchange failed: ${res.status} ${text.slice(0, 200)}`);
+  }
+  const data = await res.json();
+  if (!data.token) throw new Error("qoder PAT exchange returned no job token");
+
+  let expiresAt = Date.now() + 24 * 60 * 60 * 1000;
+  if (data.expires_at) {
+    const parsed = Date.parse(data.expires_at);
+    if (!Number.isNaN(parsed)) expiresAt = parsed;
+  } else if (typeof data.expires_in === "number" && data.expires_in > 0) {
+    expiresAt = Date.now() + data.expires_in;
+  }
+  return { jobToken: data.token, jobRefreshToken: data.refresh_token || "", expiresAt };
+}
+
+async function fetchUserIdForJobToken(jobToken, proxyOptions = null, signal = null) {
+  try {
+    const res = await proxyAwareFetch(
+      QODER_USERINFO_URL,
+      {
+        method: "GET",
+        headers: {
+          Authorization: `Bearer ${jobToken}`,
+          Accept: "application/json",
+          "User-Agent": "qodercli/1.0.0",
+        },
+        signal,
+      },
+      proxyOptions,
+    );
+    if (!res.ok) return "";
+    const info = await res.json().catch(() => ({}));
+    return info.id || info.userId || info.user_id || "";
+  } catch {
+    return "";
+  }
+}
+
+/**
+ * Exchange a PAT for a job token + userId, caching until near-expiry so repeat
+ * chat requests don't re-exchange. Returns { accessToken, userId }.
+ */
+async function resolvePatCredential(pat, proxyOptions = null, signal = null) {
+  const cached = patJobCache.get(pat);
+  if (cached && cached.expiresAt - Date.now() > PAT_REFRESH_BUFFER_MS) {
+    return cached;
+  }
+  const { jobToken, expiresAt } = await exchangeJobToken(pat, proxyOptions, signal);
+  const userId = await fetchUserIdForJobToken(jobToken, proxyOptions, signal);
+  const entry = { accessToken: jobToken, userId, expiresAt };
+  patJobCache.set(pat, entry);
+  return entry;
+}
+
 export class QoderExecutor extends BaseExecutor {
  constructor() {
    super("qoder", PROVIDERS.qoder);
@@ -338,6 +446,34 @@ export class QoderExecutor extends BaseExecutor {
  async execute({ model, body, stream, credentials, signal, log, proxyOptions = null }) {
    const url = this.buildUrl();

+    // PAT (pt-...) → exchange for short-lived job token + resolve userId so
+    // downstream COSY signing + catalog fetch work. Device tokens (dt-...) and
+    // job tokens (jt-...) skip this and are used directly.
+    const rawToken = credentials?.apiKey || credentials?.accessToken;
+    if (isQoderPat(rawToken)) {
+      try {
+        const resolved = await resolvePatCredential(rawToken, proxyOptions, signal);
+        credentials = {
+          ...credentials,
+          accessToken: resolved.accessToken,
+          apiKey: undefined,
+          providerSpecificData: {
+            authMethod: "pat",
+            ...(credentials?.providerSpecificData || {}),
+            userId: resolved.userId || credentials?.providerSpecificData?.userId || "",
+            machineId: credentials?.providerSpecificData?.machineId || "",
+          },
+        };
+      } catch (err) {
+        log?.error?.("QODER", `PAT exchange failed: ${err.message}`);
+        const fakeResp = new Response(
+          JSON.stringify({ error: { message: `qoder PAT exchange failed: ${err.message}` } }),
+          { status: 401, headers: { "Content-Type": "application/json" } },
+        );
+        return { response: fakeResp, url, headers: {}, transformedBody: body };
+      }
+    }
+
    const psd = credentials?.providerSpecificData || {};
    if (!psd.userId) {
      // No user id → no way to sign. Surface a 401 so the dashboard nudges
@@ -410,7 +546,7 @@ export class QoderExecutor extends BaseExecutor {
    };

    // Abort if upstream doesn't return response headers within connect timeout.
-    const timeoutMs = this.config?.timeoutMs || FETCH_CONNECT_TIMEOUT_MS;
+    const timeoutMs = await resolveProviderTimeoutMs(this.provider, this.config?.timeoutMs, FETCH_CONNECT_TIMEOUT_MS);
    const connectCtrl = new AbortController();
    const connectTimer = setTimeout(() => connectCtrl.abort(new Error("fetch connect timeout")), timeoutMs);
    const mergedSignal = signal ? AbortSignal.any([signal, connectCtrl.signal]) : connectCtrl.signal;
@@ -455,4 +591,6 @@ export const __test__ = {
  normalizeMessages,
  wrapQoderSSE,
  buildQoderRequestBody,
+  isQoderPat,
+  resolvePatCredential,
 };
--- a/open-sse/executors/trae.js
+++ b/open-sse/executors/trae.js
@@ -0,0 +1,339 @@
+import { BaseExecutor } from "./base.js";
+import { proxyAwareFetch } from "../utils/proxyFetch.js";
+import { PROVIDERS } from "../config/providers.js";
+
+// Trae executor — SOLO remote agent API.
+//
+// Flow:
+//   1. POST {base}/chat_sessions          → { code:0, data:{ chat_session_id, message_id } }
+//   2. GET  {base}/chat_sessions/{id}/events?reply_to_message_id={message_id}
+//        → text/event-stream. Assistant text streams in `plan_item` events under
+//          the `thought` field (cumulative per plan-item id). `token_usage` carries
+//          usage; `done` ends the turn; `error` carries upstream errors.
+//
+// Auth: header `Authorization: Cloud-IDE-JWT <jwt>` (RS256, ~14-day lifetime).
+// Identity fields for common_params live in credentials.providerSpecificData.
+
+const STREAM_TIMEOUT_MS = parseInt(process.env.TRAE_STREAM_TIMEOUT_MS || "300000", 10);
+const TRAE_UA =
+  "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 " +
+  "(KHTML, like Gecko) Chrome/149.0.0.0 Safari/537.36";
+
+function flattenQuery(messages) {
+  const parts = [];
+  for (const m of messages) {
+    let content = "";
+    if (typeof m.content === "string") content = m.content;
+    else if (Array.isArray(m.content)) {
+      content = m.content
+        .map((p) => {
+          if (typeof p === "string") return p;
+          if (p && typeof p === "object") return String(p.text ?? "");
+          return "";
+        })
+        .join("");
+    }
+    if (m.role === "system") parts.push(`[System]\n${content}`);
+    else if (m.role === "assistant") parts.push(`[Assistant]\n${content}`);
+    else parts.push(content);
+  }
+  // Trae expects query as a JSON-encoded string of typed content blocks.
+  return JSON.stringify([{ type: "text", data: { content: parts.join("\n\n") } }]);
+}
+
+export default class TraeExecutor extends BaseExecutor {
+  constructor() {
+    super("trae", PROVIDERS.trae);
+  }
+
+  base() {
+    return (this.config.baseUrl || "https://core-normal.trae.ai/api/remote/v1").replace(/\/$/, "");
+  }
+
+  buildHeaders(credentials, stream = true) {
+    const token = credentials?.accessToken || "";
+    const psd = credentials?.providerSpecificData || {};
+    return {
+      Authorization: `Cloud-IDE-JWT ${token}`,
+      "Content-Type": "application/json",
+      "X-Trae-Client-Type": "web",
+      "X-Preferenced-Language": psd.appLanguage || "en",
+      "x-user-region": psd.userRegion || "US",
+      Referer: "https://solo.trae.ai/",
+      "User-Agent": TRAE_UA,
+      Accept: stream ? "text/event-stream" : "application/json",
+    };
+  }
+
+  // SOLO session modes: "code" (model picker) vs "work" (fast auto lane).
+  resolveMode(model) {
+    const m = (model || "").trim().toLowerCase();
+    if (m === "work" || m === "auto-work" || m === "solo-work") {
+      return { mode: "work", strategy: "auto", modelName: "" };
+    }
+    const auto = !m || m === "auto";
+    return { mode: "code", strategy: auto ? "auto" : "manual", modelName: auto ? "" : model };
+  }
+
+  // common_params is a JSON-encoded string embedded inside initial_message.
+  commonParams(psd, mode, sessionId) {
+    const cp = {
+      language: "en-us",
+      app_language: psd.appLanguage || "en",
+      quality: "stable",
+      app_version: psd.appVersion || "1.0.0.1229",
+      web_id: psd.webId || "",
+      user_identity: psd.userIdentity || "Free",
+      is_freshman: "0",
+      biz_user_id: psd.bizUserId || "",
+      user_unique_id: psd.userUniqueId || "",
+      scope: psd.scope || "marscode-us",
+      tenant: psd.tenant || "marscode",
+      region: psd.region || "US-East",
+      aiRegion: psd.aiRegion || psd.region || "US-East",
+      is_privacy_mode: 0,
+      privacy_mode: "off",
+      solo_chat_mode: mode,
+    };
+    if (sessionId) cp.biz_session_id = sessionId;
+    return JSON.stringify(cp);
+  }
+
+  // POST /chat_sessions — creates a session and submits the first turn.
+  async createSession(headers, query, model, psd, signal) {
+    const { mode, strategy, modelName } = this.resolveMode(model);
+    const body = {
+      mode,
+      environment_id: "default",
+      initial_message: {
+        chat_session_id: "",
+        content: [],
+        query,
+        model_name: modelName,
+        agent_type: "solo_agent_remote",
+        model_selection_strategy: strategy,
+        common_params: this.commonParams(psd, mode),
+      },
+      env: "remote",
+      auto_create_project: false,
+      origin: "web",
+    };
+    const res = await proxyAwareFetch(`${this.base()}/chat_sessions`, {
+      method: "POST",
+      headers,
+      body: JSON.stringify(body),
+      signal,
+    }, null);
+    const text = await res.text();
+    if (!res.ok) throw new Error(`[${res.status}] ${text}`);
+    const json = JSON.parse(text);
+    if (json?.code !== 0) throw new Error(`Trae create_session: ${JSON.stringify(json)}`);
+    return { sessionId: json.data.chat_session_id, messageId: json.data.message_id };
+  }
+
+  // GET /events SSE → invoke onEvent(eventType, dataObj) per frame.
+  // Resolves when `done`/`error` arrives, the stream ends, or timeout fires.
+  async streamEvents(headers, sessionId, replyTo, onEvent, signal) {
+    const url = `${this.base()}/chat_sessions/${sessionId}/events?reply_to_message_id=${encodeURIComponent(replyTo)}`;
+    const ctrl = new AbortController();
+    if (signal?.aborted) ctrl.abort();
+    const timer = setTimeout(() => ctrl.abort(new Error("trae stream timeout")), STREAM_TIMEOUT_MS);
+    const onAbort = () => ctrl.abort();
+    if (signal) signal.addEventListener("abort", onAbort, { once: true });
+    try {
+      const res = await proxyAwareFetch(url, { method: "GET", headers, signal: ctrl.signal }, null);
+      if (!res.ok || !res.body) throw new Error(`[${res.status}] events stream failed`);
+      const reader = res.body.getReader();
+      const decoder = new TextDecoder();
+      let buf = "";
+      let ev = null;
+      for (;;) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        buf += decoder.decode(value, { stream: true });
+        let nl;
+        while ((nl = buf.indexOf("\n")) >= 0) {
+          const line = buf.slice(0, nl).replace(/\r$/, "");
+          buf = buf.slice(nl + 1);
+          if (line.startsWith("event:")) ev = line.slice(6).trim();
+          else if (line.startsWith("data:")) {
+            const payload = line.slice(5).trim();
+            let data;
+            try { data = JSON.parse(payload); } catch { data = { _raw: payload }; }
+            if (onEvent(ev, data)) {
+              await reader.cancel().catch(() => {});
+              return;
+            }
+          } else if (line === "") ev = null;
+        }
+      }
+    } finally {
+      clearTimeout(timer);
+      if (signal) signal.removeEventListener("abort", onAbort);
+    }
+  }
+
+  async execute({ model, body, stream, credentials, signal }) {
+    const headers = this.buildHeaders(credentials, stream !== false);
+    const psd = credentials?.providerSpecificData || {};
+    const query = flattenQuery(body?.messages || []);
+    const responseId = `chatcmpl-trae-${Date.now()}`;
+    const created = Math.floor(Date.now() / 1000);
+
+    const errResponse = (status, message) => new Response(
+      JSON.stringify({ error: { message, type: "api_error", code: "" } }),
+      { status, headers: { "Content-Type": "application/json" } }
+    );
+
+    let session;
+    try {
+      session = await this.createSession(headers, query, model, psd, signal);
+    } catch (err) {
+      return { response: errResponse(502, err?.message ? String(err.message) : String(err)), url: this.base(), headers, transformedBody: body };
+    }
+
+    // Shared per-turn state: plan_item thoughts (cumulative, longest wins).
+    const order = [];
+    const thoughts = {};
+    let sent = 0;
+    let usage = null;
+    let errorEvent = null;
+    const renderNewText = (data) => {
+      const pid = data.id;
+      if (!pid) return "";
+      if (!(pid in thoughts)) order.push(pid);
+      const t = data.thought || "";
+      if (t.length >= (thoughts[pid] || "").length) thoughts[pid] = t;
+      const full = order.map((i) => thoughts[i]).join("");
+      const piece = full.slice(sent);
+      sent = full.length;
+      return piece;
+    };
+
+    if (stream !== false) {
+      const enc = new TextEncoder();
+      const sse = new ReadableStream({
+        start: async (controller) => {
+          const emit = (obj) => controller.enqueue(enc.encode(`data: ${JSON.stringify(obj)}\n\n`));
+          emit({
+            id: responseId,
+            object: "chat.completion.chunk",
+            created,
+            model,
+            choices: [{ index: 0, delta: { role: "assistant" }, finish_reason: null }],
+          });
+          try {
+            await this.streamEvents(headers, session.sessionId, session.messageId, (ev, data) => {
+              if (ev === "error") { errorEvent = data; return true; }
+              if (ev === "token_usage") usage = data;
+              if (ev === "plan_item") {
+                const piece = renderNewText(data);
+                if (piece) {
+                  emit({
+                    id: responseId,
+                    object: "chat.completion.chunk",
+                    created,
+                    model,
+                    choices: [{ index: 0, delta: { content: piece }, finish_reason: null }],
+                  });
+                }
+              }
+              return ev === "done";
+            }, signal);
+            if (errorEvent) {
+              emit({
+                id: responseId,
+                object: "chat.completion.chunk",
+                created,
+                model,
+                choices: [],
+                error: { message: `trae ${errorEvent.code || ""}: ${errorEvent.message || ""}`, type: "api_error" },
+              });
+            } else {
+              emit({
+                id: responseId,
+                object: "chat.completion.chunk",
+                created,
+                model,
+                choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+              });
+              if (usage) {
+                emit({
+                  id: responseId,
+                  object: "chat.completion.chunk",
+                  created,
+                  model,
+                  choices: [],
+                  usage: {
+                    prompt_tokens: usage.prompt_tokens || 0,
+                    completion_tokens: usage.completion_tokens || 0,
+                    total_tokens: usage.total_tokens || 0,
+                  },
+                });
+              }
+            }
+            controller.enqueue(enc.encode("data: [DONE]\n\n"));
+            controller.close();
+          } catch (err) {
+            controller.error(err);
+          }
+        },
+      });
+      return {
+        response: new Response(sse, {
+          status: 200,
+          headers: {
+            "Content-Type": "text/event-stream",
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+          },
+        }),
+        url: this.base(),
+        headers,
+        transformedBody: body,
+      };
+    }
+
+    // Non-streaming: drive to completion, return chat.completion JSON.
+    try {
+      await this.streamEvents(headers, session.sessionId, session.messageId, (ev, data) => {
+        if (ev === "error") { errorEvent = data; return true; }
+        if (ev === "token_usage") usage = data;
+        if (ev === "plan_item") renderNewText(data);
+        return ev === "done";
+      }, signal);
+    } catch (err) {
+      return { response: errResponse(502, err?.message ? String(err.message) : String(err)), url: this.base(), headers, transformedBody: body };
+    }
+    if (errorEvent) {
+      return { response: errResponse(502, `trae ${errorEvent.code || ""}: ${errorEvent.message || ""}`), url: this.base(), headers, transformedBody: body };
+    }
+    const content = order.map((i) => thoughts[i]).join("");
+    const out = {
+      id: responseId,
+      object: "chat.completion",
+      created,
+      model,
+      choices: [{ index: 0, message: { role: "assistant", content }, finish_reason: "stop" }],
+    };
+    if (usage) {
+      out.usage = {
+        prompt_tokens: usage.prompt_tokens || 0,
+        completion_tokens: usage.completion_tokens || 0,
+        total_tokens: usage.total_tokens || 0,
+      };
+    }
+    return {
+      response: new Response(JSON.stringify(out), { status: 200, headers: { "Content-Type": "application/json" } }),
+      url: this.base(),
+      headers,
+      transformedBody: body,
+    };
+  }
+
+  // Refresh hook placeholder — Cloud-IDE-JWT is long-lived (~14d); refresh via
+  // ExchangeToken (refresh→access) is wired in services/tokenRefresh/providers.js.
+  async refreshCredentials() {
+    return null;
+  }
+}
--- a/open-sse/executors/windsurf.js
+++ b/open-sse/executors/windsurf.js
@@ -0,0 +1,588 @@
+import { BaseExecutor } from "./base.js";
+import { proxyAwareFetch } from "../utils/proxyFetch.js";
+import { PROVIDERS } from "../config/providers.js";
+import { randomUUID } from "node:crypto";
+
+// WindsurfExecutor — Codeium gRPC-web chat.
+//
+// Wire protocol: gRPC-web over HTTPS (Content-Type: application/grpc-web+proto).
+// Service:  exa.language_server_pb.LanguageServerService
+// Method:   GetChatMessage (unary request → streamed CompletionChunk frames)
+//
+// Auth: credentials.accessToken = Codeium apiKey (sk-ws-... or Firebase-derived)
+//       — placed in Metadata.api_key protobuf field of every request + Bearer header.
+
+const WS_BASE_URL = "https://server.codeium.com";
+const WS_SERVICE = "exa.language_server_pb.LanguageServerService";
+const WS_METHOD_CHAT = "GetChatMessage";
+const WS_CHAT_URL = `${WS_BASE_URL}/${WS_SERVICE}/${WS_METHOD_CHAT}`;
+
+const WS_IDE_NAME = "windsurf";
+const WS_IDE_VERSION = "3.14.0";
+const WS_EXT_VERSION = "3.14.0";
+const WS_LOCALE = "en-US";
+
+// ─── Model alias map (catalog name → Windsurf wire name) ─────────────────────
+const MODEL_ALIAS_MAP = {
+  // ── Cognition SWE ───────────────────────────────────────────────────────
+  "swe-1.6-fast": "swe-1-6-fast",
+  "swe-1.6": "swe-1-6",
+  "swe-1.5-fast": "swe-1-5-fast",
+  "swe-1.5": "swe-1-5",
+  // ── Claude Opus 4.7 — effort-tiered ─────────────────────────────────────
+  "claude-opus-4.7-max": "claude-opus-4-7-max",
+  "claude-opus-4.7-xhigh": "claude-opus-4-7-xhigh",
+  "claude-opus-4.7-high": "claude-opus-4-7-high",
+  "claude-opus-4.7-medium": "claude-opus-4-7-medium",
+  "claude-opus-4.7-low": "claude-opus-4-7-low",
+  "claude-opus-4.7-review": "opus-4-7-review",
+  // ── Claude Opus/Sonnet 4.6 ──────────────────────────────────────────────
+  "claude-sonnet-4.6-thinking-1m": "claude-sonnet-4-6-thinking-1m",
+  "claude-sonnet-4.6-1m": "claude-sonnet-4-6-1m",
+  "claude-sonnet-4.6-thinking": "claude-sonnet-4-6-thinking",
+  "claude-sonnet-4.6": "claude-sonnet-4-6",
+  "claude-opus-4.6-thinking": "claude-opus-4-6-thinking",
+  "claude-opus-4.6": "claude-opus-4-6",
+  // ── Claude 4.5 ──────────────────────────────────────────────────────────
+  "claude-opus-4.5-thinking": "MODEL_CLAUDE_4_5_OPUS_THINKING",
+  "claude-opus-4.5": "MODEL_CLAUDE_4_5_OPUS",
+  "claude-sonnet-4.5-thinking": "MODEL_PRIVATE_3",
+  "claude-sonnet-4.5": "MODEL_PRIVATE_2",
+  "claude-haiku-4.5": "MODEL_PRIVATE_11",
+  // ── GPT-5.5 ─────────────────────────────────────────────────────────────
+  "gpt-5.5-xhigh-fast": "gpt-5-5-xhigh-priority",
+  "gpt-5.5-high-fast": "gpt-5-5-high-priority",
+  "gpt-5.5-medium-fast": "gpt-5-5-medium-priority",
+  "gpt-5.5-low-fast": "gpt-5-5-low-priority",
+  "gpt-5.5-none-fast": "gpt-5-5-none-priority",
+  "gpt-5.5-xhigh": "gpt-5-5-xhigh",
+  "gpt-5.5-high": "gpt-5-5-high",
+  "gpt-5.5-medium": "gpt-5-5-medium",
+  "gpt-5.5-low": "gpt-5-5-low",
+  "gpt-5.5-none": "gpt-5-5-none",
+  "gpt-5.5-review": "gpt-5-5-review",
+  "gpt-5.5": "gpt-5-5-medium",
+  // ── GPT-5.4 ─────────────────────────────────────────────────────────────
+  "gpt-5.4-xhigh-fast": "gpt-5-4-xhigh-priority",
+  "gpt-5.4-high-fast": "gpt-5-4-high-priority",
+  "gpt-5.4-medium-fast": "gpt-5-4-medium-priority",
+  "gpt-5.4-low-fast": "gpt-5-4-low-priority",
+  "gpt-5.4-none-fast": "gpt-5-4-none-priority",
+  "gpt-5.4-xhigh": "gpt-5-4-xhigh",
+  "gpt-5.4-high": "gpt-5-4-high",
+  "gpt-5.4-medium": "gpt-5-4-medium",
+  "gpt-5.4-low": "gpt-5-4-low",
+  "gpt-5.4-none": "gpt-5-4-none",
+  "gpt-5.4-mini-xhigh": "gpt-5-4-mini-xhigh",
+  "gpt-5.4-mini-high": "gpt-5-4-mini-high",
+  "gpt-5.4-mini-medium": "gpt-5-4-mini-medium",
+  "gpt-5.4-mini-low": "gpt-5-4-mini-low",
+  "gpt-5.4": "gpt-5-4-medium",
+  // ── GPT-5.3-Codex ───────────────────────────────────────────────────────
+  "gpt-5.3-codex-xhigh-fast": "gpt-5-3-codex-xhigh-priority",
+  "gpt-5.3-codex-high-fast": "gpt-5-3-codex-high-priority",
+  "gpt-5.3-codex-medium-fast": "gpt-5-3-codex-medium-priority",
+  "gpt-5.3-codex-low-fast": "gpt-5-3-codex-low-priority",
+  "gpt-5.3-codex-xhigh": "gpt-5-3-codex-xhigh",
+  "gpt-5.3-codex-high": "gpt-5-3-codex-high",
+  "gpt-5.3-codex-medium": "gpt-5-3-codex-medium",
+  "gpt-5.3-codex-low": "gpt-5-3-codex-low",
+  "gpt-5.3-codex": "gpt-5-3-codex-medium",
+  // ── GPT-5.2 ─────────────────────────────────────────────────────────────
+  "gpt-5.2-xhigh": "MODEL_GPT_5_2_XHIGH",
+  "gpt-5.2-high": "MODEL_GPT_5_2_HIGH",
+  "gpt-5.2-medium": "MODEL_GPT_5_2_MEDIUM",
+  "gpt-5.2-low": "MODEL_GPT_5_2_LOW",
+  "gpt-5.2-none": "MODEL_GPT_5_2_NONE",
+  "gpt-5.2": "MODEL_GPT_5_2_MEDIUM",
+  // ── GPT-5 ───────────────────────────────────────────────────────────────
+  "gpt-5": "gpt-5",
+  // ── GPT-4.1 / 4o ────────────────────────────────────────────────────────
+  "gpt-4.1": "MODEL_CHAT_GPT_4_1_2025_04_14",
+  "gpt-4.1-mini": "gpt-4.1-mini",
+  "gpt-4o": "MODEL_CHAT_GPT_4O_2024_08_06",
+  // ── Gemini ──────────────────────────────────────────────────────────────
+  "gemini-3.1-pro-high": "gemini-3-1-pro-high",
+  "gemini-3.1-pro-low": "gemini-3-1-pro-low",
+  "gemini-3.1-pro": "gemini-3-1-pro-high",
+  "gemini-3.0-flash-high": "MODEL_GOOGLE_GEMINI_3_0_FLASH_HIGH",
+  "gemini-3.0-flash-medium": "MODEL_GOOGLE_GEMINI_3_0_FLASH_MEDIUM",
+  "gemini-3.0-flash-low": "MODEL_GOOGLE_GEMINI_3_0_FLASH_LOW",
+  "gemini-3.0-flash-minimal": "MODEL_GOOGLE_GEMINI_3_0_FLASH_MINIMAL",
+  "gemini-3.0-flash": "MODEL_GOOGLE_GEMINI_3_0_FLASH_HIGH",
+  "gemini-2.5-pro": "MODEL_GOOGLE_GEMINI_2_5_PRO",
+  // ── Others ──────────────────────────────────────────────────────────────
+  "deepseek-v4": "deepseek-v4",
+  "kimi-k2.6": "kimi-k2-6",
+  "kimi-k2.5": "kimi-k2-5",
+  "glm-5.1": "glm-5-1",
+};
+
+export function resolveWsModelId(model) {
+  return MODEL_ALIAS_MAP[model] ?? model;
+}
+
+// ─── Minimal protobuf encoder ────────────────────────────────────────────────
+// Wire types: 0 = varint, 2 = length-delimited.
+
+function encodeVarint(value) {
+  const bytes = [];
+  let v = value >>> 0;
+  while (v > 0x7f) {
+    bytes.push((v & 0x7f) | 0x80);
+    v >>>= 7;
+  }
+  bytes.push(v & 0x7f);
+  return new Uint8Array(bytes);
+}
+
+function concatBytes(arrays) {
+  const total = arrays.reduce((n, a) => n + a.length, 0);
+  const out = new Uint8Array(total);
+  let off = 0;
+  for (const a of arrays) {
+    out.set(a, off);
+    off += a.length;
+  }
+  return out;
+}
+
+const TEXT_ENC = new TextEncoder();
+const TEXT_DEC = new TextDecoder();
+
+function encodeField(fieldNum, payload) {
+  const tag = encodeVarint((fieldNum << 3) | 2);
+  const len = encodeVarint(payload.length);
+  return concatBytes([tag, len, payload]);
+}
+
+function encodeString(fieldNum, value) {
+  return encodeField(fieldNum, TEXT_ENC.encode(value));
+}
+
+function encodeMessage(fieldNum, msg) {
+  return encodeField(fieldNum, msg);
+}
+
+// ─── Protobuf message builders ───────────────────────────────────────────────
+
+function buildMetadata(apiKey, sessionId) {
+  return concatBytes([
+    encodeString(1, apiKey),
+    encodeString(2, WS_IDE_NAME),
+    encodeString(3, WS_IDE_VERSION),
+    encodeString(4, WS_EXT_VERSION),
+    encodeString(5, sessionId),
+    encodeString(6, WS_LOCALE),
+  ]);
+}
+
+function buildModelOrAlias(model) {
+  return encodeString(1, model);
+}
+
+function buildChatMessage(msg) {
+  const parts = [encodeString(1, msg.role), encodeString(2, msg.content)];
+  if (msg.toolCallId) parts.push(encodeString(3, msg.toolCallId));
+  return concatBytes(parts);
+}
+
+export function buildGetChatMessageRequest(apiKey, model, messages) {
+  const sessionId = randomUUID();
+  const cascadeId = randomUUID();
+
+  const parts = [
+    encodeMessage(1, buildMetadata(apiKey, sessionId)), // metadata
+    encodeString(2, cascadeId), // cascade_id
+    encodeMessage(3, buildModelOrAlias(model)), // model_or_alias
+  ];
+
+  for (const msg of messages) {
+    parts.push(encodeMessage(4, buildChatMessage(msg))); // repeated messages
+  }
+
+  return concatBytes(parts);
+}
+
+// ─── gRPC-web framing ────────────────────────────────────────────────────────
+
+export function grpcWebFrame(payload) {
+  const frame = new Uint8Array(5 + payload.length);
+  frame[0] = 0x00; // no compression
+  const view = new DataView(frame.buffer);
+  view.setUint32(1, payload.length, false); // big-endian length
+  frame.set(payload, 5);
+  return frame;
+}
+
+// ─── Protobuf response decoder ───────────────────────────────────────────────
+// CompletionChunk (oneof):
+//   field 1 → ContentChunk { field 1: string text }
+//   field 2 → ToolCallChunk (skipped)
+//   field 3 → DoneChunk    { field 1: UsageStats{ field1: prompt, field2: completion } }
+//   field 4 → ErrorChunk   { field 1: string message }
+
+function readVarint(buf, offset) {
+  let result = 0;
+  let shift = 0;
+  while (offset < buf.length) {
+    const b = buf[offset++];
+    result |= (b & 0x7f) << shift;
+    if ((b & 0x80) === 0) break;
+    shift += 7;
+  }
+  return [result >>> 0, offset];
+}
+
+function decodeStringField(buf, targetField) {
+  let offset = 0;
+  while (offset < buf.length) {
+    let tag;
+    [tag, offset] = readVarint(buf, offset);
+    const fieldNum = tag >>> 3;
+    const wireType = tag & 0x07;
+    if (wireType === 2) {
+      let len;
+      [len, offset] = readVarint(buf, offset);
+      const payload = buf.slice(offset, offset + len);
+      offset += len;
+      if (fieldNum === targetField) return TEXT_DEC.decode(payload);
+    } else if (wireType === 0) {
+      let v;
+      [v, offset] = readVarint(buf, offset);
+    } else if (wireType === 1) {
+      offset += 8;
+    } else if (wireType === 5) {
+      offset += 4;
+    } else {
+      break;
+    }
+  }
+  return null;
+}
+
+function decodeDoneChunk(buf) {
+  // DoneChunk: field 1 = UsageStats (nested)
+  // UsageStats: field 1 = prompt_tokens (varint), field 2 = completion_tokens (varint)
+  let offset = 0;
+  let usageBytes = null;
+  while (offset < buf.length) {
+    let tag;
+    [tag, offset] = readVarint(buf, offset);
+    const fieldNum = tag >>> 3;
+    const wireType = tag & 0x07;
+    if (wireType === 2) {
+      let len;
+      [len, offset] = readVarint(buf, offset);
+      if (fieldNum === 1) usageBytes = buf.slice(offset, offset + len);
+      offset += len;
+    } else if (wireType === 0) {
+      let v;
+      [v, offset] = readVarint(buf, offset);
+    } else {
+      break;
+    }
+  }
+  if (!usageBytes) return [0, 0];
+  let promptTokens = 0;
+  let completionTokens = 0;
+  offset = 0;
+  while (offset < usageBytes.length) {
+    let tag;
+    [tag, offset] = readVarint(usageBytes, offset);
+    const fieldNum = tag >>> 3;
+    const wireType = tag & 0x07;
+    if (wireType === 0) {
+      let v;
+      [v, offset] = readVarint(usageBytes, offset);
+      if (fieldNum === 1) promptTokens = v;
+      else if (fieldNum === 2) completionTokens = v;
+    } else if (wireType === 2) {
+      let len;
+      [len, offset] = readVarint(usageBytes, offset);
+      offset += len;
+    } else {
+      break;
+    }
+  }
+  return [promptTokens, completionTokens];
+}
+
+export function decodeCompletionChunk(buf) {
+  let offset = 0;
+  while (offset < buf.length) {
+    let tag;
+    [tag, offset] = readVarint(buf, offset);
+    const fieldNum = tag >>> 3;
+    const wireType = tag & 0x07;
+
+    if (wireType === 2) {
+      let len;
+      [len, offset] = readVarint(buf, offset);
+      const payload = buf.slice(offset, offset + len);
+      offset += len;
+
+      if (fieldNum === 1) {
+        const text = decodeStringField(payload, 1);
+        if (text !== null) return { kind: "content", text };
+      } else if (fieldNum === 3) {
+        const usage = decodeDoneChunk(payload);
+        return { kind: "done", promptTokens: usage[0], completionTokens: usage[1] };
+      } else if (fieldNum === 4) {
+        const msg = decodeStringField(payload, 1);
+        return { kind: "error", message: msg ?? "unknown windsurf error" };
+      }
+      // field 2 = ToolCallChunk — not yet handled; skip
+    } else if (wireType === 0) {
+      let v;
+      [v, offset] = readVarint(buf, offset);
+    } else if (wireType === 1) {
+      offset += 8;
+    } else if (wireType === 5) {
+      offset += 4;
+    } else {
+      break;
+    }
+  }
+  return { kind: "unknown" };
+}
+
+// ─── OpenAI messages → Windsurf wire ─────────────────────────────────────────
+
+function openAIMessagesToWs(messages) {
+  const out = [];
+  for (const m of messages) {
+    const role = String(m.role || "user");
+    let content = "";
+    if (typeof m.content === "string") {
+      content = m.content;
+    } else if (Array.isArray(m.content)) {
+      for (const part of m.content) {
+        if (part && typeof part === "object" && part.type === "text") {
+          content += String(part.text || "");
+        }
+      }
+    }
+    out.push({ role, content, toolCallId: m.tool_call_id });
+  }
+  return out;
+}
+
+// ─── WindsurfExecutor ────────────────────────────────────────────────────────
+
+export class WindsurfExecutor extends BaseExecutor {
+  constructor() {
+    super("windsurf", PROVIDERS.windsurf || { id: "windsurf", baseUrl: WS_CHAT_URL });
+  }
+
+  buildUrl() {
+    return WS_CHAT_URL;
+  }
+
+  buildHeaders(credentials, stream = true) {
+    const token = credentials?.accessToken || credentials?.apiKey || "";
+    return {
+      "Content-Type": "application/grpc-web+proto",
+      Accept: "application/grpc-web+proto",
+      // Codeium apiKey also goes in Metadata.api_key (protobuf field) — see request body.
+      ...(token ? { Authorization: `Bearer ${token}` } : {}),
+      "User-Agent": `windsurf/${WS_IDE_VERSION}`,
+      "X-Grpc-Web": "1",
+    };
+  }
+
+  // Request body is built manually in execute() — requires model + messages.
+  transformRequest() {
+    return null;
+  }
+
+  async execute({ model, body, stream, credentials, signal, log, upstreamExtraHeaders, proxyOptions = null }) {
+    const apiKey = credentials?.accessToken || credentials?.apiKey || "";
+    const wsModel = resolveWsModelId(model);
+
+    const b = body ?? {};
+    const rawMessages = Array.isArray(b.messages) ? b.messages : [];
+    let wsMessages = openAIMessagesToWs(rawMessages);
+    if (wsMessages.length === 0) {
+      wsMessages.push({ role: "user", content: "" });
+    }
+
+    const protoPayload = buildGetChatMessageRequest(apiKey, wsModel, wsMessages);
+    const framedPayload = grpcWebFrame(protoPayload);
+
+    const url = this.buildUrl();
+    const headers = this.buildHeaders(credentials);
+    if (upstreamExtraHeaders) Object.assign(headers, upstreamExtraHeaders);
+
+    log?.debug?.("WS", `Windsurf → ${wsModel} (${wsMessages.length} messages)`);
+
+    const upstream = await proxyAwareFetch(url, {
+      method: "POST",
+      headers,
+      body: framedPayload,
+      signal,
+    }, proxyOptions);
+
+    if (!upstream.ok && upstream.status !== 200) {
+      return { response: upstream, url, headers, transformedBody: protoPayload };
+    }
+
+    const sseResponse = this.transformToSSE(upstream, model);
+    return { response: sseResponse, url, headers, transformedBody: protoPayload };
+  }
+
+  // Convert a gRPC-web binary response into an OpenAI-compatible SSE stream.
+  transformToSSE(upstream, model) {
+    const responseId = `chatcmpl-ws-${Date.now()}`;
+    const created = Math.floor(Date.now() / 1000);
+    const executor = this;
+
+    const sseStream = new ReadableStream({
+      async start(controller) {
+        const enc = new TextEncoder();
+        let roleEmitted = false;
+        let totalText = "";
+        let promptTokens = 0;
+        let completionTokens = 0;
+        let hadError = null;
+
+        const emit = (data) => controller.enqueue(enc.encode(data));
+
+        try {
+          let pending = new Uint8Array(0);
+          const reader = upstream.body?.getReader();
+
+          const handleFrame = (flag, payload) => {
+            if (flag === 0x80) {
+              // Trailer frame — contains grpc-status, grpc-message
+              const trailer = TEXT_DEC.decode(payload);
+              const statusMatch = /grpc-status:\s*(\d+)/i.exec(trailer);
+              if (statusMatch && statusMatch[1] !== "0") {
+                const msgMatch = /grpc-message:\s*(.+)/i.exec(trailer);
+                hadError = msgMatch
+                  ? decodeURIComponent(msgMatch[1].trim())
+                  : `gRPC status ${statusMatch[1]}`;
+              }
+              return;
+            }
+            if (flag !== 0x00) return; // skip unknown flags
+
+            const chunk = executor.constructor.decodeCompletionChunk
+              ? executor.constructor.decodeCompletionChunk(payload)
+              : decodeCompletionChunk(payload);
+
+            if (chunk.kind === "content" && chunk.text) {
+              totalText += chunk.text;
+              if (!roleEmitted) {
+                emit(`data: ${JSON.stringify({
+                  id: responseId, object: "chat.completion.chunk", created, model,
+                  choices: [{ index: 0, delta: { role: "assistant", content: "" }, finish_reason: null }],
+                })}\n\n`);
+                roleEmitted = true;
+              }
+              emit(`data: ${JSON.stringify({
+                id: responseId, object: "chat.completion.chunk", created, model,
+                choices: [{ index: 0, delta: { content: chunk.text }, finish_reason: null }],
+              })}\n\n`);
+            } else if (chunk.kind === "done") {
+              promptTokens = chunk.promptTokens;
+              completionTokens = chunk.completionTokens;
+            } else if (chunk.kind === "error") {
+              hadError = chunk.message;
+            }
+          };
+
+          const drainFrames = () => {
+            let offset = 0;
+            while (offset + 5 <= pending.length) {
+              const flag = pending[offset];
+              const len =
+                (pending[offset + 1] << 24) |
+                (pending[offset + 2] << 16) |
+                (pending[offset + 3] << 8) |
+                pending[offset + 4];
+              if (len < 0 || offset + 5 + len > pending.length) break;
+              handleFrame(flag, pending.slice(offset + 5, offset + 5 + len));
+              offset += 5 + len;
+            }
+            if (offset > 0) pending = pending.slice(offset);
+          };
+
+          if (reader) {
+            try {
+              while (true) {
+                const { done, value } = await reader.read();
+                if (done) break;
+                if (!value) continue;
+                pending = pending.length === 0 ? value : concatBytes([pending, value]);
+                drainFrames();
+              }
+            } finally {
+              reader.releaseLock();
+            }
+          }
+          drainFrames();
+
+          if (hadError) {
+            emit(`data: ${JSON.stringify({
+              error: { message: hadError, type: "windsurf_error", code: "upstream_error" },
+            })}\n\n`);
+            emit("data: [DONE]\n\n");
+            controller.close();
+            return;
+          }
+
+          // Unary fallback: nothing streamed but text decoded → emit as one chunk.
+          if (!roleEmitted && totalText) {
+            emit(`data: ${JSON.stringify({
+              id: responseId, object: "chat.completion.chunk", created, model,
+              choices: [{ index: 0, delta: { role: "assistant", content: "" }, finish_reason: null }],
+            })}\n\n`);
+            emit(`data: ${JSON.stringify({
+              id: responseId, object: "chat.completion.chunk", created, model,
+              choices: [{ index: 0, delta: { content: totalText }, finish_reason: null }],
+            })}\n\n`);
+          }
+
+          const finishPayload = {
+            id: responseId, object: "chat.completion.chunk", created, model,
+            choices: [{ index: 0, delta: {}, finish_reason: "stop" }],
+          };
+          if (promptTokens > 0 || completionTokens > 0) {
+            finishPayload.usage = {
+              prompt_tokens: promptTokens,
+              completion_tokens: completionTokens,
+              total_tokens: promptTokens + completionTokens,
+            };
+          }
+          emit(`data: ${JSON.stringify(finishPayload)}\n\n`);
+          emit("data: [DONE]\n\n");
+        } catch (err) {
+          const msg = err?.message ? String(err.message) : String(err);
+          emit(`data: ${JSON.stringify({
+            error: { message: `Windsurf stream error: ${msg}`, type: "windsurf_error" },
+          })}\n\n`);
+          emit("data: [DONE]\n\n");
+        }
+
+        controller.close();
+      },
+    });
+
+    return new Response(sseStream, {
+      status: 200,
+      headers: {
+        "Content-Type": "text/event-stream",
+        "Cache-Control": "no-cache",
+        Connection: "keep-alive",
+      },
+    });
+  }
+
+  // apiKey is long-lived (Firebase-derived or Devin ide_token); refresh handled out-of-band.
+  async refreshCredentials() {
+    return null;
+  }
+}
+
+export default WindsurfExecutor;
--- a/open-sse/executors/zed.js
+++ b/open-sse/executors/zed.js
@@ -0,0 +1,304 @@
+// ZedHostedExecutor — routes requests to Zed's hosted LLM aggregator
+// (cloud.zed.dev/completions), a multi-format proxy fronting
+// Anthropic/OpenAI/Google/xAI depending on the requested model.
+//
+// Wire protocol: POST /completions with an NDJSON/SSE-ish body-per-line
+// response stream (`{"event": <provider-shaped-chunk>}` / `{"status": ...}` /
+// `[DONE]`), authenticated with a short-lived LLM bearer token exchanged from
+// the RSA-decrypted access_token (see open-sse/shared/zedAuth.js). The
+// provider-shaped chunk is Claude/Gemini/OpenAI-Responses/xAI(OpenAI-shaped)
+// depending on which upstream Zed fronts for the model — translated back to
+// OpenAI Chat Completions by reusing the existing translators.
+//
+// Overrides execute() entirely (does NOT use DefaultExecutor's pipeline) because the Zed wire
+// shape (thread envelope, LLM-token exchange, NDJSON status frames) doesn't
+// fit the generic transformRequest/buildUrl contract.
+
+import { BaseExecutor } from "./base.js";
+import { FORMATS } from "../translator/formats.js";
+import { initState } from "../translator/index.js";
+import { openaiToClaudeRequest } from "../translator/request/openai-to-claude.js";
+import { openaiToGeminiRequest } from "../translator/request/openai-to-gemini.js";
+import { openaiToOpenAIResponsesRequest } from "../translator/request/openai-responses.js";
+import { claudeToOpenAIResponse } from "../translator/response/claude-to-openai.js";
+import { geminiToOpenAIResponse } from "../translator/response/gemini-to-openai.js";
+import { openaiResponsesToOpenAIResponse } from "../translator/response/openai-responses.js";
+import {
+  ZED_HEADERS,
+  resolveZedModels,
+  zedLlmFetch,
+} from "../shared/zedAuth.js";
+
+const ZED_PROVIDER = {
+  anthropic: "Anthropic",
+  openai: "OpenAi",
+  google: "Google",
+  xai: "XAi",
+};
+
+function normalizeZedProvider(value, model) {
+  const raw = String(value || "").toLowerCase();
+  if (raw === "anthropic") return ZED_PROVIDER.anthropic;
+  if (raw === "openai" || raw === "open_ai") return ZED_PROVIDER.openai;
+  if (raw === "google" || raw === "gemini") return ZED_PROVIDER.google;
+  if (raw === "xai" || raw === "x_ai" || raw === "x-ai") return ZED_PROVIDER.xai;
+
+  const m = String(model || "").toLowerCase();
+  if (m.includes("claude")) return ZED_PROVIDER.anthropic;
+  if (m.includes("gemini")) return ZED_PROVIDER.google;
+  if (m.includes("grok") || m.includes("xai")) return ZED_PROVIDER.xai;
+  return ZED_PROVIDER.openai;
+}
+
+function buildProviderRequest(provider, model, body, stream, credentials) {
+  if (provider === ZED_PROVIDER.anthropic) {
+    return openaiToClaudeRequest(model, body, true);
+  }
+  if (provider === ZED_PROVIDER.google) {
+    return openaiToGeminiRequest(model, body, true);
+  }
+  if (provider === ZED_PROVIDER.openai) {
+    return openaiToOpenAIResponsesRequest(model, body, true, credentials);
+  }
+  // xAI is OpenAI-shaped — forward as-is.
+  return { ...(body || {}), model, stream: stream !== false };
+}
+
+function initProviderState(provider, model) {
+  if (provider === ZED_PROVIDER.anthropic) return initState(FORMATS.CLAUDE);
+  if (provider === ZED_PROVIDER.google) return initState(FORMATS.GEMINI);
+  if (provider === ZED_PROVIDER.openai) return initState(FORMATS.OPENAI_RESPONSES);
+  const state = initState(FORMATS.OPENAI);
+  state.model = model;
+  return state;
+}
+
+function convertProviderEvent(provider, event, state) {
+  if (provider === ZED_PROVIDER.anthropic) return claudeToOpenAIResponse(event, state);
+  if (provider === ZED_PROVIDER.google) return geminiToOpenAIResponse(event, state);
+  if (provider === ZED_PROVIDER.openai) return openaiResponsesToOpenAIResponse(event, state);
+  return event;
+}
+
+function createErrorChunk(model, message) {
+  return {
+    id: `chatcmpl-zed-error-${Date.now()}`,
+    object: "chat.completion.chunk",
+    created: Math.floor(Date.now() / 1000),
+    model,
+    choices: [
+      { index: 0, delta: { content: `[Zed error] ${message}` }, finish_reason: "stop" },
+    ],
+  };
+}
+
+function enqueueSseObject(controller, encoder, chunk) {
+  if (!chunk) return;
+  const items = Array.isArray(chunk) ? chunk : [chunk];
+  for (const item of items) {
+    if (!item) continue;
+    controller.enqueue(encoder.encode(`data: ${JSON.stringify(item)}\n\n`));
+  }
+}
+
+function unwrapZedLine(line) {
+  let text = line.replace(/\r$/, "").trim();
+  if (!text) return null;
+  if (text.startsWith("data:")) text = text.slice(5).trimStart();
+  if (text === "[DONE]") return { done: true };
+  try {
+    const parsed = JSON.parse(text);
+    if (parsed && Object.prototype.hasOwnProperty.call(parsed, "event")) {
+      return { event: parsed.event };
+    }
+    if (parsed && Object.prototype.hasOwnProperty.call(parsed, "status")) {
+      return { status: parsed.status };
+    }
+    return { event: parsed };
+  } catch {
+    return null;
+  }
+}
+
+function normalizeStatus(status) {
+  if (!status) return null;
+  if (typeof status === "string") return { type: status };
+  if (typeof status === "object") {
+    const key = Object.keys(status)[0];
+    if (key && typeof status[key] === "object") return { type: key, ...status[key] };
+    return status;
+  }
+  return null;
+}
+
+function wrapZedCompletionStream(response, provider, model) {
+  if (!response.ok || !response.body) return response;
+
+  const decoder = new TextDecoder();
+  const encoder = new TextEncoder();
+  const state = initProviderState(provider, model);
+  let buffer = "";
+  let done = false;
+
+  const finish = (controller) => {
+    if (done) return;
+    const finalChunk = convertProviderEvent(provider, null, state);
+    enqueueSseObject(controller, encoder, finalChunk);
+    controller.enqueue(encoder.encode("data: [DONE]\n\n"));
+    done = true;
+  };
+
+  const processLine = (line, controller) => {
+    if (done) return;
+    const payload = unwrapZedLine(line);
+    if (!payload) return;
+    if (payload.done) {
+      finish(controller);
+      return;
+    }
+    if (payload.status) {
+      const status = normalizeStatus(payload.status);
+      if (status?.type === "failed" || status?.failed) {
+        const failed = status.failed || status;
+        const message = String(failed.message || failed.error || failed.code || "request failed");
+        enqueueSseObject(controller, encoder, createErrorChunk(model, message));
+        finish(controller);
+      } else if (status?.type === "stream_ended" || status === "stream_ended") {
+        finish(controller);
+      }
+      return;
+    }
+    const converted = convertProviderEvent(provider, payload.event, state);
+    enqueueSseObject(controller, encoder, converted);
+  };
+
+  const transformed = response.body.pipeThrough(
+    new TransformStream({
+      transform(chunk, controller) {
+        buffer += decoder.decode(chunk, { stream: true });
+        let nl;
+        while ((nl = buffer.indexOf("\n")) !== -1) {
+          const line = buffer.slice(0, nl);
+          buffer = buffer.slice(nl + 1);
+          processLine(line, controller);
+        }
+      },
+      flush(controller) {
+        buffer += decoder.decode();
+        if (buffer) {
+          processLine(buffer, controller);
+          buffer = "";
+        }
+        finish(controller);
+      },
+    }),
+  );
+
+  return new Response(transformed, {
+    status: response.status,
+    statusText: response.statusText,
+    headers: {
+      "Content-Type": "text/event-stream",
+      "Cache-Control": "no-cache",
+    },
+  });
+}
+
+class ZedExecutor extends BaseExecutor {
+  constructor() {
+    super("zed");
+  }
+
+  async resolveModel(model, credentials, signal, log) {
+    try {
+      const catalog = await resolveZedModels(credentials, { config: this.config, signal });
+      let raw = catalog?.rawById?.get(model) ?? null;
+      if (!raw) {
+        const refreshed = await resolveZedModels(credentials, {
+          config: this.config,
+          signal,
+          forceRefresh: true,
+        });
+        raw = refreshed?.rawById?.get(model) ?? null;
+      }
+      return { raw, provider: normalizeZedProvider(raw?.provider, model) };
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      log?.warn?.("ZED", `model catalog unavailable, inferring provider for ${model}: ${message}`);
+      return { raw: null, provider: normalizeZedProvider(null, model) };
+    }
+  }
+
+  async execute({ model, body, stream, credentials, signal, log, proxyOptions = null }) {
+    const { provider } = await this.resolveModel(model, credentials, signal, log);
+    const providerRequest = buildProviderRequest(provider, model, body, stream, credentials);
+    const bodyRecord = body || {};
+    const payload = {
+      thread_id: bodyRecord.thread_id || credentials?._clientSessionId,
+      prompt_id: bodyRecord.prompt_id,
+      provider,
+      model,
+      provider_request: providerRequest,
+    };
+
+    const response = await zedLlmFetch(credentials, "/completions", {
+      config: this.config,
+      signal,
+      fetchOptions: {
+        method: "POST",
+        headers: {
+          "Content-Type": "application/json",
+          Accept: "application/x-ndjson, text/event-stream, */*",
+          "User-Agent": "9router/zed",
+          "x-zed-version": this.config?.appVersion?.toString() || "0.200.0",
+          [ZED_HEADERS.clientSupportsStatus]: "true",
+          [ZED_HEADERS.clientSupportsStreamEnded]: "true",
+        },
+        body: JSON.stringify(payload),
+      },
+    });
+
+    const wrapped = response.ok ? wrapZedCompletionStream(response, provider, model) : response;
+    return {
+      response: wrapped,
+      url: `${this.config?.llmBaseUrl || "https://cloud.zed.dev"}/completions`,
+      headers: { "Content-Type": "application/json", Authorization: "Bearer <zed-llm-token>" },
+      transformedBody: payload,
+    };
+  }
+
+  parseError(response, bodyText) {
+    let parsed = null;
+    try {
+      parsed = JSON.parse(bodyText || "{}");
+    } catch {
+      parsed = null;
+    }
+
+    const errorObj = parsed?.error || undefined;
+    const code = parsed?.code || errorObj?.code || "";
+    const rawMessage =
+      parsed?.message || errorObj?.message || bodyText || response.statusText;
+    if (code === "trial_blocked") {
+      return {
+        status: response.status,
+        message: `Zed trial access is blocked upstream. The account can list hosted models, but Zed is refusing completions until trial/billing access is enabled or unblocked. Zed says: ${rawMessage}`,
+      };
+    }
+    if (code) {
+      return { status: response.status, message: `Zed ${code}: ${rawMessage}` };
+    }
+    return { status: response.status, message: rawMessage || `Zed upstream error: ${response.status}` };
+  }
+
+  async refreshCredentials() {
+    // Zed uses a long-lived RSA-decrypted access_token — no OAuth refresh.
+    return null;
+  }
+
+  needsRefresh() {
+    return false;
+  }
+}
+
+export default ZedExecutor;
--- a/open-sse/handlers/chatCore.js
+++ b/open-sse/handlers/chatCore.js
--- a/open-sse/handlers/chatCore/nonStreamingHandler.js
+++ b/open-sse/handlers/chatCore/nonStreamingHandler.js
@@ -1,19 +1,74 @@
 import { FORMATS } from "../../translator/formats.js";
 import { needsTranslation } from "../../translator/index.js";
+import { fromOpenAIFinish } from "../../translator/concerns/finishReason.js";
 import { ollamaBodyToOpenAI } from "../../translator/response/ollama-to-openai.js";
 import { addBufferToUsage, filterUsageForFormat } from "../../utils/usageTracking.js";
 import { createErrorResult } from "../../utils/error.js";
 import { HTTP_STATUS } from "../../config/runtimeConfig.js";
 import { parseSSEToOpenAIResponse } from "./sseToJsonHandler.js";
-import { buildRequestDetail, extractRequestConfig, extractUsageFromResponse, saveUsageStats } from "./requestDetail.js";
+import { buildRequestDetail, extractRequestConfig, extractUsageFromResponse, saveUsageStats, formatDoneLine } from "./requestDetail.js";
 import { appendRequestLog, saveRequestDetail } from "@/lib/usageDb.js";
 import { decloakToolNames } from "../../utils/claudeCloaking.js";

+function parseToolArguments(value) {
+  if (!value) return {};
+  if (typeof value === "object") return value;
+  try {
+    return JSON.parse(value);
+  } catch {
+    return {};
+  }
+}
+
+function openAICompletionToClaudeMessage(responseBody) {
+  if (!responseBody?.choices?.[0]) return responseBody;
+  const choice = responseBody.choices[0];
+  const message = choice.message || {};
+  const content = [];
+
+  const reasoning = message.reasoning_content || message.provider_specific_fields?.reasoning_content || "";
+  if (reasoning) {
+    content.push({ type: "thinking", thinking: reasoning });
+  }
+  if (typeof message.content === "string" && message.content.length > 0) {
+    content.push({ type: "text", text: message.content });
+  }
+  for (const toolCall of message.tool_calls || []) {
+    const fn = toolCall.function || {};
+    content.push({
+      type: "tool_use",
+      id: toolCall.id || `toolu_${Date.now()}_${content.length}`,
+      name: fn.name || toolCall.name || "",
+      input: parseToolArguments(fn.arguments || toolCall.arguments),
+    });
+  }
+  if (content.length === 0) content.push({ type: "text", text: "" });
+
+  const usage = responseBody.usage || {};
+  return {
+    id: String(responseBody.id || `msg_${Date.now()}`).replace(/^chatcmpl-/, ""),
+    type: "message",
+    role: "assistant",
+    model: responseBody.model || "unknown",
+    content,
+    stop_reason: fromOpenAIFinish(choice.finish_reason, FORMATS.CLAUDE),
+    stop_sequence: null,
+    usage: {
+      input_tokens: usage.prompt_tokens || usage.input_tokens || 0,
+      output_tokens: usage.completion_tokens || usage.output_tokens || 0,
+    },
+  };
+}
+
 /**
 * Translate non-streaming response body from provider format → OpenAI format.
 */
 export function translateNonStreamingResponse(responseBody, targetFormat, sourceFormat) {
-  if (targetFormat === sourceFormat || targetFormat === FORMATS.OPENAI) return responseBody;
+  if (targetFormat === sourceFormat) return responseBody;
+  if (targetFormat === FORMATS.OPENAI && sourceFormat === FORMATS.CLAUDE) {
+    return openAICompletionToClaudeMessage(responseBody);
+  }
+  if (targetFormat === FORMATS.OPENAI) return responseBody;

  // Gemini / Antigravity
  if (targetFormat === FORMATS.GEMINI || targetFormat === FORMATS.ANTIGRAVITY || targetFormat === FORMATS.GEMINI_CLI || targetFormat === FORMATS.VERTEX) {
@@ -143,7 +198,7 @@ export function translateNonStreamingResponse(responseBody, targetFormat, source
 /**
 * Handle non-streaming response from provider.
 */
-export async function handleNonStreamingResponse({ providerResponse, provider, model, sourceFormat, targetFormat, body, stream, translatedBody, finalBody, requestStartTime, connectionId, apiKey, clientRawRequest, onRequestSuccess, reqLogger, toolNameMap, trackDone, appendLog }) {
+export async function handleNonStreamingResponse({ providerResponse, provider, model, sourceFormat, targetFormat, body, stream, translatedBody, finalBody, requestStartTime, connectionId, apiKey, clientRawRequest, onRequestSuccess, reqLogger, toolNameMap, trackDone, appendLog, pxpipe, reqTag, log }) {
  trackDone();
  const contentType = providerResponse.headers.get("content-type") || "";
  let responseBody;
@@ -180,11 +235,13 @@ export async function handleNonStreamingResponse({ providerResponse, provider, m

  const usage = extractUsageFromResponse(responseBody);
  appendLog({ tokens: usage, status: "200 OK" });
-  saveUsageStats({ provider, model, tokens: usage, connectionId, apiKey, endpoint: clientRawRequest?.endpoint });
+  saveUsageStats({ provider, model, tokens: usage, connectionId, apiKey, endpoint: clientRawRequest?.endpoint, silent: true });
+  if (log?.line) log.line(reqTag, "📊", formatDoneLine({ usage, latency: { total: Date.now() - requestStartTime } }));

  const translatedResponse = needsTranslation(targetFormat, sourceFormat)
    ? translateNonStreamingResponse(responseBody, targetFormat, sourceFormat)
    : responseBody;
+  const isClaudeMessageResponse = sourceFormat === FORMATS.CLAUDE && translatedResponse?.type === "message";

  // Fix finish_reason for tool_calls: some providers return non-standard values (e.g. "other")
  if (translatedResponse?.choices?.[0]) {
@@ -197,13 +254,17 @@ export async function handleNonStreamingResponse({ providerResponse, provider, m
  }

  // Ensure OpenAI-required fields
-  if (!translatedResponse.object) translatedResponse.object = "chat.completion";
-  if (!translatedResponse.created) translatedResponse.created = Math.floor(Date.now() / 1000);
+  if (!isClaudeMessageResponse) {
+    if (!translatedResponse.object) translatedResponse.object = "chat.completion";
+    if (!translatedResponse.created) translatedResponse.created = Math.floor(Date.now() / 1000);
+  }

  // Strip Azure-specific fields
-  delete translatedResponse.prompt_filter_results;
-  if (translatedResponse?.choices) {
-    for (const choice of translatedResponse.choices) delete choice.content_filter_results;
+  if (!isClaudeMessageResponse) {
+    delete translatedResponse.prompt_filter_results;
+    if (translatedResponse?.choices) {
+      for (const choice of translatedResponse.choices) delete choice.content_filter_results;
+    }
  }

  if (translatedResponse?.usage) {
@@ -213,7 +274,7 @@ export async function handleNonStreamingResponse({ providerResponse, provider, m
  // Strip reasoning_content only when content is non-empty.
  // When content is empty (e.g. thinking models that used all tokens for reasoning),
  // reasoning_content is the only useful output and must be preserved.
-  if (translatedResponse?.choices) {
+  if (!isClaudeMessageResponse && translatedResponse?.choices) {
    for (const choice of translatedResponse.choices) {
      if (choice?.message?.reasoning_content && choice.message.content) {
        delete choice.message.reasoning_content;
@@ -236,6 +297,7 @@ export async function handleNonStreamingResponse({ providerResponse, provider, m
      thinking: translatedResponse?.choices?.[0]?.message?.reasoning_content || translatedResponse?.reasoning_content || null,
      finish_reason: translatedResponse?.choices?.[0]?.finish_reason || "unknown"
    },
+    pxpipe,
    status: "success"
  }, { endpoint: clientRawRequest?.endpoint || null })).catch(err => {
    console.error("[RequestDetail] Failed to save:", err.message);
--- a/open-sse/handlers/chatCore/requestDetail.js
+++ b/open-sse/handlers/chatCore/requestDetail.js
@@ -1,5 +1,6 @@
 import { saveRequestUsage, appendRequestLog, saveRequestDetail } from "@/lib/usageDb.js";
 import { COLORS } from "../../utils/stream.js";
+import { canonicalizeUsage } from "../../utils/usageTracking.js";

 const OPTIONAL_PARAMS = [
  "temperature", "top_p", "top_k",
@@ -48,7 +49,8 @@ export function extractUsageFromResponse(responseBody) {
    return {
      prompt_tokens: responseBody.usageMetadata.promptTokenCount || 0,
      completion_tokens: responseBody.usageMetadata.candidatesTokenCount || 0,
-      reasoning_tokens: responseBody.usageMetadata.thoughtsTokenCount
+      cached_tokens: responseBody.usageMetadata.cachedContentTokenCount || 0,
+      reasoning_tokens: responseBody.usageMetadata.thoughtsTokenCount || 0
    };
  }

@@ -67,12 +69,31 @@ export function buildRequestDetail(base, overrides = {}) {
    providerRequest: base.providerRequest || null,
    providerResponse: base.providerResponse || null,
    response: base.response || {},
+    pxpipe: base.pxpipe || undefined,
    status: base.status || "success",
    ...overrides
  };
 }

-export function saveUsageStats({ provider, model, tokens, connectionId, apiKey, endpoint, label = "USAGE" }) {
+// Build the "done" summary: duration, ttft, in/out tokens with cache breakdown
+export function formatDoneLine({ usage, latency }) {
+  const u = usage || {};
+  const inTok = u.prompt_tokens ?? u.input_tokens ?? 0;
+  const outTok = u.completion_tokens ?? u.output_tokens ?? 0;
+  const cacheRead = u.cache_read_input_tokens ?? u.cached_tokens ?? u.prompt_tokens_details?.cached_tokens ?? 0;
+  const cacheCreate = u.cache_creation_input_tokens ?? 0;
+  let inStr = `IN ${inTok}`;
+  if (cacheRead || cacheCreate) {
+    const parts = [];
+    if (cacheRead) parts.push(`↻${cacheRead}`);
+    if (cacheCreate) parts.push(`+${cacheCreate}`);
+    inStr += ` (CACHE ${parts.join(" ")})`;
+  }
+  const ttftStr = latency?.ttft ? ` · TTFT ${latency.ttft}ms` : "";
+  return `DONE ${latency?.total ?? 0}ms${ttftStr} · ${inStr} · OUT ${outTok}`;
+}
+
+export function saveUsageStats({ provider, model, tokens, connectionId, apiKey, endpoint, label = "USAGE", silent = false }) {
  if (!tokens || typeof tokens !== "object") return;

  const inTokens = tokens.input_tokens ?? tokens.prompt_tokens ?? 0;
@@ -80,12 +101,15 @@ export function saveUsageStats({ provider, model, tokens, connectionId, apiKey,

  if (inTokens === 0 && outTokens === 0) return;

-  const time = new Date().toLocaleTimeString("en-US", { hour12: false, hour: "2-digit", minute: "2-digit", second: "2-digit" });
-  const accountSuffix = connectionId ? ` | account=${connectionId.slice(0, 8)}...` : "";
-  console.log(`${COLORS.green}[${time}] 📊 [${label}] ${provider.toUpperCase()} | in=${inTokens} | out=${outTokens}${accountSuffix}${COLORS.reset}`);
+  if (!silent) {
+    const time = new Date().toLocaleTimeString("en-US", { hour12: false, hour: "2-digit", minute: "2-digit", second: "2-digit" });
+    const accountSuffix = connectionId ? ` | account=${connectionId.slice(0, 8)}...` : "";
+    console.log(`${COLORS.green}[${time}] 📊 [${label}] ${provider.toUpperCase()} | in=${inTokens} | out=${outTokens}${accountSuffix}${COLORS.reset}`);
+  }

-  // Normalize to OpenAI token shape for storage
-  const normalized = {
+  // Canonicalize to one storage convention (prompt_tokens cache-inclusive) so
+  // cached/cache-creation tokens survive to cost calc + stats. See canonicalizeUsage.
+  const normalized = canonicalizeUsage(tokens) || {
    prompt_tokens: tokens.prompt_tokens ?? tokens.input_tokens ?? 0,
    completion_tokens: tokens.completion_tokens ?? tokens.output_tokens ?? 0
  };
--- a/open-sse/handlers/chatCore/sseToJsonHandler.js
+++ b/open-sse/handlers/chatCore/sseToJsonHandler.js
@@ -1,21 +1,28 @@
 import { convertResponsesStreamToJson } from "../../transformer/streamToJsonConverter.js";
+import { matchStreamErrorPatterns } from "../../utils/streamErrorPatterns.js";
 import { createErrorResult } from "../../utils/error.js";
 import { HTTP_STATUS } from "../../config/runtimeConfig.js";
 import { FORMATS } from "../../translator/formats.js";
 import { PROVIDERS } from "../../config/providers.js";
-import { buildRequestDetail, extractRequestConfig, saveUsageStats } from "./requestDetail.js";
+import {
+	buildRequestDetail,
+	extractRequestConfig,
+	saveUsageStats,
+	formatDoneLine,
+} from "./requestDetail.js";

 // Responses-API providers (e.g. codex) may emit SSE without content-type + use Responses output shape
-const isResponsesProvider = (p) => PROVIDERS[p]?.format === FORMATS.OPENAI_RESPONSES;
+const isResponsesProvider = (p) =>
+	PROVIDERS[p]?.format === FORMATS.OPENAI_RESPONSES;
 import { saveRequestDetail, appendRequestLog } from "@/lib/usageDb.js";

 function textFromResponsesMessageItem(item) {
-  if (!item?.content || !Array.isArray(item.content)) return "";
-  const byType = item.content.find((c) => c.type === "output_text");
-  if (typeof byType?.text === "string") return byType.text;
-  const anyText = item.content.find((c) => typeof c.text === "string");
-  if (typeof anyText?.text === "string") return anyText.text;
-  return "";
+	if (!item?.content || !Array.isArray(item.content)) return "";
+	const byType = item.content.find((c) => c.type === "output_text");
+	if (typeof byType?.text === "string") return byType.text;
+	const anyText = item.content.find((c) => typeof c.text === "string");
+	if (typeof anyText?.text === "string") return anyText.text;
+	return "";
 }

 /**
@@ -23,15 +30,15 @@ function textFromResponsesMessageItem(item) {
 * Early message blocks often have empty output_text; the user-visible answer is usually in the last non-empty message.
 */
 function pickAssistantMessageForChatCompletion(output) {
-  if (!Array.isArray(output)) return { msgItem: null, textContent: null };
-  const messages = output.filter((item) => item?.type === "message");
-  if (messages.length === 0) return { msgItem: null, textContent: null };
-  for (let i = messages.length - 1; i >= 0; i--) {
-    const text = textFromResponsesMessageItem(messages[i]);
-    if (text.length > 0) return { msgItem: messages[i], textContent: text };
-  }
-  const last = messages[messages.length - 1];
-  return { msgItem: last, textContent: textFromResponsesMessageItem(last) };
+	if (!Array.isArray(output)) return { msgItem: null, textContent: null };
+	const messages = output.filter((item) => item?.type === "message");
+	if (messages.length === 0) return { msgItem: null, textContent: null };
+	for (let i = messages.length - 1; i >= 0; i--) {
+		const text = textFromResponsesMessageItem(messages[i]);
+		if (text.length > 0) return { msgItem: messages[i], textContent: text };
+	}
+	const last = messages[messages.length - 1];
+	return { msgItem: last, textContent: textFromResponsesMessageItem(last) };
 }

 /**
@@ -39,197 +46,388 @@ function pickAssistantMessageForChatCompletion(output) {
 * Used when provider forces streaming but client wants non-streaming.
 */
 export function parseSSEToOpenAIResponse(rawSSE, fallbackModel) {
-  const chunks = [];
+	const chunks = [];
+	let streamError = null;

-  for (const line of String(rawSSE || "").split("\n")) {
-    const trimmed = line.trim();
-    if (!trimmed.startsWith("data:")) continue;
-    const payload = trimmed.slice(5).trim();
-    if (!payload || payload === "[DONE]") continue;
-    try { chunks.push(JSON.parse(payload)); } catch { /* ignore malformed lines */ }
-  }
+	for (const line of String(rawSSE || "").split("\n")) {
+		const trimmed = line.trim();
+		if (!trimmed.startsWith("data:")) continue;
+		const payload = trimmed.slice(5).trim();
+		if (!payload || payload === "[DONE]") continue;
+		try {
+			const chunk = JSON.parse(payload);
+			if (chunk?.error) streamError = chunk.error;
+			else chunks.push(chunk);
+		} catch {
+			/* ignore malformed lines */
+		}
+	}

-  if (chunks.length === 0) return null;
+	if (streamError) return { error: streamError };
+	if (chunks.length === 0) return null;

-  const first = chunks[0];
-  const contentParts = [];
-  const reasoningParts = [];
-  const toolCallMap = new Map(); // index -> { id, type, function: { name, arguments } }
-  let finishReason = "stop";
-  let usage = null;
+	const first = chunks[0];
+	const contentParts = [];
+	const reasoningParts = [];
+	const toolCallMap = new Map(); // index -> { id, type, function: { name, arguments } }
+	let finishReason = "stop";
+	let usage = null;

-  for (const chunk of chunks) {
-    const choice = chunk?.choices?.[0];
-    const delta = choice?.delta || {};
-    if (typeof delta.content === "string" && delta.content.length > 0) contentParts.push(delta.content);
-    if (typeof delta.reasoning_content === "string" && delta.reasoning_content.length > 0) reasoningParts.push(delta.reasoning_content);
-    if (choice?.finish_reason) finishReason = choice.finish_reason;
-    if (chunk?.usage && typeof chunk.usage === "object") usage = chunk.usage;
+	for (const chunk of chunks) {
+		const choice = chunk?.choices?.[0];
+		const delta = choice?.delta || {};
+		if (typeof delta.content === "string" && delta.content.length > 0)
+			contentParts.push(delta.content);
+		if (
+			typeof delta.reasoning_content === "string" &&
+			delta.reasoning_content.length > 0
+		)
+			reasoningParts.push(delta.reasoning_content);
+		if (choice?.finish_reason) finishReason = choice.finish_reason;
+		if (chunk?.usage && typeof chunk.usage === "object") usage = chunk.usage;

-    // Accumulate tool_calls from streaming deltas
-    if (Array.isArray(delta.tool_calls)) {
-      for (const tc of delta.tool_calls) {
-        const idx = tc.index ?? 0;
-        if (!toolCallMap.has(idx)) {
-          toolCallMap.set(idx, { id: tc.id || "", type: "function", function: { name: "", arguments: "" } });
-        }
-        const existing = toolCallMap.get(idx);
-        if (tc.id) existing.id = tc.id;
-        if (tc.function?.name) existing.function.name += tc.function.name;
-        if (tc.function?.arguments) existing.function.arguments += tc.function.arguments;
-      }
-    }
-  }
+		// Accumulate tool_calls from streaming deltas
+		if (Array.isArray(delta.tool_calls)) {
+			for (const tc of delta.tool_calls) {
+				const idx = tc.index ?? 0;
+				if (!toolCallMap.has(idx)) {
+					toolCallMap.set(idx, {
+						id: tc.id || "",
+						type: "function",
+						function: { name: "", arguments: "" },
+					});
+				}
+				const existing = toolCallMap.get(idx);
+				if (tc.id) existing.id = tc.id;
+				if (tc.function?.name) existing.function.name += tc.function.name;
+				if (tc.function?.arguments)
+					existing.function.arguments += tc.function.arguments;
+			}
+		}
+	}

-  const message = { role: "assistant", content: contentParts.join("") || (toolCallMap.size > 0 ? null : "") };
-  if (reasoningParts.length > 0) message.reasoning_content = reasoningParts.join("");
-  if (toolCallMap.size > 0) {
-    message.tool_calls = [...toolCallMap.entries()].sort((a, b) => a[0] - b[0]).map(([, tc]) => tc);
-  }
+	const message = {
+		role: "assistant",
+		content: contentParts.join("") || (toolCallMap.size > 0 ? null : ""),
+	};
+	if (reasoningParts.length > 0)
+		message.reasoning_content = reasoningParts.join("");
+	if (toolCallMap.size > 0) {
+		message.tool_calls = [...toolCallMap.entries()]
+			.sort((a, b) => a[0] - b[0])
+			.map(([, tc]) => tc);
+	}

-  const result = {
-    id: first.id || `chatcmpl-${Date.now()}`,
-    object: "chat.completion",
-    created: first.created || Math.floor(Date.now() / 1000),
-    model: first.model || fallbackModel || "unknown",
-    choices: [{ index: 0, message, finish_reason: finishReason }]
-  };
-  if (usage) result.usage = usage;
-  return result;
+	const result = {
+		id: first.id || `chatcmpl-${Date.now()}`,
+		object: "chat.completion",
+		created: first.created || Math.floor(Date.now() / 1000),
+		model: first.model || fallbackModel || "unknown",
+		choices: [{ index: 0, message, finish_reason: finishReason }],
+	};
+	if (usage) result.usage = usage;
+	return result;
 }

 /**
 * Handle case: provider forced streaming but client wants JSON.
 * Supports both Codex/Responses API SSE and standard Chat Completions SSE.
 */
-export async function handleForcedSSEToJson({ providerResponse, sourceFormat, provider, model, body, stream, translatedBody, finalBody, requestStartTime, connectionId, apiKey, clientRawRequest, onRequestSuccess, trackDone, appendLog }) {
-  const contentType = providerResponse.headers.get("content-type") || "";
-  const isSSE = contentType.includes("text/event-stream") || (contentType === "" && isResponsesProvider(provider));
-  if (!isSSE) return null; // not handled here
+export async function handleForcedSSEToJson({
+	providerResponse,
+	sourceFormat,
+	provider,
+	model,
+	body,
+	stream,
+	translatedBody,
+	finalBody,
+	requestStartTime,
+	connectionId,
+	apiKey,
+	clientRawRequest,
+	onRequestSuccess,
+	trackDone,
+	appendLog,
+	reqTag,
+	log,
+	streamErrorPatterns,
+}) {
+	const contentType = providerResponse.headers.get("content-type") || "";
+	const isSSE =
+		contentType.includes("text/event-stream") ||
+		(contentType === "" && isResponsesProvider(provider));
+	if (!isSSE) return null; // not handled here

-  trackDone();
+	trackDone();

-  const ctx = {
-    provider, model, connectionId,
-    request: extractRequestConfig(body, stream),
-    providerRequest: finalBody || translatedBody || null
-  };
+	const ctx = {
+		provider,
+		model,
+		connectionId,
+		request: extractRequestConfig(body, stream),
+		providerRequest: finalBody || translatedBody || null,
+	};

-  // Codex/Responses API SSE path
-  const isCodexResponsesApi = isResponsesProvider(provider) || sourceFormat === FORMATS.OPENAI_RESPONSES;
-  if (isCodexResponsesApi) {
-    try {
-      const jsonResponse = await convertResponsesStreamToJson(providerResponse.body);
-      if (onRequestSuccess) await onRequestSuccess();
+	// Codex/Responses API SSE path
+	const isCodexResponsesApi =
+		isResponsesProvider(provider) || sourceFormat === FORMATS.OPENAI_RESPONSES;
+	if (isCodexResponsesApi) {
+		try {
+			const jsonResponse = await convertResponsesStreamToJson(
+				providerResponse.body,
+			);
+			if (onRequestSuccess) await onRequestSuccess();

-      const usage = jsonResponse.usage || {};
-      appendLog({ tokens: usage, status: "200 OK" });
-      saveUsageStats({ provider, model, tokens: usage, connectionId, apiKey, endpoint: clientRawRequest?.endpoint });
+			const usage = jsonResponse.usage || {};
+			appendLog({ tokens: usage, status: "200 OK" });
+			saveUsageStats({
+				provider,
+				model,
+				tokens: usage,
+				connectionId,
+				apiKey,
+				endpoint: clientRawRequest?.endpoint,
+				silent: true,
+			});
+			if (log?.line)
+				log.line(
+					reqTag,
+					"📊",
+					formatDoneLine({
+						usage,
+						latency: { total: Date.now() - requestStartTime },
+					}),
+				);

-      const { msgItem, textContent } = pickAssistantMessageForChatCompletion(jsonResponse.output);
-      const totalLatency = Date.now() - requestStartTime;
+			const { msgItem, textContent } = pickAssistantMessageForChatCompletion(
+				jsonResponse.output,
+			);
+			const totalLatency = Date.now() - requestStartTime;

-      saveRequestDetail(buildRequestDetail({
-        ...ctx,
-        latency: { ttft: totalLatency, total: totalLatency },
-        tokens: { prompt_tokens: usage.input_tokens || 0, completion_tokens: usage.output_tokens || 0 },
-        response: { content: textContent, thinking: null, finish_reason: jsonResponse.status || "unknown" },
-        status: "success"
-      }, { endpoint: clientRawRequest?.endpoint || null })).catch(() => {});
+			saveRequestDetail(
+				buildRequestDetail(
+					{
+						...ctx,
+						latency: { ttft: totalLatency, total: totalLatency },
+						tokens: {
+							prompt_tokens: usage.input_tokens || 0,
+							completion_tokens: usage.output_tokens || 0,
+						},
+						response: {
+							content: textContent,
+							thinking: null,
+							finish_reason: jsonResponse.status || "unknown",
+						},
+						status: "success",
+					},
+					{ endpoint: clientRawRequest?.endpoint || null },
+				),
+			).catch(() => {});

-      // Client is Responses API → return as-is
-      if (sourceFormat === FORMATS.OPENAI_RESPONSES) {
-        return { success: true, response: new Response(JSON.stringify(jsonResponse), { headers: { "Content-Type": "application/json", "Access-Control-Allow-Origin": "*" } }) };
-      }
+			// Client is Responses API → return as-is
+			if (sourceFormat === FORMATS.OPENAI_RESPONSES) {
+				return {
+					success: true,
+					response: new Response(JSON.stringify(jsonResponse), {
+						headers: {
+							"Content-Type": "application/json",
+							"Access-Control-Allow-Origin": "*",
+						},
+					}),
+				};
+			}

-      // Build client-format response
-      const inTokens = usage.input_tokens || 0;
-      const outTokens = usage.output_tokens || 0;
-      let finalResp;
+			// Build client-format response
+			const inTokens = usage.input_tokens || 0;
+			const outTokens = usage.output_tokens || 0;
+			let finalResp;

-      // Extract tool calls from Responses API output (function_call items)
-      const funcCallItems = (jsonResponse.output || []).filter(item => item.type === "function_call");
-      const toolCalls = funcCallItems.map((item, idx) => ({
-        id: item.call_id || `call_${item.name}_${Date.now()}_${idx}`,
-        type: "function",
-        function: {
-          name: item.name,
-          arguments: typeof item.arguments === "string" ? item.arguments : JSON.stringify(item.arguments || {})
-        }
-      }));
-      const hasToolCalls = toolCalls.length > 0;
+			// Extract tool calls from Responses API output (function_call items)
+			const funcCallItems = (jsonResponse.output || []).filter(
+				(item) => item.type === "function_call",
+			);
+			const toolCalls = funcCallItems.map((item, idx) => ({
+				id: item.call_id || `call_${item.name}_${Date.now()}_${idx}`,
+				type: "function",
+				function: {
+					name: item.name,
+					arguments:
+						typeof item.arguments === "string"
+							? item.arguments
+							: JSON.stringify(item.arguments || {}),
+				},
+			}));
+			const hasToolCalls = toolCalls.length > 0;

-      if (sourceFormat === FORMATS.ANTIGRAVITY || sourceFormat === FORMATS.GEMINI || sourceFormat === FORMATS.GEMINI_CLI) {
-        finalResp = {
-          response: {
-            candidates: [{ content: { role: "model", parts: [{ text: textContent || "" }] }, finishReason: "STOP", index: 0 }],
-            usageMetadata: { promptTokenCount: inTokens, candidatesTokenCount: outTokens, totalTokenCount: inTokens + outTokens },
-            modelVersion: model,
-            responseId: jsonResponse.id || `resp_${Date.now()}`
-          }
-        };
-      } else {
-        const message = { role: "assistant", content: textContent || (hasToolCalls ? null : "") };
-        if (hasToolCalls) message.tool_calls = toolCalls;
-        const responseDone = jsonResponse.status === "completed" || jsonResponse.status === "done";
-        const finishReason = hasToolCalls ? "tool_calls" : (responseDone ? "stop" : (jsonResponse.status || "stop"));
-        finalResp = {
-          id: jsonResponse.id || `chatcmpl-${Date.now()}`,
-          object: "chat.completion",
-          created: jsonResponse.created_at || Math.floor(Date.now() / 1000),
-          model: jsonResponse.model || model,
-          choices: [{ index: 0, message, finish_reason: finishReason }],
-          usage: { prompt_tokens: inTokens, completion_tokens: outTokens, total_tokens: inTokens + outTokens }
-        };
-      }
+			if (
+				sourceFormat === FORMATS.ANTIGRAVITY ||
+				sourceFormat === FORMATS.GEMINI ||
+				sourceFormat === FORMATS.GEMINI_CLI
+			) {
+				finalResp = {
+					response: {
+						candidates: [
+							{
+								content: {
+									role: "model",
+									parts: [{ text: textContent || "" }],
+								},
+								finishReason: "STOP",
+								index: 0,
+							},
+						],
+						usageMetadata: {
+							promptTokenCount: inTokens,
+							candidatesTokenCount: outTokens,
+							totalTokenCount: inTokens + outTokens,
+						},
+						modelVersion: model,
+						responseId: jsonResponse.id || `resp_${Date.now()}`,
+					},
+				};
+			} else {
+				const message = {
+					role: "assistant",
+					content: textContent || (hasToolCalls ? null : ""),
+				};
+				if (hasToolCalls) message.tool_calls = toolCalls;
+				const responseDone =
+					jsonResponse.status === "completed" || jsonResponse.status === "done";
+				const finishReason = hasToolCalls
+					? "tool_calls"
+					: responseDone
+						? "stop"
+						: jsonResponse.status || "stop";
+				finalResp = {
+					id: jsonResponse.id || `chatcmpl-${Date.now()}`,
+					object: "chat.completion",
+					created: jsonResponse.created_at || Math.floor(Date.now() / 1000),
+					model: jsonResponse.model || model,
+					choices: [{ index: 0, message, finish_reason: finishReason }],
+					usage: {
+						prompt_tokens: inTokens,
+						completion_tokens: outTokens,
+						total_tokens: inTokens + outTokens,
+					},
+				};
+			}

-      return { success: true, response: new Response(JSON.stringify(finalResp), { headers: { "Content-Type": "application/json", "Access-Control-Allow-Origin": "*" } }) };
-    } catch (err) {
-      console.error("[ChatCore] Responses API SSE→JSON failed:", err);
-      return createErrorResult(HTTP_STATUS.BAD_GATEWAY, "Failed to convert streaming response to JSON");
-    }
-  }
+			return {
+				success: true,
+				response: new Response(JSON.stringify(finalResp), {
+					headers: {
+						"Content-Type": "application/json",
+						"Access-Control-Allow-Origin": "*",
+					},
+				}),
+			};
+		} catch (err) {
+			console.error("[ChatCore] Responses API SSE→JSON failed:", err);
+			return createErrorResult(
+				HTTP_STATUS.BAD_GATEWAY,
+				"Failed to convert streaming response to JSON",
+			);
+		}
+	}

-  // Standard Chat Completions SSE path
-  try {
-    const sseText = await providerResponse.text();
-    const parsed = parseSSEToOpenAIResponse(sseText, model);
-    if (!parsed) return createErrorResult(HTTP_STATUS.BAD_GATEWAY, "Invalid SSE response for non-streaming request");
+	// Standard Chat Completions SSE path
+	try {
+		const sseText = await providerResponse.text();
+		const parsed = parseSSEToOpenAIResponse(sseText, model);
+		if (!parsed)
+			return createErrorResult(
+				HTTP_STATUS.BAD_GATEWAY,
+				"Invalid SSE response for non-streaming request",
+			);
+		if (parsed.error) {
+			return createErrorResult(
+				HTTP_STATUS.BAD_GATEWAY,
+				parsed.error.message || "Upstream SSE stream failed",
+			);
+		}

-    if (onRequestSuccess) await onRequestSuccess();
+		// Config-driven in-stream error detection: the request "succeeded" at the
+		// HTTP level, but the content signals an upstream failure — treat it as an
+		// error so account/combo fallback and FAILED logging kick in.
+		const matchedPattern = matchStreamErrorPatterns(
+			streamErrorPatterns?.[provider],
+			parsed.choices?.[0]?.message?.content || "",
+		);
+		if (matchedPattern) {
+			return createErrorResult(
+				HTTP_STATUS.BAD_GATEWAY,
+				`Stream error pattern matched: ${matchedPattern}`,
+			);
+		}

-    const usage = parsed.usage || {};
-    appendLog({ tokens: usage, status: "200 OK" });
-    saveUsageStats({ provider, model, tokens: usage, connectionId, apiKey, endpoint: clientRawRequest?.endpoint });
+		if (onRequestSuccess) await onRequestSuccess();

-    const totalLatency = Date.now() - requestStartTime;
-    saveRequestDetail(buildRequestDetail({
-      ...ctx,
-      latency: { ttft: totalLatency, total: totalLatency },
-      tokens: usage,
-      response: {
-        content: parsed.choices?.[0]?.message?.content || null,
-        thinking: parsed.choices?.[0]?.message?.reasoning_content || null,
-        finish_reason: parsed.choices?.[0]?.finish_reason || "unknown"
-      },
-      status: "success"
-    }, { endpoint: clientRawRequest?.endpoint || null })).catch(() => {});
+		const usage = parsed.usage || {};
+		appendLog({ tokens: usage, status: "200 OK" });
+		saveUsageStats({
+			provider,
+			model,
+			tokens: usage,
+			connectionId,
+			apiKey,
+			endpoint: clientRawRequest?.endpoint,
+			silent: true,
+		});
+		if (log?.line)
+			log.line(
+				reqTag,
+				"📊",
+				formatDoneLine({
+					usage,
+					latency: { total: Date.now() - requestStartTime },
+				}),
+			);

-    // Strip reasoning_content only when content is non-empty.
-    // When content is empty (e.g. thinking models that used all tokens for reasoning),
-    // reasoning_content is the only useful output and must be preserved.
-    // Previously this was unconditional, which broke Qwen3.5, Claude extended thinking, etc.
-    if (parsed?.choices) {
-      for (const choice of parsed.choices) {
-        if (choice?.message?.reasoning_content && choice.message.content) {
-          delete choice.message.reasoning_content;
-        }
-      }
-    }
+		const totalLatency = Date.now() - requestStartTime;
+		saveRequestDetail(
+			buildRequestDetail(
+				{
+					...ctx,
+					latency: { ttft: totalLatency, total: totalLatency },
+					tokens: usage,
+					response: {
+						content: parsed.choices?.[0]?.message?.content || null,
+						thinking: parsed.choices?.[0]?.message?.reasoning_content || null,
+						finish_reason: parsed.choices?.[0]?.finish_reason || "unknown",
+					},
+					status: "success",
+				},
+				{ endpoint: clientRawRequest?.endpoint || null },
+			),
+		).catch(() => {});

-    return { success: true, response: new Response(JSON.stringify(parsed), { headers: { "Content-Type": "application/json", "Access-Control-Allow-Origin": "*" } }) };
-  } catch (err) {
-    console.error("[ChatCore] Chat Completions SSE→JSON failed:", err);
-    return createErrorResult(HTTP_STATUS.BAD_GATEWAY, "Failed to convert streaming response to JSON");
-  }
+		// Strip reasoning_content only when content is non-empty.
+		// When content is empty (e.g. thinking models that used all tokens for reasoning),
+		// reasoning_content is the only useful output and must be preserved.
+		// Previously this was unconditional, which broke Qwen3.5, Claude extended thinking, etc.
+		if (parsed?.choices) {
+			for (const choice of parsed.choices) {
+				if (choice?.message?.reasoning_content && choice.message.content) {
+					delete choice.message.reasoning_content;
+				}
+			}
+		}
+
+		return {
+			success: true,
+			response: new Response(JSON.stringify(parsed), {
+				headers: {
+					"Content-Type": "application/json",
+					"Access-Control-Allow-Origin": "*",
+				},
+			}),
+		};
+	} catch (err) {
+		console.error("[ChatCore] Chat Completions SSE→JSON failed:", err);
+		return createErrorResult(
+			HTTP_STATUS.BAD_GATEWAY,
+			"Failed to convert streaming response to JSON",
+		);
+	}
 }
--- a/open-sse/handlers/chatCore/streamingHandler.js
+++ b/open-sse/handlers/chatCore/streamingHandler.js
@@ -1,122 +1,336 @@
 import { FORMATS } from "../../translator/formats.js";
 import { needsTranslation } from "../../translator/index.js";
-import { createSSETransformStreamWithLogger, createPassthroughStreamWithLogger } from "../../utils/stream.js";
+import {
+	createSSETransformStreamWithLogger,
+	createPassthroughStreamWithLogger,
+} from "../../utils/stream.js";
 import { pipeWithDisconnect } from "../../utils/streamHandler.js";
 import { PROVIDERS } from "../../config/providers.js";
 import { STREAM_STALL_TIMEOUT_MS } from "../../config/runtimeConfig.js";
 import { buildAbortedResponsesTerminalBytes } from "../../utils/responsesStreamHelpers.js";
-import { buildRequestDetail, extractRequestConfig, saveUsageStats } from "./requestDetail.js";
+import {
+	buildRequestDetail,
+	extractRequestConfig,
+	saveUsageStats,
+	formatDoneLine,
+} from "./requestDetail.js";
+import { streamStatusForContent } from "../../utils/streamErrorPatterns.js";
 import { saveRequestDetail } from "@/lib/usageDb.js";
 import { SSE_HEADERS_CORS as SSE_HEADERS } from "../../utils/sseConstants.js";

 // Codex returns Responses API SSE → which client format to translate INTO, by request sourceFormat.
 // Gemini-family all map to ANTIGRAVITY decoder; unknown sources fall back to OPENAI.
 const CODEX_SOURCE_TO_TARGET = {
-  [FORMATS.OPENAI_RESPONSES]: FORMATS.OPENAI_RESPONSES,
-  [FORMATS.CLAUDE]: FORMATS.CLAUDE,
-  [FORMATS.ANTIGRAVITY]: FORMATS.ANTIGRAVITY,
-  [FORMATS.GEMINI]: FORMATS.ANTIGRAVITY,
-  [FORMATS.GEMINI_CLI]: FORMATS.ANTIGRAVITY,
+	[FORMATS.OPENAI_RESPONSES]: FORMATS.OPENAI_RESPONSES,
+	[FORMATS.CLAUDE]: FORMATS.CLAUDE,
+	[FORMATS.ANTIGRAVITY]: FORMATS.ANTIGRAVITY,
+	[FORMATS.GEMINI]: FORMATS.ANTIGRAVITY,
+	[FORMATS.GEMINI_CLI]: FORMATS.ANTIGRAVITY,
 };

 /**
 * Determine which SSE transform stream to use based on provider/format.
 */
-function buildTransformStream({ provider, sourceFormat, targetFormat, userAgent, reqLogger, toolNameMap, model, connectionId, body, onStreamComplete, apiKey }) {
-  const isDroidCLI = userAgent?.toLowerCase().includes("droid") || userAgent?.toLowerCase().includes("codex-cli");
-  // Responses-API providers (e.g. codex) emit Responses SSE → translate into client format
-  const isResponsesProvider = PROVIDERS[provider]?.format === FORMATS.OPENAI_RESPONSES;
-  const needsCodexTranslation = isResponsesProvider && targetFormat === FORMATS.OPENAI_RESPONSES && !isDroidCLI;
+function buildTransformStream({
+	provider,
+	sourceFormat,
+	targetFormat,
+	userAgent,
+	reqLogger,
+	toolNameMap,
+	model,
+	connectionId,
+	body,
+	onStreamComplete,
+	apiKey,
+}) {
+	const isDroidCLI =
+		userAgent?.toLowerCase().includes("droid") ||
+		userAgent?.toLowerCase().includes("codex-cli");
+	// Responses-API providers (e.g. codex) emit Responses SSE → translate into client format
+	const isResponsesProvider =
+		PROVIDERS[provider]?.format === FORMATS.OPENAI_RESPONSES;
+	const needsCodexTranslation =
+		isResponsesProvider &&
+		targetFormat === FORMATS.OPENAI_RESPONSES &&
+		!isDroidCLI;

-  if (needsCodexTranslation) {
-    const codexTarget = CODEX_SOURCE_TO_TARGET[sourceFormat] || FORMATS.OPENAI;
-    return createSSETransformStreamWithLogger(FORMATS.OPENAI_RESPONSES, codexTarget, provider, reqLogger, toolNameMap, model, connectionId, body, onStreamComplete, apiKey);
-  }
+	if (needsCodexTranslation) {
+		const codexTarget = CODEX_SOURCE_TO_TARGET[sourceFormat] || FORMATS.OPENAI;
+		return createSSETransformStreamWithLogger(
+			FORMATS.OPENAI_RESPONSES,
+			codexTarget,
+			provider,
+			reqLogger,
+			toolNameMap,
+			model,
+			connectionId,
+			body,
+			onStreamComplete,
+			apiKey,
+		);
+	}

-  if (needsTranslation(targetFormat, sourceFormat)) {
-    return createSSETransformStreamWithLogger(targetFormat, sourceFormat, provider, reqLogger, toolNameMap, model, connectionId, body, onStreamComplete, apiKey);
-  }
+	if (needsTranslation(targetFormat, sourceFormat)) {
+		return createSSETransformStreamWithLogger(
+			targetFormat,
+			sourceFormat,
+			provider,
+			reqLogger,
+			toolNameMap,
+			model,
+			connectionId,
+			body,
+			onStreamComplete,
+			apiKey,
+		);
+	}

-  return createPassthroughStreamWithLogger(provider, reqLogger, model, connectionId, body, onStreamComplete, apiKey);
+	return createPassthroughStreamWithLogger(
+		provider,
+		reqLogger,
+		model,
+		connectionId,
+		body,
+		onStreamComplete,
+		apiKey,
+	);
 }

 /**
 * Handle streaming response — pipe provider SSE through transform stream to client.
 */
-export function handleStreamingResponse({ providerResponse, provider, model, sourceFormat, targetFormat, userAgent, body, stream, translatedBody, finalBody, requestStartTime, connectionId, apiKey, clientRawRequest, onRequestSuccess, reqLogger, toolNameMap, streamController, onStreamComplete }) {
-  if (onRequestSuccess) {
-    Promise.resolve()
-      .then(onRequestSuccess)
-      .catch(err => {
-        console.error("[ChatCore] onRequestSuccess failed:", err?.message || err);
-      });
-  }
+export async function handleStreamingResponse({
+	providerResponse,
+	provider,
+	model,
+	sourceFormat,
+	targetFormat,
+	userAgent,
+	body,
+	stream,
+	translatedBody,
+	finalBody,
+	requestStartTime,
+	connectionId,
+	apiKey,
+	clientRawRequest,
+	onRequestSuccess,
+	reqLogger,
+	toolNameMap,
+	streamController,
+	onStreamComplete,
+	streamDetailId,
+	pxpipe,
+	reqTag,
+	log,
+}) {
+	if (onRequestSuccess) {
+		Promise.resolve()
+			.then(onRequestSuccess)
+			.catch((err) => {
+				console.error(
+					"[ChatCore] onRequestSuccess failed:",
+					err?.message || err,
+				);
+			});
+	}

-  // Warn when upstream returns unexpected Content-Type for a streaming response.
-  // This often means the provider returned an HTML error page or plain-text error
-  // that the SSE transform stream would forward as garbage to the client.
-  const upstreamContentType = (providerResponse.headers.get('content-type') || '').toLowerCase();
-  if (upstreamContentType && !upstreamContentType.includes('text/event-stream') && !upstreamContentType.includes('application/json')) {
-    console.warn('[STREAM] ' + provider + ' | ' + model + ' | unexpected Content-Type: ' + upstreamContentType);
-  }
+	// When upstream returns HTML/text instead of SSE (e.g. Cloudflare 5xx error
+	// page), piping it through the SSE transform stream causes Next.js
+	// "failed to pipe response" and crashes the chat router. Read the body,
+	// pull a short human-readable message from the <title>, sanitize it, and
+	// return a clean JSON error instead. The message is stripped of HTML tags
+	// and clamped so untrusted upstream text never reaches the client verbatim
+	// (the UI may render error.message as HTML).
+	const upstreamContentType = (
+		providerResponse.headers.get("content-type") || ""
+	).toLowerCase();
+	if (
+		upstreamContentType &&
+		!upstreamContentType.includes("text/event-stream") &&
+		!upstreamContentType.includes("application/json")
+	) {
+		const bodyText = await providerResponse.text().catch(() => "");
+		const titleMatch = bodyText.match(/<title>([^<]+)<\/title>/i);
+		const sanitizedTitle = (titleMatch?.[1] || "")
+			.replace(/<[^>]*>/g, "")
+			.replace(/[\r\n]+/g, " ")
+			.trim()
+			.slice(0, 160);
+		const shortMsg =
+			sanitizedTitle ||
+			(bodyText.length < 200
+				? bodyText
+						.replace(/<[^>]*>/g, "")
+						.trim()
+						.slice(0, 160)
+				: `Upstream returned non-SSE response (${upstreamContentType})`);
+		const status = providerResponse.status || 502;
+		if (log?.errorLine)
+			log.errorLine(
+				reqTag,
+				"✗",
+				`BLOCKED ${status} · ${provider}/${model} · non-SSE (${upstreamContentType})\n    ${shortMsg}`,
+			);
+		else
+			console.warn(
+				`[STREAM] ${provider} | ${model} | blocked pipe: ${shortMsg} [${status}]`,
+			);
+		streamController?.handleError?.(new Error(`upstream non-SSE: ${status}`));
+		return {
+			success: false,
+			response: new Response(
+				JSON.stringify({ error: { message: `[${status}]: ${shortMsg}` } }),
+				{
+					status,
+					headers: {
+						"Content-Type": "application/json",
+						"Access-Control-Allow-Origin": "*",
+					},
+				},
+			),
+		};
+	}

-  const transformStream = buildTransformStream({ provider, sourceFormat, targetFormat, userAgent, reqLogger, toolNameMap, model, connectionId, body, onStreamComplete, apiKey });
+	const transformStream = buildTransformStream({
+		provider,
+		sourceFormat,
+		targetFormat,
+		userAgent,
+		reqLogger,
+		toolNameMap,
+		model,
+		connectionId,
+		body,
+		onStreamComplete,
+		apiKey,
+	});

-  // Responses passthrough: synthesize response.failed + [DONE] if the stream aborts/stalls before a terminal event
-  const isResponsesPassthrough = sourceFormat === FORMATS.OPENAI_RESPONSES && targetFormat === FORMATS.OPENAI_RESPONSES;
-  const onAbortTerminal = isResponsesPassthrough ? buildAbortedResponsesTerminalBytes : null;
-  const stallTimeoutMs = PROVIDERS[provider]?.stallTimeoutMs || STREAM_STALL_TIMEOUT_MS;
-  const transformedBody = pipeWithDisconnect(providerResponse, transformStream, streamController, onAbortTerminal, stallTimeoutMs);
+	// Responses passthrough: synthesize response.failed + [DONE] if the stream aborts/stalls before a terminal event
+	const isResponsesPassthrough =
+		sourceFormat === FORMATS.OPENAI_RESPONSES &&
+		targetFormat === FORMATS.OPENAI_RESPONSES;
+	const onAbortTerminal = isResponsesPassthrough
+		? buildAbortedResponsesTerminalBytes
+		: null;
+	const stallTimeoutMs =
+		PROVIDERS[provider]?.stallTimeoutMs || STREAM_STALL_TIMEOUT_MS;
+	const transformedBody = pipeWithDisconnect(
+		providerResponse,
+		transformStream,
+		streamController,
+		onAbortTerminal,
+		stallTimeoutMs,
+	);

-  const streamDetailId = `${Date.now()}-${Math.random().toString(36).slice(2, 11)}`;
-  saveRequestDetail(buildRequestDetail({
-    provider, model, connectionId,
-    latency: { ttft: 0, total: Date.now() - requestStartTime },
-    tokens: { prompt_tokens: 0, completion_tokens: 0 },
-    request: extractRequestConfig(body, stream),
-    providerRequest: finalBody || translatedBody || null,
-    providerResponse: "[Streaming - raw response not captured]",
-    response: { content: "[Streaming in progress...]", thinking: null, type: "streaming" },
-    status: "success"
-  }, { id: streamDetailId })).catch(err => {
-    console.error("[RequestDetail] Failed to save streaming request:", err.message);
-  });
+	saveRequestDetail(
+		buildRequestDetail(
+			{
+				provider,
+				model,
+				connectionId,
+				latency: { ttft: 0, total: Date.now() - requestStartTime },
+				tokens: { prompt_tokens: 0, completion_tokens: 0 },
+				request: extractRequestConfig(body, stream),
+				providerRequest: finalBody || translatedBody || null,
+				providerResponse: "[Streaming - raw response not captured]",
+				response: {
+					content: "[Streaming in progress...]",
+					thinking: null,
+					type: "streaming",
+				},
+				pxpipe,
+				status: "success",
+			},
+			{ id: streamDetailId },
+		),
+	).catch((err) => {
+		console.error(
+			"[RequestDetail] Failed to save streaming request:",
+			err.message,
+		);
+	});

-  return {
-    success: true,
-    response: new Response(transformedBody, { headers: SSE_HEADERS })
-  };
+	return {
+		success: true,
+		response: new Response(transformedBody, { headers: SSE_HEADERS }),
+	};
 }

 /**
 * Build onStreamComplete callback for streaming usage tracking.
 */
-export function buildOnStreamComplete({ provider, model, connectionId, apiKey, requestStartTime, body, stream, finalBody, translatedBody, clientRawRequest }) {
-  const streamDetailId = `${Date.now()}-${Math.random().toString(36).slice(2, 11)}`;
+export function buildOnStreamComplete({
+	provider,
+	model,
+	connectionId,
+	apiKey,
+	requestStartTime,
+	body,
+	stream,
+	finalBody,
+	translatedBody,
+	clientRawRequest,
+	pxpipe,
+	reqTag,
+	log,
+	streamErrorPatterns,
+}) {
+	const streamDetailId = `${Date.now()}-${Math.random().toString(36).slice(2, 11)}`;

-  const onStreamComplete = (contentObj, usage, ttftAt) => {
-    const latency = {
-      ttft: ttftAt ? ttftAt - requestStartTime : Date.now() - requestStartTime,
-      total: Date.now() - requestStartTime
-    };
-    const safeContent = contentObj?.content || "[Empty streaming response]";
-    const safeThinking = contentObj?.thinking || null;
+	const onStreamComplete = (contentObj, usage, ttftAt) => {
+		const latency = {
+			ttft: ttftAt ? ttftAt - requestStartTime : Date.now() - requestStartTime,
+			total: Date.now() - requestStartTime,
+		};
+		const safeContent = contentObj?.content || "[Empty streaming response]";
+		const safeThinking = contentObj?.thinking || null;

-    saveRequestDetail(buildRequestDetail({
-      provider, model, connectionId,
-      latency,
-      tokens: usage || { prompt_tokens: 0, completion_tokens: 0 },
-      request: extractRequestConfig(body, stream),
-      providerRequest: finalBody || translatedBody || null,
-      providerResponse: safeContent,
-      response: { content: safeContent, thinking: safeThinking, type: "streaming" },
-      status: "success"
-    }, { id: streamDetailId })).catch(err => {
-      console.error("[RequestDetail] Failed to update streaming content:", err.message);
-    });
+		saveRequestDetail(
+			buildRequestDetail(
+				{
+					provider,
+					model,
+					connectionId,
+					latency,
+					tokens: usage || { prompt_tokens: 0, completion_tokens: 0 },
+					request: extractRequestConfig(body, stream),
+					providerRequest: finalBody || translatedBody || null,
+					providerResponse: safeContent,
+					response: {
+						content: safeContent,
+						thinking: safeThinking,
+						type: "streaming",
+					},
+					pxpipe,
+					status: streamStatusForContent(
+						streamErrorPatterns?.[provider],
+						safeContent,
+					),
+				},
+				{ id: streamDetailId },
+			),
+		).catch((err) => {
+			console.error(
+				"[RequestDetail] Failed to update streaming content:",
+				err.message,
+			);
+		});

-    saveUsageStats({ provider, model, tokens: usage, connectionId, apiKey, endpoint: clientRawRequest?.endpoint, label: "STREAM USAGE" });
-  };
+		// Persist stream usage to DB (no console line; the "📊 done" line below is authoritative)
+		saveUsageStats({
+			provider,
+			model,
+			tokens: usage,
+			connectionId,
+			apiKey,
+			endpoint: clientRawRequest?.endpoint,
+			label: "STREAM USAGE",
+			silent: true,
+		});
+		if (log?.line) log.line(reqTag, "📊", formatDoneLine({ usage, latency }));
+	};

-  return { onStreamComplete, streamDetailId };
+	return { onStreamComplete, streamDetailId };
 }
--- a/open-sse/handlers/embeddingsCore.js
+++ b/open-sse/handlers/embeddingsCore.js
@@ -116,6 +116,7 @@ export async function handleEmbeddingsCore({

  return {
    success: true,
+    usage: normalized.usage || null,
    response: new Response(JSON.stringify(normalized), {
      headers: {
        "Content-Type": "application/json",
--- a/open-sse/handlers/fetch/index.js
+++ b/open-sse/handlers/fetch/index.js
@@ -49,7 +49,10 @@ function truncate(text, max) {
 }

 function parseJinaTitle(text) {
-  const m = String(text || "").match(/^\s*#\s+(.+)$/m);
+  const source = String(text || "");
+  const metadataTitle = source.match(/^\s*Title:\s*(.+)$/mi);
+  if (metadataTitle) return metadataTitle[1].trim();
+  const m = source.match(/^\s*#\s+(.+)$/m);
  return m ? m[1].trim() : null;
 }

@@ -151,11 +154,14 @@ async function runFirecrawl({ url, fmt, timeoutMs, apiKey, maxCharacters, costPe
 }

 async function runJina({ url, fmt, timeoutMs, apiKey, maxCharacters, costPerQuery, startedAt }) {
-  const target = `https://r.jina.ai/${encodeURIComponent(url)}`;
  const upstreamStart = Date.now();
-  const r = await tryFetch(target, {
-    method: "GET",
-    headers: apiKey ? { authorization: `Bearer ${apiKey}` } : {}
+  const r = await tryFetch("https://r.jina.ai/", {
+    method: "POST",
+    headers: {
+      "content-type": "application/json",
+      ...(apiKey ? { authorization: `Bearer ${apiKey}` } : {})
+    },
+    body: JSON.stringify({ url })
  }, timeoutMs);

  if (!r.ok) {
--- a/open-sse/handlers/imageGenerationCore.js
+++ b/open-sse/handlers/imageGenerationCore.js
@@ -96,7 +96,7 @@ export async function handleImageGenerationCore({
  let requestBody;

  try {
-    url = adapter.buildUrl(model, credentials);
+    url = adapter.buildUrl(model, credentials, body);
    requestBody = await adapter.buildBody(model, body);
    headers = adapter.buildHeaders(credentials, requestBody, model, body);
  } catch (error) {
@@ -140,7 +140,7 @@ export async function handleImageGenerationCore({
      try {
        const retryBody = await adapter.buildBody(model, body);
        const retryHeaders = adapter.buildHeaders(credentials, retryBody, model, body);
-        const retryUrl = adapter.buildUrl(model, credentials);
+        const retryUrl = adapter.buildUrl(model, credentials, body);
        providerResponse = await fetch(retryUrl, {
          method: "POST",
          headers: retryHeaders,
--- a/open-sse/handlers/imageProviders/index.js
+++ b/open-sse/handlers/imageProviders/index.js
@@ -12,6 +12,7 @@ import blackForestLabs from "./blackForestLabs.js";
 import runwayml from "./runwayml.js";
 import cloudflareAi from "./cloudflareAi.js";
 import antigravity from "./antigravity.js";
+import xai from "./xai.js";

 const ADAPTERS = {
  openai: createOpenAIAdapter("openai"),
@@ -19,7 +20,7 @@ const ADAPTERS = {
  openrouter: createOpenAIAdapter("openrouter"),
  recraft: createOpenAIAdapter("recraft"),
  "vercel-ai-gateway": createOpenAIAdapter("vercel-ai-gateway"),
-  xai: createOpenAIAdapter("xai"),
+  xai,
  gemini,
  codex,
  sdwebui,
--- a/open-sse/handlers/imageProviders/xai.js
+++ b/open-sse/handlers/imageProviders/xai.js
@@ -0,0 +1,137 @@
+// xAI Grok Imagine — text-to-image + single/multi image editing
+// Docs:
+//   https://docs.x.ai/developers/model-capabilities/images/generation
+//   https://docs.x.ai/developers/model-capabilities/images/editing
+//   https://docs.x.ai/developers/model-capabilities/images/multi-image-editing
+import { sizeToAspectRatio } from "./_base.js";
+import { PROVIDER_MEDIA } from "../../providers/index.js";
+
+const IMG_CFG = PROVIDER_MEDIA["xai"]?.imageConfig || {};
+const GENERATIONS_URL = IMG_CFG.baseUrl || "https://api.x.ai/v1/images/generations";
+const EDITS_URL = IMG_CFG.editsUrl || "https://api.x.ai/v1/images/edits";
+
+const ASPECT_RATIOS = new Set([
+  "auto",
+  "1:1",
+  "16:9",
+  "9:16",
+  "4:3",
+  "3:2",
+  "2:3",
+  "9:19.5",
+  "20:9",
+]);
+
+function hasEditInput(body) {
+  if (!body || typeof body !== "object") return false;
+  if (body.image) return true;
+  return Array.isArray(body.images) && body.images.some(Boolean);
+}
+
+/** Normalize client image input → xAI image ref object */
+function toXaiImageRef(input) {
+  if (!input) return null;
+
+  if (typeof input === "object") {
+    // Already xAI-shaped or partial
+    if (input.file_id) {
+      return {
+        type: input.type || "image_url",
+        file_id: input.file_id,
+        ...(input.url ? { url: input.url } : {}),
+      };
+    }
+    if (input.url) {
+      return { type: input.type || "image_url", url: input.url };
+    }
+    return null;
+  }
+
+  if (typeof input !== "string") return null;
+  const trimmed = input.trim();
+  if (!trimmed) return null;
+
+  // Public URL or data URI
+  if (/^https?:\/\//i.test(trimmed) || /^data:image\//i.test(trimmed)) {
+    return { type: "image_url", url: trimmed };
+  }
+
+  // Raw base64 → data URI
+  return { type: "image_url", url: `data:image/png;base64,${trimmed}` };
+}
+
+function collectImageRefs(body) {
+  const refs = [];
+  if (Array.isArray(body.images)) {
+    for (const item of body.images) {
+      const ref = toXaiImageRef(item);
+      if (ref) refs.push(ref);
+    }
+  }
+  if (body.image) {
+    const ref = toXaiImageRef(body.image);
+    if (ref) refs.push(ref);
+  }
+  // xAI multi-edit supports up to 3 source images
+  return refs.slice(0, 3);
+}
+
+function resolveAspectRatio(body) {
+  if (typeof body.aspect_ratio === "string" && body.aspect_ratio.trim()) {
+    const ratio = body.aspect_ratio.trim();
+    if (ASPECT_RATIOS.has(ratio)) return ratio;
+    // Pass through unknown ratio strings (upstream will validate)
+    return ratio;
+  }
+  // OpenAI-style size → aspect ratio (skip auto)
+  if (body.size && body.size !== "auto") {
+    return sizeToAspectRatio(body.size);
+  }
+  return undefined;
+}
+
+function resolveResolution(body) {
+  if (typeof body.resolution !== "string") return undefined;
+  const value = body.resolution.trim().toLowerCase();
+  if (!value || value === "auto") return undefined;
+  return value; // "1k" | "2k"
+}
+
+export default {
+  buildUrl: (_model, _credentials, body) => (hasEditInput(body) ? EDITS_URL : GENERATIONS_URL),
+
+  buildHeaders: (creds) => {
+    const headers = { "Content-Type": "application/json", ...(IMG_CFG.headers || {}) };
+    const key = creds?.apiKey || creds?.accessToken;
+    if (key) headers["Authorization"] = `Bearer ${key}`;
+    return headers;
+  },
+
+  buildBody: (model, body) => {
+    const req = {
+      model,
+      prompt: body.prompt,
+    };
+
+    if (body.n != null) req.n = body.n;
+    if (body.response_format) req.response_format = body.response_format;
+
+    const aspectRatio = resolveAspectRatio(body);
+    if (aspectRatio) req.aspect_ratio = aspectRatio;
+
+    const resolution = resolveResolution(body);
+    if (resolution) req.resolution = resolution;
+
+    const refs = collectImageRefs(body);
+    if (refs.length === 1) {
+      req.image = refs[0];
+    } else if (refs.length > 1) {
+      req.images = refs;
+    }
+
+    return req;
+  },
+
+  // xAI already returns OpenAI-compatible { created, data: [{ url | b64_json }] }
+  normalize: (responseBody) => responseBody,
+};
--- a/open-sse/handlers/search/callers.js
+++ b/open-sse/handlers/search/callers.js
@@ -1,7 +1,6 @@
 /**
 * Search Provider Request Builders
 *
- * Ported from OmniRoute open-sse/handlers/search.ts (lines 223-610).
 * Builds HTTP request `{ url, init }` for 10 search providers.
 *
 * @typedef {Object} SearchProviderConfig
--- a/open-sse/handlers/search/chatSearch.js
+++ b/open-sse/handlers/search/chatSearch.js
@@ -273,6 +273,53 @@ const CHAT_SEARCH_CONFIG = {
      const tokens = data?.usage?.total_tokens || 0;
      return { text, citations, tokens };
    }
+  },
+
+  "perplexity-agent": {
+    endpoint: () => searchEndpoint("perplexity-agent"),
+    buildBody: (query, model) => ({
+      model,
+      input: query,
+      tools: [{ type: "web_search" }]
+    }),
+    buildHeaders: (token) => ({
+      "Content-Type": "application/json",
+      Authorization: `Bearer ${token}`
+    }),
+    extractAnswer: (data) => {
+      const output = Array.isArray(data?.output) ? data.output : [];
+      let text = "";
+      const citations = [];
+      for (const item of output) {
+        const parts = Array.isArray(item?.content) ? item.content : [];
+        for (const p of parts) {
+          if (typeof p?.text === "string") text += p.text;
+          const anns = Array.isArray(p?.annotations) ? p.annotations : [];
+          for (const a of anns) {
+            const c = normalizeCitation(a?.url ? a : a?.url_citation);
+            if (c) citations.push(c);
+          }
+        }
+        const results = Array.isArray(item?.results) ? item.results : [];
+        for (const r of results) {
+          const url = r?.url || r?.link;
+          if (!url) continue;
+          citations.push({
+            url,
+            title: r?.title || "",
+            snippet: r?.snippet || ""
+          });
+        }
+      }
+      if (!citations.length && Array.isArray(data?.citations)) {
+        for (const c of data.citations) {
+          const n = normalizeCitation(c);
+          if (n) citations.push(n);
+        }
+      }
+      const tokens = data?.usage?.total_tokens || 0;
+      return { text, citations, tokens };
+    }
  }
 };

--- a/open-sse/handlers/search/normalizers.js
+++ b/open-sse/handlers/search/normalizers.js
@@ -1,7 +1,6 @@
 /**
 * Search Response Normalizers
 *
- * Ported from OmniRoute open-sse/handlers/search.ts.
 * Each normalizer maps a provider-specific response into the unified SearchResult shape.
 */

--- a/open-sse/handlers/videoCore.js
+++ b/open-sse/handlers/videoCore.js
@@ -0,0 +1,166 @@
+import { createErrorResult } from "../utils/error.js";
+import { HTTP_STATUS } from "../config/runtimeConfig.js";
+import { refreshTokenByProvider } from "../services/tokenRefresh.js";
+import { PROVIDER_MEDIA } from "../providers/index.js";
+
+// Upstream fetch deadline for video job submission/polling (the job itself is
+// async upstream — this only bounds the HTTP round-trip, not video rendering).
+const VIDEO_FETCH_TIMEOUT_MS = Number(process.env.VIDEO_FETCH_TIMEOUT_MS || 120000);
+
+// POST /videos/* creates a billable upstream job. A network error after the
+// request left the socket may still have created the job, so creation is NEVER
+// auto-retried (the only re-send is the auth retry after a 401/403 refresh,
+// which upstream rejects before job creation).
+export const VIDEO_ACTIONS = new Set(["generations", "edits", "extensions"]);
+
+export function getVideoConfig(provider) {
+  return PROVIDER_MEDIA[provider]?.videoConfig || null;
+}
+
+/** Strip bearer tokens / obvious secrets from text destined for clients or logs. */
+export function sanitizeSecrets(text, credentials = null) {
+  if (!text) return text;
+  let out = String(text).replace(/Bearer\s+[A-Za-z0-9._~+/=-]{8,}/gi, "Bearer [redacted]");
+  for (const key of ["accessToken", "refreshToken", "apiKey"]) {
+    const secret = credentials?.[key];
+    if (typeof secret === "string" && secret.length >= 8) {
+      out = out.split(secret).join("[redacted]");
+    }
+  }
+  return out;
+}
+
+function buildUpstreamUrl(config, action, requestId) {
+  const base = config.baseUrl.replace(/\/$/, "");
+  return requestId ? `${base}/${encodeURIComponent(requestId)}` : `${base}/${action}`;
+}
+
+function buildHeaders({ token, contentType, idempotencyKey }) {
+  const headers = { Accept: "application/json" };
+  if (token) headers.Authorization = `Bearer ${token}`;
+  if (contentType) headers["Content-Type"] = contentType;
+  if (idempotencyKey) headers["Idempotency-Key"] = idempotencyKey;
+  return headers;
+}
+
+function combineSignals(signal, timeoutMs) {
+  const timeoutSignal = typeof AbortSignal?.timeout === "function" ? AbortSignal.timeout(timeoutMs) : null;
+  if (signal && timeoutSignal && typeof AbortSignal.any === "function") {
+    return AbortSignal.any([signal, timeoutSignal]);
+  }
+  return signal || timeoutSignal || undefined;
+}
+
+/**
+ * Transparent proxy for async video jobs (xAI Grok Imagine shape).
+ *
+ * - Forwards the raw body byte-for-byte (JSON or multipart) — no reshaping.
+ * - Passes upstream JSON (request_id, status, video.url, error) back verbatim.
+ * - 401/403 with a refresh token: refresh ONCE, retry ONCE. No other retry.
+ * - Upstream error text is sanitized before it reaches the client.
+ *
+ * @param {object} options
+ * @param {string} options.provider - Provider id (must have registry videoConfig)
+ * @param {"generations"|"edits"|"extensions"|null} options.action - Creation action (POST)
+ * @param {string|null} [options.requestId] - Poll target (GET /videos/{id})
+ * @param {Buffer|string|null} [options.rawBody] - Exact body to forward
+ * @param {string|null} [options.contentType] - Original Content-Type header
+ * @param {string|null} [options.idempotencyKey] - Forwarded Idempotency-Key
+ * @param {object} options.credentials - { accessToken?, apiKey?, refreshToken?, authType? }
+ * @param {AbortSignal} [options.signal] - Client cancellation signal
+ * @param {number} [options.timeoutMs]
+ * @param {object} [options.log]
+ * @param {function} [options.onCredentialsRefreshed]
+ * @returns {Promise<{ success: boolean, response: Response, status?: number, error?: string }>}
+ */
+export async function handleVideoProxyCore({
+  provider,
+  action = null,
+  requestId = null,
+  rawBody = null,
+  contentType = null,
+  idempotencyKey = null,
+  credentials,
+  signal,
+  timeoutMs = VIDEO_FETCH_TIMEOUT_MS,
+  log,
+  onCredentialsRefreshed,
+}) {
+  const config = getVideoConfig(provider);
+  if (!config) {
+    return createErrorResult(HTTP_STATUS.BAD_REQUEST, `Provider '${provider}' does not support video generation`);
+  }
+  if (!requestId && !VIDEO_ACTIONS.has(action)) {
+    return createErrorResult(HTTP_STATUS.BAD_REQUEST, `Unknown video action: ${action}`);
+  }
+
+  const method = requestId ? "GET" : "POST";
+  const url = buildUpstreamUrl(config, action, requestId);
+  const fetchSignal = combineSignals(signal, timeoutMs);
+
+  const doFetch = (token) =>
+    fetch(url, {
+      method,
+      headers: buildHeaders({ token, contentType: method === "POST" ? contentType : null, idempotencyKey: method === "POST" ? idempotencyKey : null }),
+      body: method === "POST" ? rawBody : undefined,
+      signal: fetchSignal,
+    });
+
+  let upstream;
+  try {
+    upstream = await doFetch(credentials?.accessToken || credentials?.apiKey);
+  } catch (error) {
+    if (error?.name === "AbortError" || error?.name === "TimeoutError") {
+      return createErrorResult(HTTP_STATUS.REQUEST_TIMEOUT, `[${provider}] video ${method} aborted: ${error.message}`);
+    }
+    // Never re-send a creation POST on network error — the job may already exist upstream.
+    return createErrorResult(HTTP_STATUS.BAD_GATEWAY, sanitizeSecrets(`[${provider}] video upstream fetch failed: ${error.message}`, credentials));
+  }
+
+  // 401/403 → refresh once → retry once (OAuth accounts only; API keys can't refresh)
+  if (
+    (upstream.status === HTTP_STATUS.UNAUTHORIZED || upstream.status === HTTP_STATUS.FORBIDDEN) &&
+    credentials?.refreshToken
+  ) {
+    let refreshed = null;
+    try {
+      refreshed = await refreshTokenByProvider(provider, credentials, log);
+    } catch (error) {
+      log?.warn?.("TOKEN", `${provider} | video refresh error: ${sanitizeSecrets(error.message, credentials)}`);
+    }
+    if (refreshed?.accessToken) {
+      log?.info?.("TOKEN", `${provider.toUpperCase()} | refreshed for video ${method}`);
+      Object.assign(credentials, refreshed);
+      if (onCredentialsRefreshed) await onCredentialsRefreshed(refreshed);
+      try {
+        await upstream.body?.cancel?.();
+      } catch { /* noop */ }
+      try {
+        upstream = await doFetch(credentials.accessToken || credentials.apiKey);
+      } catch (error) {
+        return createErrorResult(HTTP_STATUS.BAD_GATEWAY, sanitizeSecrets(`[${provider}] video retry after refresh failed: ${error.message}`, credentials));
+      }
+    } else {
+      log?.warn?.("TOKEN", `${provider.toUpperCase()} | video refresh failed — account needs re-auth`);
+    }
+  }
+
+  const bodyText = await upstream.text().catch(() => "");
+
+  if (!upstream.ok) {
+    const message = sanitizeSecrets(bodyText || `HTTP ${upstream.status}`, credentials);
+    return createErrorResult(upstream.status, `[${provider}] ${message.slice(0, 2000)}`);
+  }
+
+  // Success: pass the upstream JSON through untouched (request_id / status / video.url).
+  return {
+    success: true,
+    response: new Response(bodyText, {
+      status: upstream.status,
+      headers: {
+        "Content-Type": upstream.headers.get("content-type") || "application/json",
+        "Access-Control-Allow-Origin": "*",
+      },
+    }),
+  };
+}
--- a/open-sse/providers/capabilities.js
+++ b/open-sse/providers/capabilities.js
@@ -71,7 +71,11 @@ export function capabilitiesFromServiceKind(kind) {
 * otherwise mis-match. Only declare deltas vs DEFAULT.
 */
 export const MODEL_CAPABILITIES = {
-  // Claude 4.6/4.7/4.8 have 1M context + adaptive thinking (override generic claude pattern)
+  // Claude Opus 5, 4.6/4.7/4.8, and Kiro Sonnet 5 have 1M context + adaptive thinking (override generic claude pattern)
+  "claude-opus-5":     { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-opus-5-thinking": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-opus-5-agentic": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-opus-5-thinking-agentic": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
  "claude-opus-4.6":   { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
  "claude-opus-4.7":   { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
  "claude-opus-4-7":   { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
@@ -82,6 +86,10 @@ export const MODEL_CAPABILITIES = {
  "claude-opus-4-8-thinking": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
  "claude-sonnet-4.6": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
  "claude-sonnet-4-6": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-sonnet-5": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-sonnet-5-thinking": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-sonnet-5-agentic": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },
+  "claude-sonnet-5-thinking-agentic": { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 },

  // Gemini image-gen / OpenAI image / xai image variants
  "gpt-image-1":       { imageOutput: true, tools: false },
@@ -92,12 +100,58 @@ export const MODEL_CAPABILITIES = {
  // Qwen plain coder/text (no vision) — registry "vision-model" / "coder-model" aliases
  "vision-model":      { vision: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000 },
  "coder-model":       { reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000 },
+
+  // Kimi flagship + coding (platform + Kimi Code ids) — vision/video native
+  "kimi-k3":           { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 1048576, maxOutput: 131072 },
+  "k3":                { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 1048576, maxOutput: 131072 },
+  "kimi-for-coding":   { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 },
+  "kimi-for-coding-highspeed": { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 },
+  "kimi-k2.7-code":    { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 },
+  "kimi-k2.7-code-highspeed": { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 },
 };

+const KIRO_GPT_5_6_CAPABILITIES = { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 272000, maxOutput: 128000 };
+
+// Codex OAuth (ChatGPT backend) — per-model context window reported by upstream
+// (lower than OpenAI API's 1.05M). Sol differs from Terra/Luna. #2720
+const CODEX_GPT_56_SOL_CAPS  = { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 372000, maxOutput: 128000 };
+const CODEX_GPT_56_DEFAULT_CAPS = { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 272000, maxOutput: 128000 };
+
 /**
 * Provider-specific capability overrides. Keyed by provider alias/id.
 */
 export const PROVIDER_CAPABILITIES = {
+  // NVIDIA NIM is OpenAI-compatible → rejects MiniMax/GLM native `thinking` field.
+  // Force openai reasoning_effort format for its reasoning models. #issue
+  "nvidia": {
+    "minimaxai/minimax-m2.7": { reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 200000, maxOutput: 131072 },
+    "minimaxai/minimax-m3": { vision: true, reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 512000, maxOutput: 131072 },
+    "z-ai/glm-5.2": { reasoning: true, thinkingFormat: "openai", contextWindow: 200000, maxOutput: 128000 },
+    "deepseek-ai/deepseek-v4-pro": { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 65536 },
+    "deepseek-ai/deepseek-v4-flash": { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 65536 },
+  },
+  "codex": {
+    "gpt-5.6-sol":               CODEX_GPT_56_SOL_CAPS,
+    "gpt-5.6-sol-review":        CODEX_GPT_56_SOL_CAPS,
+    "gpt-5.6-terra":             CODEX_GPT_56_DEFAULT_CAPS,
+    "gpt-5.6-terra-review":      CODEX_GPT_56_DEFAULT_CAPS,
+    "gpt-5.6-luna":              CODEX_GPT_56_DEFAULT_CAPS,
+    "gpt-5.6-luna-review":       CODEX_GPT_56_DEFAULT_CAPS,
+  },
+  "kiro": {
+    "gpt-5.6-sol": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-terra": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-luna": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-sol-thinking": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-terra-thinking": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-luna-thinking": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-sol-agentic": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-terra-agentic": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-luna-agentic": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-sol-thinking-agentic": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-terra-thinking-agentic": KIRO_GPT_5_6_CAPABILITIES,
+    "gpt-5.6-luna-thinking-agentic": KIRO_GPT_5_6_CAPABILITIES,
+  },
  // CodeBuddy.cn — authoritative per-model metadata from the gateway's model
  // config (contextWindow=maxInputTokens, maxOutput=maxOutputTokens, vision=
  // supportsImages). Every model reasons via OpenAI-style reasoning_effort
@@ -120,6 +174,11 @@ export const PROVIDER_CAPABILITIES = {
    "deepseek-v4-flash":  { vision: true, reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 1000000, maxOutput: 50000 },
    "deepseek-v3-2-volc": { reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 96000, maxOutput: 32000 },
  },
+  // Poolside Laguna — OpenAI-compatible, all reasoning-capable (32K max output).
+  "poolside": {
+    "laguna-s-2.1":  { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 32000 },
+    "laguna-xs-2.1": { reasoning: true, thinkingFormat: "openai", contextWindow: 200000, maxOutput: 32000 },
+  },
 };

 /**
@@ -130,6 +189,7 @@ export const PROVIDER_CAPABILITIES = {
 */
 export const PATTERN_CAPABILITIES = [
  // ── Claude (4.6+ = adaptive thinking; older/haiku = budget) ──────
+  { pattern: "*claude*opus-5*",     caps: { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive", contextWindow: 1000000, maxOutput: 128000 } },
  { pattern: "*claude*opus-4.6*",   caps: { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive" } },
  { pattern: "*claude*opus-4.7*",   caps: { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive" } },
  { pattern: "*claude*opus-4.8*",   caps: { vision: true, reasoning: true, search: true, thinkingFormat: "claude-adaptive" } },
@@ -173,21 +233,29 @@ export const PATTERN_CAPABILITIES = [
  // ── Grok (vision + Live Search) ──────────────────────────────────
  { pattern: "*grok*image*",    caps: { imageOutput: true } },
  { pattern: "*grok-code*",     caps: { reasoning: true, thinkingFormat: "openai", contextWindow: 256000 } },
+  // Grok 4.5 (Grok CLI / Grok Build): 500k context per cli-chat-proxy /v1/models
+  { pattern: "*grok-4.5*",      caps: { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 500000, maxOutput: 64000 } },
  { pattern: "*grok-4*",        caps: { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 256000 } },
  { pattern: "*grok-3*",        caps: { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 131072 } },
  { pattern: "*grok*",          caps: { vision: true, reasoning: true, search: true, thinkingFormat: "openai", contextWindow: 256000 } },

-  // ── Qwen (enable_thinking + thinking_budget; QwQ = thinking-only) ─
+  // ── Qwen (3.5+ = native vision/video; coder & max = text-only; QwQ = thinking-only) ─
  { pattern: "*qwen*vl*",       caps: { vision: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 262144 } },
-  { pattern: "*qwen*max*",      caps: { vision: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
+  { pattern: "*qwen*omni*",     caps: { vision: true, audioInput: true, videoInput: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 262144, maxOutput: 65536 } },
+  { pattern: "*qwen*coder*",    caps: { reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000 } },
+  { pattern: "*qwen*max*",      caps: { reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
+  { pattern: "*qwen3.5*",       caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
+  { pattern: "*qwen3.6*",       caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
+  { pattern: "*qwen3.7*",       caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
  { pattern: "*qwen*plus*",     caps: { vision: true, reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000, maxOutput: 65536 } },
  { pattern: "*qwen*235b*",     caps: { reasoning: true, thinkingFormat: "qwen", contextWindow: 262144 } },
-  { pattern: "*qwen*coder*",    caps: { reasoning: true, thinkingFormat: "qwen", contextWindow: 1000000 } },
  { pattern: "*qwq*",           caps: { reasoning: true, thinkingFormat: "qwen", thinkingCanDisable: false, contextWindow: 131072 } },
  { pattern: "*qwen*",          caps: { reasoning: true, thinkingFormat: "qwen", contextWindow: 262144 } },

  // ── Kimi (enabled→reasoning_effort; K2.7-code cannot disable) ─────
-  { pattern: "*kimi*k2.7*code*", caps: { vision: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 262144 } },
+  { pattern: "*kimi*k3*",       caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 1048576, maxOutput: 131072 } },
+  { pattern: "*kimi*for-coding*", caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 } },
+  { pattern: "*kimi*k2.7*code*", caps: { vision: true, videoInput: true, reasoning: true, thinkingFormat: "kimi", thinkingCanDisable: false, contextWindow: 262144, maxOutput: 65536 } },
  { pattern: "*kimi*k2*",       caps: { vision: true, reasoning: true, thinkingFormat: "kimi", contextWindow: 262144, maxOutput: 262144 } },
  { pattern: "*kimi*",          caps: { reasoning: true, thinkingFormat: "kimi", contextWindow: 262144 } },

@@ -233,6 +301,13 @@ export const PATTERN_CAPABILITIES = [
  { pattern: "*pplx*",          caps: { search: true, contextWindow: 128000 } },
  { pattern: "*perplexity*",    caps: { search: true, contextWindow: 128000 } },

+  // ── Poolside Laguna (resellers: openrouter/nvidia/kilocode/vercel/...) ──
+  // Free tiers cap S 2.1 well below the paid 1M window → match the free suffix
+  // (":free" or "-free", depending on reseller) before the plain id.
+  { pattern: "*laguna-s-2.1*free*", caps: { reasoning: true, thinkingFormat: "openai", contextWindow: 200000, maxOutput: 32000 } },
+  { pattern: "*laguna-s-2.1*",  caps: { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 32000 } },
+  { pattern: "*laguna*",        caps: { reasoning: true, thinkingFormat: "openai", contextWindow: 200000, maxOutput: 32000 } },
+
  // ── Others ───────────────────────────────────────────────────────
  { pattern: "*hunyuan*",       caps: { reasoning: true, thinkingFormat: "hunyuan", contextWindow: 262144, maxOutput: 262144 } },
  { pattern: "hy3*",            caps: { reasoning: true, thinkingFormat: "hunyuan", contextWindow: 262144, maxOutput: 262144 } },
@@ -252,13 +327,17 @@ export const PATTERN_CAPABILITIES = [
 export function getCapabilitiesForModel(provider, model) {
  if (!model) return { ...DEFAULT_CAPABILITIES };

+  // Canonical exact lookup strips vendor prefix: "anthropic/claude-opus-4.7" -> "claude-opus-4.7".
+  const baseModel = model.includes("/") ? model.split("/").pop() : model;
+
  // 1. Provider-specific override
-  if (provider && PROVIDER_CAPABILITIES[provider]?.[model]) {
-    return { ...DEFAULT_CAPABILITIES, ...PROVIDER_CAPABILITIES[provider][model] };
+  if (provider) {
+    const providerCaps = PROVIDER_CAPABILITIES[provider];
+    if (providerCaps?.[model]) return { ...DEFAULT_CAPABILITIES, ...providerCaps[model] };
+    if (providerCaps?.[baseModel]) return { ...DEFAULT_CAPABILITIES, ...providerCaps[baseModel] };
  }

-  // 2. Canonical exact (strip vendor prefix: "anthropic/claude-opus-4.7" -> "claude-opus-4.7")
-  const baseModel = model.includes("/") ? model.split("/").pop() : model;
+  // 2. Canonical exact
  if (MODEL_CAPABILITIES[baseModel]) return { ...DEFAULT_CAPABILITIES, ...MODEL_CAPABILITIES[baseModel] };
  if (MODEL_CAPABILITIES[model]) return { ...DEFAULT_CAPABILITIES, ...MODEL_CAPABILITIES[model] };

--- a/open-sse/providers/models/schema.js
+++ b/open-sse/providers/models/schema.js
@@ -1,5 +1,14 @@
 import { deriveModelName } from "./namePatterns.js";

+// Normalize version separators in a model id: hyphen between two digits becomes a dot.
+// Registry ids use dots for versions ("claude-sonnet-4.5") but clients (CLIs, aliases)
+// often send them with dashes ("claude-sonnet-4-5"). Only digit-digit hyphens are
+// touched, so word/suffix hyphens stay intact ("-thinking", "-agentic", "qwen3-coder-next").
+export function normalizeModelId(modelId) {
+  if (typeof modelId !== "string") return modelId;
+  return modelId.replace(/(\d)-(\d)/g, "$1.$2");
+}
+
 // Model defaults centralized (was scattered as `m.kind || "llm"`, `quotaFamily || "normal"`, etc.)
 export const MODEL_DEFAULTS = {
  kind: "llm",
--- a/open-sse/providers/pricing.js
+++ b/open-sse/providers/pricing.js
@@ -28,6 +28,7 @@ export const MODEL_PRICING = {
  "claude-sonnet-4.6":            { input: 3.00,  output: 15.00, cached: 0.30,  reasoning: 22.50,  cache_creation: 3.00  },
  "claude-opus-4-5-thinking":     { input: 5.00,  output: 25.00, cached: 0.50,  reasoning: 37.50,  cache_creation: 5.00  },
  "claude-opus-4-6-thinking":     { input: 5.00,  output: 25.00, cached: 0.50,  reasoning: 37.50,  cache_creation: 5.00  },
+  "claude-fable-5":               { input: 10.00, output: 50.00, cached: 1.00,  reasoning: 50.00,  cache_creation: 12.50 },

  // === OpenAI / GPT ===
  "gpt-3.5-turbo":                { input: 0.50,  output: 1.50,  cached: 0.25,  reasoning: 2.25,   cache_creation: 0.50  },
@@ -36,27 +37,33 @@ export const MODEL_PRICING = {
  "gpt-4o":                       { input: 2.50,  output: 10.00, cached: 1.25,  reasoning: 15.00,  cache_creation: 2.50  },
  "gpt-4o-mini":                  { input: 0.15,  output: 0.60,  cached: 0.075, reasoning: 0.90,   cache_creation: 0.15  },
  "gpt-4.1":                      { input: 2.50,  output: 10.00, cached: 1.25,  reasoning: 15.00,  cache_creation: 2.50  },
-  "gpt-5":                        { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  },
-  "gpt-5-mini":                   { input: 0.75,  output: 3.00,  cached: 0.375, reasoning: 4.50,   cache_creation: 0.75  },
-  "gpt-5-codex":                  { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  },
-  "gpt-5.1":                      { input: 4.00,  output: 16.00, cached: 2.00,  reasoning: 24.00,  cache_creation: 4.00  },
-  "gpt-5.1-codex":                { input: 4.00,  output: 16.00, cached: 2.00,  reasoning: 24.00,  cache_creation: 4.00  },
+  "gpt-5":                        { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  },
+  "gpt-5-mini":                   { input: 0.25,  output: 2.00,  cached: 0.125, reasoning: 2.00,   cache_creation: 0.25  },
+  "gpt-5-codex":                  { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  },
+  "gpt-5.1":                      { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  },
+  "gpt-5.1-codex":                { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  },
  "gpt-5.1-codex-mini":           { input: 1.50,  output: 6.00,  cached: 0.75,  reasoning: 9.00,   cache_creation: 1.50  },
  "gpt-5.1-codex-mini-high":      { input: 2.00,  output: 8.00,  cached: 1.00,  reasoning: 12.00,  cache_creation: 2.00  },
  "gpt-5.1-codex-max":            { input: 8.00,  output: 32.00, cached: 4.00,  reasoning: 48.00,  cache_creation: 8.00  },
-  "gpt-5.2":                      { input: 5.00,  output: 20.00, cached: 2.50,  reasoning: 30.00,  cache_creation: 5.00  },
-  "gpt-5.2-codex":                { input: 5.00,  output: 20.00, cached: 2.50,  reasoning: 30.00,  cache_creation: 5.00  },
-  "gpt-5.3-codex":                { input: 6.00,  output: 24.00, cached: 3.00,  reasoning: 36.00,  cache_creation: 6.00  },
-  "gpt-5.3-codex-xhigh":         { input: 10.00, output: 40.00, cached: 5.00,  reasoning: 60.00,  cache_creation: 10.00 },
-  "gpt-5.3-codex-high":          { input: 8.00,  output: 32.00, cached: 4.00,  reasoning: 48.00,  cache_creation: 8.00  },
-  "gpt-5.3-codex-low":           { input: 4.00,  output: 16.00, cached: 2.00,  reasoning: 24.00,  cache_creation: 4.00  },
-  "gpt-5.3-codex-none":          { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  },
+  "gpt-5.2":                      { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  },
+  "gpt-5.2-codex":                { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  },
+  "gpt-5.3-codex":                { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  },
  "gpt-5.3-codex-spark":         { input: 3.00,  output: 12.00, cached: 0.30,  reasoning: 12.00,  cache_creation: 3.00  },
+  "gpt-5.6":                      { input: 2.50,  output: 15.00, cached: 0.25,  reasoning: 15.00,  cache_creation: 2.50  },
+  "gpt-5.6-luna":                 { input: 1.00,  output: 6.00,  cached: 0.10,  reasoning: 6.00,   cache_creation: 1.00  },
+  "gpt-5.6-terra":                { input: 2.50,  output: 15.00, cached: 0.25,  reasoning: 15.00,  cache_creation: 2.50  },
+  "gpt-5.6-sol":                  { input: 5.00,  output: 30.00, cached: 0.50,  reasoning: 30.00,  cache_creation: 5.00  },
  "o1":                           { input: 15.00, output: 60.00, cached: 7.50,  reasoning: 90.00,  cache_creation: 15.00 },
  "o1-mini":                      { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  },

  // === Gemini ===
-  "gemini-3-flash-preview":       { input: 0.50,  output: 3.00,  cached: 0.03,  reasoning: 4.50,   cache_creation: 0.50  },
+  "gemini-3.6-flash":              { input: 1.50,  output: 7.50,  cached: 0.15,  reasoning: 11.25,  cache_creation: 1.875 },
+  "gemini-3.6-flash-high":         { input: 1.50,  output: 7.50,  cached: 0.15,  reasoning: 11.25,  cache_creation: 1.875 },
+  "gemini-3.6-flash-medium":       { input: 1.50,  output: 7.50,  cached: 0.15,  reasoning: 11.25,  cache_creation: 1.875 },
+  "gemini-3.6-flash-low":          { input: 1.50,  output: 7.50,  cached: 0.15,  reasoning: 11.25,  cache_creation: 1.875 },
+  "gemini-3.5-flash-lite":         { input: 0.30,  output: 2.50,  cached: 0.03,  reasoning: 3.75,   cache_creation: 0.375 },
+  "gemini-3.5-flash-high":         { input: 0.50,  output: 3.00,  cached: 0.03,  reasoning: 4.50,   cache_creation: 0.50  },
+  "gemini-3-flash-preview":        { input: 0.50,  output: 3.00,  cached: 0.03,  reasoning: 4.50,   cache_creation: 0.50  },
  "gemini-3-pro-preview":         { input: 2.00,  output: 12.00, cached: 0.25,  reasoning: 18.00,  cache_creation: 2.00  },
  "gemini-3.1-pro-low":           { input: 2.00,  output: 12.00, cached: 0.25,  reasoning: 18.00,  cache_creation: 2.00  },
  "gemini-3.1-pro-high":          { input: 4.00,  output: 18.00, cached: 0.50,  reasoning: 27.00,  cache_creation: 4.00  },
@@ -74,10 +81,18 @@ export const MODEL_PRICING = {
  "qwen3-coder-flash":            { input: 0.50,  output: 2.00,  cached: 0.25,  reasoning: 3.00,   cache_creation: 0.50  },

  // === Kimi ===
+  // Official platform.kimi.ai: cache-hit / cache-miss / output per 1M tokens
+  "kimi-k3":                      { input: 3.00,  output: 15.00, cached: 0.30,  reasoning: 15.00,  cache_creation: 3.00  },
+  "k3":                           { input: 3.00,  output: 15.00, cached: 0.30,  reasoning: 15.00,  cache_creation: 3.00  },
+  "kimi-k2.7-code":               { input: 0.95,  output: 4.00,  cached: 0.19,  reasoning: 4.00,   cache_creation: 0.95  },
+  "kimi-k2.7-code-highspeed":     { input: 1.90,  output: 8.00,  cached: 0.38,  reasoning: 8.00,   cache_creation: 1.90  },
+  "kimi-for-coding":              { input: 0.95,  output: 4.00,  cached: 0.19,  reasoning: 4.00,   cache_creation: 0.95  },
+  "kimi-for-coding-highspeed":    { input: 1.90,  output: 8.00,  cached: 0.38,  reasoning: 8.00,   cache_creation: 1.90  },
  "kimi-k2":                      { input: 1.00,  output: 4.00,  cached: 0.50,  reasoning: 6.00,   cache_creation: 1.00  },
  "kimi-k2-thinking":             { input: 1.50,  output: 6.00,  cached: 0.75,  reasoning: 9.00,   cache_creation: 1.50  },
  "kimi-k2.5":                    { input: 1.20,  output: 4.80,  cached: 0.60,  reasoning: 7.20,   cache_creation: 1.20  },
  "kimi-k2.5-thinking":           { input: 1.80,  output: 7.20,  cached: 0.90,  reasoning: 10.80,  cache_creation: 1.80  },
+  "kimi-k2.6":                    { input: 1.00,  output: 4.00,  cached: 0.50,  reasoning: 6.00,   cache_creation: 1.00  },
  "kimi-latest":                  { input: 1.00,  output: 4.00,  cached: 0.50,  reasoning: 6.00,   cache_creation: 1.00  },

  // === DeepSeek ===
@@ -122,7 +137,7 @@ export const MODEL_PRICING = {
 * Keyed by provider alias (cc, cx, gc, gh, ...) or provider id (openai, anthropic, ...).
 */
 export const PROVIDER_PRICING = {
-  // GitHub Copilot (gh) — gpt-5.3-codex has different rate than canonical
+  // GitHub Copilot (gh) — explicit override, matches canonical gpt-5.3-codex rate
  gh: {
    "gpt-5.3-codex": { input: 1.75, output: 14.00, cached: 0.175, reasoning: 14.00, cache_creation: 1.75 },
  },
@@ -140,11 +155,11 @@ export const PATTERN_PRICING = [
  { pattern: "*-codex-max",     pricing: { input: 8.00,  output: 32.00, cached: 4.00,  reasoning: 48.00,  cache_creation: 8.00  } },
  { pattern: "*-codex-mini-*",  pricing: { input: 1.50,  output: 6.00,  cached: 0.75,  reasoning: 9.00,   cache_creation: 1.50  } },
  { pattern: "*-codex-mini",    pricing: { input: 1.50,  output: 6.00,  cached: 0.75,  reasoning: 9.00,   cache_creation: 1.50  } },
-  { pattern: "*-codex-low",     pricing: { input: 4.00,  output: 16.00, cached: 2.00,  reasoning: 24.00,  cache_creation: 4.00  } },
-  { pattern: "*-codex-none",    pricing: { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  } },
+  { pattern: "*-codex-low",     pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },
+  { pattern: "*-codex-none",    pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },
  { pattern: "*-codex-spark",   pricing: { input: 3.00,  output: 12.00, cached: 0.30,  reasoning: 12.00,  cache_creation: 3.00  } },
-  { pattern: "codex-*",         pricing: { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  } },
-  { pattern: "*-codex",         pricing: { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  } },
+  { pattern: "codex-*",         pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },
+  { pattern: "*-codex",         pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },

  // --- Claude ---
  { pattern: "claude-opus-*",   pricing: { input: 5.00,  output: 25.00, cached: 0.50,  reasoning: 25.00,  cache_creation: 6.25  } },
@@ -161,11 +176,12 @@ export const PATTERN_PRICING = [
  { pattern: "gemini-*",        pricing: { input: 0.50,  output: 3.00,  cached: 0.03,  reasoning: 4.50,   cache_creation: 0.50  } },

  // --- GPT (specific first, generic last) ---
-  { pattern: "gpt-5.3-*",       pricing: { input: 6.00,  output: 24.00, cached: 3.00,  reasoning: 36.00,  cache_creation: 6.00  } },
-  { pattern: "gpt-5.2-*",       pricing: { input: 5.00,  output: 20.00, cached: 2.50,  reasoning: 30.00,  cache_creation: 5.00  } },
-  { pattern: "gpt-5.1-*",       pricing: { input: 4.00,  output: 16.00, cached: 2.00,  reasoning: 24.00,  cache_creation: 4.00  } },
-  { pattern: "gpt-5-*",         pricing: { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  } },
-  { pattern: "gpt-5*",          pricing: { input: 3.00,  output: 12.00, cached: 1.50,  reasoning: 18.00,  cache_creation: 3.00  } },
+  { pattern: "gpt-5.6-*",       pricing: { input: 2.50,  output: 15.00, cached: 0.25,  reasoning: 15.00,  cache_creation: 2.50  } },
+  { pattern: "gpt-5.3-*",       pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },
+  { pattern: "gpt-5.2-*",       pricing: { input: 1.75,  output: 14.00, cached: 0.175, reasoning: 14.00,  cache_creation: 1.75  } },
+  { pattern: "gpt-5.1-*",       pricing: { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  } },
+  { pattern: "gpt-5-*",         pricing: { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  } },
+  { pattern: "gpt-5*",          pricing: { input: 1.25,  output: 10.00, cached: 0.625, reasoning: 10.00,  cache_creation: 1.25  } },
  { pattern: "gpt-4o-*",        pricing: { input: 0.15,  output: 0.60,  cached: 0.075, reasoning: 0.90,   cache_creation: 0.15  } },
  { pattern: "gpt-4o",          pricing: { input: 2.50,  output: 10.00, cached: 1.25,  reasoning: 15.00,  cache_creation: 2.50  } },
  { pattern: "gpt-4*",          pricing: { input: 2.50,  output: 10.00, cached: 1.25,  reasoning: 15.00,  cache_creation: 2.50  } },
@@ -183,6 +199,7 @@ export const PATTERN_PRICING = [

  // --- Kimi ---
  { pattern: "kimi-*-thinking",  pricing: { input: 1.80,  output: 7.20,  cached: 0.90,  reasoning: 10.80,  cache_creation: 1.80  } },
+  { pattern: "kimi-k3*",        pricing: { input: 3.00,  output: 15.00, cached: 0.30,  reasoning: 15.00,  cache_creation: 3.00  } },
  { pattern: "kimi-k2*",        pricing: { input: 1.20,  output: 4.80,  cached: 0.60,  reasoning: 7.20,   cache_creation: 1.20  } },
  { pattern: "kimi-*",          pricing: { input: 1.00,  output: 4.00,  cached: 0.50,  reasoning: 6.00,   cache_creation: 1.00  } },

@@ -279,7 +296,10 @@ export function calculateCostFromTokens(tokens, pricing) {

  const inputTokens = tokens.prompt_tokens || tokens.input_tokens || 0;
  const cachedTokens = tokens.cached_tokens || tokens.cache_read_input_tokens || 0;
-  const nonCachedInput = Math.max(0, inputTokens - cachedTokens);
+  const cacheCreationTokens = tokens.cache_creation_input_tokens || 0;
+  // prompt_tokens is cache-inclusive (see canonicalizeUsage): cached + cache_creation
+  // are subsets, so subtract both to avoid charging them at the full input rate.
+  const nonCachedInput = Math.max(0, inputTokens - cachedTokens - cacheCreationTokens);

  cost += nonCachedInput * (pricing.input / 1000000);

@@ -295,7 +315,6 @@ export function calculateCostFromTokens(tokens, pricing) {
    cost += reasoningTokens * ((pricing.reasoning || pricing.output) / 1000000);
  }

-  const cacheCreationTokens = tokens.cache_creation_input_tokens || 0;
  if (cacheCreationTokens > 0) {
    cost += cacheCreationTokens * ((pricing.cache_creation || pricing.input) / 1000000);
  }
--- a/open-sse/providers/registry/alicode-intl.js
+++ b/open-sse/providers/registry/alicode-intl.js
@@ -3,19 +3,20 @@ export default {
  priority: 10,
  alias: "alicode-intl",
  display: {
-    name: "Alibaba Intl",
+    name: "Alibaba Coding",
    icon: "cloud",
    color: "#FF6A00",
    textIcon: "ALi",
-    website: "https://modelstudio.console.alibabacloud.com",
+    website: "https://www.alibabacloud.com/product/coding",
    notice: {
-      apiKeyUrl: "https://modelstudio.console.alibabacloud.com/?apiKey=1",
+      apiKeyUrl: "https://www.alibabacloud.com/product/coding",
    },
  },
  category: "apikey",
  transport: {
    baseUrl: "https://coding-intl.dashscope.aliyuncs.com/v1/chat/completions",
    headers: {},
+    quirks: { preserveCacheControl: true },
  },
  models: [
    { id: "qwen3.5-plus", name: "Qwen3.5 Plus" },
--- a/open-sse/providers/registry/alicode.js
+++ b/open-sse/providers/registry/alicode.js
@@ -16,6 +16,7 @@ export default {
  transport: {
    baseUrl: "https://coding.dashscope.aliyuncs.com/v1/chat/completions",
    headers: {},
+    quirks: { preserveCacheControl: true },
  },
  models: [
    { id: "qwen3.5-plus", name: "Qwen3.5 Plus" },
--- a/open-sse/providers/registry/alims-intl.js
+++ b/open-sse/providers/registry/alims-intl.js
@@ -0,0 +1,32 @@
+// Model Studio Intl — standard DashScope API keys (sk-...), NOT Coding Plan keys.
+// Sibling of alicode-intl (Coding Plan). Two key types use two different hosts.
+export default {
+  id: "alims-intl",
+  priority: 11,
+  alias: "alims-intl",
+  display: {
+    name: "Alibaba Studio",
+    icon: "cloud",
+    color: "#FF6A00",
+    textIcon: "ALi",
+    website: "https://modelstudio.console.alibabacloud.com",
+    notice: {
+      apiKeyUrl: "https://modelstudio.console.alibabacloud.com/?apiKey=1",
+    },
+  },
+  category: "apikey",
+  transport: {
+    baseUrl: "https://dashscope-intl.aliyuncs.com/compatible-mode/v1/chat/completions",
+    headers: {},
+    quirks: { preserveCacheControl: true },
+  },
+  models: [
+    { id: "qwen3.5-plus", name: "Qwen3.5 Plus" },
+    { id: "kimi-k2.5", name: "Kimi K2.5" },
+    { id: "glm-5", name: "GLM 5" },
+    { id: "MiniMax-M2.5", name: "MiniMax M2.5" },
+    { id: "qwen3-coder-next", name: "Qwen3 Coder Next" },
+    { id: "qwen3-coder-plus", name: "Qwen3 Coder Plus" },
+    { id: "glm-4.7", name: "GLM 4.7" },
+  ],
+};
--- a/open-sse/providers/registry/anthropic.js
+++ b/open-sse/providers/registry/anthropic.js
@@ -1,5 +1,3 @@
-import { CLAUDE_API_HEADERS } from "../shared.js";
-
 export default {
  id: "anthropic",
  priority: 30,
@@ -19,7 +17,7 @@ export default {
    baseUrl: "https://api.anthropic.com/v1/messages",
    format: "claude",
    headers: {
-      "Anthropic-Version": "2023-06-01",
+      "anthropic-version": "2023-06-01",
      "Anthropic-Beta": "claude-code-20250219,interleaved-thinking-2025-05-14",
    },
  },
--- a/open-sse/providers/registry/antigravity.js
+++ b/open-sse/providers/registry/antigravity.js
@@ -1,5 +1,4 @@
-import { platform, arch } from "os";
-import { ANTIGRAVITY_OAUTH_CLIENT } from "../shared.js";
+import { ANTIGRAVITY_IDE_BASE_URL, ANTIGRAVITY_IDE_USER_AGENT, ANTIGRAVITY_OAUTH_CLIENT } from "../shared.js";

 export default {
  id: "antigravity",
@@ -20,13 +19,10 @@ export default {
  category: "oauth",
  serviceKinds: ["llm", "image"],
  transport: {
-    baseUrls: [
-      "https://daily-cloudcode-pa.googleapis.com",
-      "https://daily-cloudcode-pa.sandbox.googleapis.com",
-    ],
+    baseUrls: [ANTIGRAVITY_IDE_BASE_URL],
    format: "antigravity",
    headers: {
-      "User-Agent": "antigravity/1.107.0 darwin/arm64",
+      "User-Agent": ANTIGRAVITY_IDE_USER_AGENT,
    },
    retry: {
      "429": {
@@ -40,6 +36,7 @@ export default {
      },
    },
    usage: {
+      // Discovery (quota/project) on PROD; daily host rejects these.
      quotaApiUrl: "https://cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels",
      loadProjectApiUrl: "https://cloudcode-pa.googleapis.com/v1internal:loadCodeAssist",
      tokenUrl: "https://oauth2.googleapis.com/token",
@@ -48,6 +45,10 @@ export default {
    clientSecret: "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",
  },
  models: [
+    { id: "gemini-3.6-flash-high", name: "Gemini 3.6 Flash (High)", upstreamModelId: "gemini-3.6-flash-tiered(high)" },
+    { id: "gemini-3.6-flash-medium", name: "Gemini 3.6 Flash (Medium)", upstreamModelId: "gemini-3.6-flash-tiered(medium)" },
+    { id: "gemini-3.6-flash-low", name: "Gemini 3.6 Flash (Low)", upstreamModelId: "gemini-3.6-flash-tiered(low)" },
+    { id: "gemini-3.5-flash-high", name: "Gemini 3.5 Flash (High)" },
    { id: "gemini-3-flash-agent", name: "Gemini 3.5 Flash (High)" },
    { id: "gemini-3.5-flash-low", name: "Gemini 3.5 Flash (Medium)" },
    { id: "gemini-3.5-flash-extra-low", name: "Gemini 3.5 Flash (Low)" },
@@ -71,7 +72,7 @@ export default {
      "https://www.googleapis.com/auth/cclog",
      "https://www.googleapis.com/auth/experimentsandconfigs",
    ],
-    apiEndpoint: "https://cloudcode-pa.googleapis.com",
+    apiEndpoint: "https://daily-cloudcode-pa.googleapis.com",
    apiVersion: "v1internal",
    loadCodeAssistEndpoint: "https://cloudcode-pa.googleapis.com/v1internal:loadCodeAssist",
    onboardUserEndpoint: "https://cloudcode-pa.googleapis.com/v1internal:onboardUser",
--- a/open-sse/providers/registry/api-airforce.js
+++ b/open-sse/providers/registry/api-airforce.js
@@ -0,0 +1,36 @@
+export default {
+  id: "api-airforce",
+  alias: "af",
+  aliases: [
+    "airforce",
+  ],
+  uiAlias: "af",
+  display: {
+    name: "API.airforce",
+    icon: "flight",
+    color: "#0EA5E9",
+    textIcon: "AF",
+    website: "https://api.airforce",
+    notice: {
+      apiKeyUrl: "https://api.airforce",
+    },
+  },
+  category: "freeTier",
+  authType: "apikey",
+  authModes: [
+    "apikey",
+  ],
+  transport: {
+    baseUrl: "https://api.airforce/v1/chat/completions",
+    validateUrl: "https://api.airforce/v1/models",
+    headers: {
+      "HTTP-Referer": "https://endpoint-proxy.local",
+      "X-Title": "Endpoint Proxy",
+    },
+  },
+  models: [
+    { id: "anthropic/claude-3.7-sonnet", name: "Claude 3.7 Sonnet (Free)", contextLength: 200000 },
+    { id: "moonshot/kimi-k2.6", name: "Kimi K2.6 (Free)", contextLength: 262144 },
+    { id: "google/gemini-2.5-flash", name: "Gemini 2.5 Flash (Free)", contextLength: 1048576 },
+  ],
+};
--- a/open-sse/providers/registry/baidu.js
+++ b/open-sse/providers/registry/baidu.js
@@ -0,0 +1,33 @@
+export default {
+  id: "baidu",
+  alias: "qianfan",
+  aliases: ["qianfan", "ernie", "baidu-qianfan"],
+  uiAlias: "qianfan",
+  category: "apikey",
+  authType: "apikey",
+  authModes: ["apikey"],
+  display: {
+    name: "Baidu Qianfan",
+    icon: "search",
+    color: "#2932E1",
+    textIcon: "BD",
+    website: "https://cloud.baidu.com/product/qianfan.html",
+    notice: {
+      apiKeyUrl:
+        "https://console.bce.baidu.com/qianfan/ais/console/applicationConsole/application",
+    },
+  },
+  transport: {
+    baseUrl: "https://qianfan.baidubce.com/v2/chat/completions",
+    validateUrl: "https://qianfan.baidubce.com/v2/models",
+  },
+  models: [
+    { id: "deepseek-v4-pro", name: "DeepSeek V4 Pro", contextLength: 1048576 },
+    { id: "deepseek-v4-flash", name: "DeepSeek V4 Flash", contextLength: 1048576 },
+    { id: "glm-5.2", name: "GLM 5.2", contextLength: 512000 },
+    { id: "glm-5.1", name: "GLM 5.1", contextLength: 198000 },
+    { id: "kimi-k2.6", name: "Kimi K2.6", contextLength: 262144 },
+    { id: "qwen3.5-397b-a17b", name: "Qwen 3.5 397B A17B", contextLength: 262144 },
+    { id: "qwen3.5-27b", name: "Qwen 3.5 27B", contextLength: 262144 },
+  ],
+};
--- a/open-sse/providers/registry/bazaarlink.js
+++ b/open-sse/providers/registry/bazaarlink.js
@@ -0,0 +1,47 @@
+export default {
+  id: "bazaarlink",
+  alias: "bzl",
+  aliases: ["bazaar-link"],
+  uiAlias: "bzl",
+  category: "freeTier",
+  authType: "apikey",
+  authModes: ["apikey"],
+  display: {
+    name: "Bazaarlink",
+    icon: "storefront",
+    color: "#DC2626",
+    textIcon: "BZ",
+    website: "https://bazaarlink.ai",
+    notice: { apiKeyUrl: "https://bazaarlink.ai" },
+  },
+  transport: {
+    baseUrl: "https://bazaarlink.ai/api/v1/chat/completions",
+    validateUrl: "https://bazaarlink.ai/api/v1/models",
+  },
+  models: [
+    { id: "auto:free", name: "Auto Free (Zero Cost)" },
+    { id: "claude-opus-4.7", name: "Claude Opus 4.7", contextLength: 1000000 },
+    { id: "claude-sonnet-4.6", name: "Claude Sonnet 4.6", contextLength: 1000000 },
+    { id: "claude-haiku-4.5", name: "Claude Haiku 4.5", contextLength: 200000 },
+    { id: "gpt-5.5", name: "GPT-5.5", contextLength: 1050000 },
+    { id: "gpt-5.4", name: "GPT-5.4", contextLength: 1050000 },
+    { id: "gpt-5.4-mini", name: "GPT-5.4 Mini", contextLength: 400000 },
+    { id: "gpt-5.4-nano", name: "GPT-5.4 Nano", contextLength: 400000 },
+    { id: "grok-4.3", name: "Grok 4.3", contextLength: 1000000 },
+    { id: "grok-4.20", name: "Grok 4.20", contextLength: 2000000 },
+    { id: "gemini-3.1-pro-preview", name: "Gemini 3.1 Pro", contextLength: 1048576 },
+    { id: "gemini-3-flash-preview", name: "Gemini 3 Flash", contextLength: 1048576 },
+    { id: "gemini-3.1-flash-lite-preview", name: "Gemini 3.1 Flash Lite", contextLength: 1048576 },
+    { id: "kimi-k2.6", name: "Kimi K2.6", contextLength: 262144 },
+    { id: "kimi-k2.5", name: "Kimi K2.5", contextLength: 262144 },
+    { id: "glm-5.1", name: "GLM 5.1", contextLength: 204800 },
+    { id: "glm-5", name: "GLM 5", contextLength: 204800 },
+    { id: "mimo-v2.5-pro", name: "MiMo-V2.5-Pro", contextLength: 1050000 },
+    { id: "mimo-v2.5", name: "MiMo-V2.5", contextLength: 1050000 },
+    { id: "minimax-m3", name: "MiniMax M3", contextLength: 1048576 },
+    { id: "minimax-m2.7", name: "MiniMax M2.7", contextLength: 204800 },
+    { id: "minimax-m2.5", name: "MiniMax M2.5", contextLength: 204800 },
+    { id: "qwen3.6-plus", name: "Qwen 3.6 Plus", contextLength: 1000000 },
+    { id: "nemotron-3-super-120b-a12b", name: "Nemotron 3 Super", contextLength: 1000000 },
+  ],
+};
--- a/open-sse/providers/registry/bluesminds.js
+++ b/open-sse/providers/registry/bluesminds.js
@@ -0,0 +1,38 @@
+export default {
+  id: "bluesminds",
+  alias: "bm",
+  aliases: ["blue-sminds"],
+  uiAlias: "bm",
+  hidden: true,
+  display: {
+    name: "BluesMinds",
+    icon: "psychology",
+    color: "#2563EB",
+    textIcon: "BM",
+    website: "https://bluesminds.com",
+    notice: { apiKeyUrl: "https://bluesminds.com" },
+  },
+  category: "apikey",
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://api.bluesminds.com/v1/chat/completions",
+    validateUrl: "https://api.bluesminds.com/v1/models",
+  },
+  models: [
+    { id: "gpt-4.1", name: "GPT-4.1", contextLength: 1048576 },
+    { id: "gpt-4.1-mini", name: "GPT-4.1 Mini", contextLength: 1048576 },
+    { id: "gpt-4.1-nano", name: "GPT-4.1 Nano", contextLength: 1048576 },
+    { id: "claude-sonnet-4-5", name: "Claude Sonnet 4.5", contextLength: 200000 },
+    { id: "claude-haiku-4-5", name: "Claude Haiku 4.5", contextLength: 200000 },
+    { id: "gemini-2.0-flash", name: "Gemini 2.0 Flash", contextLength: 1048576 },
+    { id: "gemini-2.0-flash-exp", name: "Gemini 2.0 Flash (Exp)", contextLength: 1048576 },
+    { id: "qwen-turbo", name: "Qwen Turbo", contextLength: 1000000 },
+    { id: "kimi-k2", name: "Kimi K2", contextLength: 262144 },
+    { id: "kimi-k2-thinking", name: "Kimi K2 Thinking", contextLength: 262144 },
+    { id: "glm-4.7", name: "GLM 4.7", contextLength: 204800 },
+    { id: "minimax-m2.5", name: "MiniMax M2.5", contextLength: 204800 },
+    { id: "claude-opus-4-5", name: "Claude Opus 4.5 (VIP)", contextLength: 200000 },
+    { id: "gemini-2.5-pro", name: "Gemini 2.5 Pro (VIP)", contextLength: 1048576 },
+  ],
+};
--- a/open-sse/providers/registry/claude.js
+++ b/open-sse/providers/registry/claude.js
@@ -60,12 +60,9 @@ export default {
    },
  },
  models: [
-    { id: "claude-opus-4-8", name: "Claude Opus 4.8" },
-    { id: "claude-opus-4-7", name: "Claude Opus 4.7" },
-    { id: "claude-opus-4-6", name: "Claude Opus 4.6" },
-    { id: "claude-sonnet-4-6", name: "Claude Sonnet 4.6" },
-    { id: "claude-opus-4-5-20251101", name: "Claude 4.5 Opus" },
-    { id: "claude-sonnet-4-5-20250929", name: "Claude 4.5 Sonnet" },
+    { id: "claude-opus-5", name: "Claude Opus 5" },
+    { id: "claude-fable-5", name: "Claude Fable 5" },
+    { id: "claude-sonnet-5", name: "Claude Sonnet 5" },
    { id: "claude-haiku-4-5-20251001", name: "Claude 4.5 Haiku" },
  ],
  oauth: {
--- a/open-sse/providers/registry/clinepass.js
+++ b/open-sse/providers/registry/clinepass.js
@@ -0,0 +1,57 @@
+export default {
+  id: "clinepass",
+  priority: 85,
+  alias: "clinepass",
+  uiAlias: "clinepass",
+  display: {
+    name: "ClinePass",
+    icon: "vpn_key",
+    color: "#5B9BD5",
+    textIcon: "CP",
+    website: "https://cline.bot",
+    notice: {
+      signupUrl: "https://app.cline.bot",
+    },
+  },
+  category: "oauth",
+  authModes: ["oauth", "apikey"],
+  hasOAuth: true,
+  transport: {
+    baseUrl: "https://api.cline.bot/api/v1/chat/completions",
+    headers: {
+      "HTTP-Referer": "https://cline.bot",
+      "X-Title": "Cline",
+    },
+    auth: {
+      combined: true,
+      header: "Authorization",
+      scheme: "bearer",
+      hooks: [
+        "clineHeaders",
+      ],
+    },
+  },
+  models: [
+    { id: "cline-pass/glm-5.2", name: "GLM-5.2 (ClinePass)" },
+    { id: "cline-pass/kimi-k2.7-code", name: "Kimi K2.7 Code (ClinePass)" },
+    { id: "cline-pass/kimi-k2.6", name: "Kimi K2.6 (ClinePass)" },
+    { id: "cline-pass/deepseek-v4-pro", name: "DeepSeek V4 Pro (ClinePass)" },
+    { id: "cline-pass/deepseek-v4-flash", name: "DeepSeek V4 Flash (ClinePass)" },
+    { id: "cline-pass/mimo-v2.5", name: "MiMo-V2.5 (ClinePass)" },
+    { id: "cline-pass/mimo-v2.5-pro", name: "MiMo-V2.5-Pro (ClinePass)" },
+    { id: "cline-pass/minimax-m3", name: "MiniMax M3 (ClinePass)" },
+    { id: "cline-pass/qwen3.7-max", name: "Qwen3.7 Max (ClinePass)" },
+    { id: "cline-pass/qwen3.7-plus", name: "Qwen3.7 Plus (ClinePass)" },
+  ],
+  oauth: {
+    appBaseUrl: "https://app.cline.bot",
+    apiBaseUrl: "https://api.cline.bot",
+    authorizeUrl: "https://api.cline.bot/api/v1/auth/authorize",
+    tokenUrl: "https://api.cline.bot/api/v1/auth/token",
+    refreshUrl: "https://api.cline.bot/api/v1/auth/refresh",
+  },
+  thinkingConfig: {
+    options: ["auto", "on", "off"],
+    defaultMode: "auto",
+  },
+};
--- a/open-sse/providers/registry/codebuddy-intl.js
+++ b/open-sse/providers/registry/codebuddy-intl.js
@@ -0,0 +1,73 @@
+// CodeBuddy international (codebuddy.ai) — mirrors codebuddy-cn registry shape,
+// swapping the Tencent CN domain for the .ai endpoint set. All OAuth/plugin URLs
+// use the /v2/plugin prefix with platform=ide (CN uses platform=CLI).
+export default {
+  id: "codebuddy-intl",
+  alias: "cbai",
+  uiAlias: "cbai",
+  hidden: false,
+  priority: 90,
+  display: {
+    name: "CodeBuddy",
+    icon: "smart_toy",
+    color: "#006EFF",
+    website: "https://www.codebuddy.ai",
+    notice: {
+      signupUrl: "https://www.codebuddy.ai",
+    },
+  },
+  category: "oauth",
+  authModes: ["oauth", "apikey"],
+  hasOAuth: true,
+  transport: {
+    // Chat gateway is OpenAI-compatible SSE (same /v2/chat/completions path as CN).
+    baseUrl: "https://www.codebuddy.ai/v2/chat/completions",
+    forceStream: true,
+    // CodeBuddy intl speaks the same unified OpenAI reasoning_effort shape as CN.
+    thinkingFormat: "openai",
+    headers: {
+      "User-Agent": "IDE/2.108.1 CodeBuddy/2.108.1",
+      "X-Product": "SaaS",
+      "X-IDE-Type": "IDE",
+      "X-IDE-Name": "IDE",
+      "x-requested-with": "XMLHttpRequest",
+      "x-codebuddy-request": "1",
+    },
+    auth: {
+      combined: true,
+      header: "Authorization",
+      scheme: "bearer",
+    },
+  },
+  // Same model lineup exposed by the CN gateway — intl backend is the same catalog.
+  models: [
+    { id: "glm-5.2", name: "GLM-5.2" },
+    { id: "glm-5.1", name: "GLM-5.1" },
+    { id: "glm-5.0", name: "GLM-5.0" },
+    { id: "glm-5.0-turbo", name: "GLM-5.0-Turbo" },
+    { id: "glm-5v-turbo", name: "GLM-5v-Turbo" },
+    { id: "glm-4.7", name: "GLM-4.7" },
+    { id: "minimax-m3", name: "MiniMax-M3" },
+    { id: "minimax-m2.7", name: "MiniMax-M2.7" },
+    { id: "kimi-k2.7", name: "Kimi-K2.7-Code" },
+    { id: "kimi-k2.6", name: "Kimi-K2.6" },
+    { id: "kimi-k2.5", name: "Kimi-K2.5" },
+    { id: "hy3-preview", name: "Hy3 Preview" },
+    { id: "deepseek-v4-pro", name: "DeepSeek-V4-Pro" },
+    { id: "deepseek-v4-flash", name: "DeepSeek-V4-Flash" },
+    { id: "deepseek-v3-2-volc", name: "DeepSeek-V3.2" },
+  ],
+  oauth: {
+    baseUrl: "https://www.codebuddy.ai",
+    stateUrl: "https://www.codebuddy.ai/v2/plugin/auth/state",
+    tokenUrl: "https://www.codebuddy.ai/v2/plugin/auth/token",
+    refreshUrl: "https://www.codebuddy.ai/v2/plugin/auth/token/refresh",
+    userAgent: "IDE/2.63.2 CodeBuddy/2.63.2",
+    platform: "ide",
+    pollInterval: 5000,
+  },
+  features: {
+    usage: true,
+    usageApikey: true,
+  },
+};
--- a/open-sse/providers/registry/codex.js
+++ b/open-sse/providers/registry/codex.js
@@ -40,26 +40,23 @@ export default {
    },
    usage: {
      url: "https://chatgpt.com/backend-api/wham/usage",
+      resetCreditsUrl: "https://chatgpt.com/backend-api/wham/rate-limit-reset-credits",
      resetCreditsConsumeUrl: "https://chatgpt.com/backend-api/wham/rate-limit-reset-credits/consume",
    },
  },
  models: [
+    { id: "gpt-5.6-sol", name: "GPT 5.6 Sol" },
+    { id: "gpt-5.6-sol-review", name: "GPT 5.6 Sol Review", upstreamModelId: "gpt-5.6-sol", quotaFamily: "review" },
+    { id: "gpt-5.6-terra", name: "GPT 5.6 Terra" },
+    { id: "gpt-5.6-terra-review", name: "GPT 5.6 Terra Review", upstreamModelId: "gpt-5.6-terra", quotaFamily: "review" },
+    { id: "gpt-5.6-luna", name: "GPT 5.6 Luna" },
+    { id: "gpt-5.6-luna-review", name: "GPT 5.6 Luna Review", upstreamModelId: "gpt-5.6-luna", quotaFamily: "review" },
    { id: "gpt-5.5", name: "GPT 5.5" },
    { id: "gpt-5.5-review", name: "GPT 5.5 Review", upstreamModelId: "gpt-5.5", quotaFamily: "review" },
    { id: "gpt-5.4", name: "GPT 5.4" },
    { id: "gpt-5.4-review", name: "GPT 5.4 Review", upstreamModelId: "gpt-5.4", quotaFamily: "review" },
    { id: "gpt-5.4-mini", name: "GPT 5.4 Mini" },
    { id: "gpt-5.4-mini-review", name: "GPT 5.4 Mini Review", upstreamModelId: "gpt-5.4-mini", quotaFamily: "review" },
-    { id: "gpt-5.3-codex", name: "GPT 5.3 Codex" },
-    { id: "gpt-5.3-codex-review", name: "GPT 5.3 Codex Review", upstreamModelId: "gpt-5.3-codex", quotaFamily: "review" },
-    { id: "gpt-5.3-codex-xhigh", name: "GPT 5.3 Codex (xHigh)" },
-    { id: "gpt-5.3-codex-xhigh-review", name: "GPT 5.3 Codex (xHigh) Review", upstreamModelId: "gpt-5.3-codex-xhigh", quotaFamily: "review" },
-    { id: "gpt-5.3-codex-high", name: "GPT 5.3 Codex (High)" },
-    { id: "gpt-5.3-codex-high-review", name: "GPT 5.3 Codex (High) Review", upstreamModelId: "gpt-5.3-codex-high", quotaFamily: "review" },
-    { id: "gpt-5.3-codex-low", name: "GPT 5.3 Codex (Low)" },
-    { id: "gpt-5.3-codex-low-review", name: "GPT 5.3 Codex (Low) Review", upstreamModelId: "gpt-5.3-codex-low", quotaFamily: "review" },
-    { id: "gpt-5.3-codex-none", name: "GPT 5.3 Codex (None)" },
-    { id: "gpt-5.3-codex-none-review", name: "GPT 5.3 Codex (None) Review", upstreamModelId: "gpt-5.3-codex-none", quotaFamily: "review" },
    { id: "gpt-5.3-codex-spark", name: "GPT 5.3 Codex Spark" },
    { id: "gpt-5.3-codex-spark-review", name: "GPT 5.3 Codex Spark Review", upstreamModelId: "gpt-5.3-codex-spark", quotaFamily: "review" },
    { id: "gpt-5.5-image", name: "GPT 5.5 Image", capabilities: ["text2img","edit"], params: ["size","quality","background","image_detail","output_format"], kind: "image" },
--- a/open-sse/providers/registry/commandcode.js
+++ b/open-sse/providers/registry/commandcode.js
@@ -23,9 +23,24 @@ export default {
    format: "commandcode",
    forceStream: true,
    headers: {
-      "x-command-code-version": "0.25.7",
+      "x-command-code-version": "1.10.0",
      "x-cli-environment": "cli",
+      "User-Agent": "cli",
    },
+    // Quota/billing endpoints (same alpha API the official CLI /usage calls).
+    // whoami resolves orgId; credits+subscription+usage/summary then report the
+    // 5-hour/weekly windows, plan, and period credits. See services/usage/commandcode.js.
+    usage: {
+      baseUrl: "https://api.commandcode.ai",
+      whoamiUrl: "/alpha/whoami",
+      creditsUrl: "/alpha/billing/credits",
+      subscriptionsUrl: "/alpha/billing/subscriptions",
+      summaryUrl: "/alpha/usage/summary",
+    },
+  },
+  features: {
+    usage: true,
+    usageApikey: true,
  },
  models: [
    { id: "deepseek/deepseek-v4-pro", name: "DeepSeek V4 Pro" },
--- a/open-sse/providers/registry/cursor.js
+++ b/open-sse/providers/registry/cursor.js
@@ -23,7 +23,7 @@ export default {
      "Content-Type": "application/connect+proto",
      "User-Agent": "connect-es/1.6.1",
    },
-    clientVersion: "3.1.0",
+    clientVersion: "3.12.17",
  },
  models: [
    { id: "default", name: "Auto (Server Picks)" },
@@ -44,11 +44,11 @@ export default {
  oauth: {
    apiEndpoint: "https://api2.cursor.sh",
    chatEndpoint: "/aiserver.v1.ChatService/StreamUnifiedChatWithTools",
-    modelsEndpoint: "/aiserver.v1.AiService/GetDefaultModelNudgeData",
+    modelsEndpoint: "/agent.v1.AgentService/GetUsableModels",
    api3Endpoint: "https://api3.cursor.sh",
    agentEndpoint: "https://agent.api5.cursor.sh",
    agentNonPrivacyEndpoint: "https://agentn.api5.cursor.sh",
-    clientVersion: "3.1.0",
+    clientVersion: "3.12.17",
    clientType: "ide",
    dbKeys: {
      accessToken: "cursorAuth/accessToken",
--- a/open-sse/providers/registry/deepseek.js
+++ b/open-sse/providers/registry/deepseek.js
@@ -48,4 +48,8 @@ export default {
    { id: "deepseek-chat", name: "DeepSeek V3.2 Chat" },
    { id: "deepseek-reasoner", name: "DeepSeek V3.2 Reasoner" },
  ],
+  features: {
+    usage: true,
+    usageApikey: true,
+  },
 };
--- a/open-sse/providers/registry/devin-cli.js
+++ b/open-sse/providers/registry/devin-cli.js
@@ -0,0 +1,63 @@
+export default {
+  id: "devin-cli",
+  alias: "dv",
+  aliases: ["devin"],
+  uiAlias: "dv",
+  hidden: true,
+  display: {
+    name: "Devin CLI",
+    icon: "smart_toy",
+    color: "#6366F1",
+    textIcon: "DV",
+    website: "https://devin.ai",
+    notice: {
+      signupUrl: "https://cli.devin.ai",
+      text: "Install: `curl -fsSL https://cli.devin.ai/install.sh | bash` (macOS: `brew install --cask devin-cli`, Windows PowerShell: `irm https://static.devin.ai/cli/setup.ps1 | iex`). Then run `devin auth login`. No API key needed.",
+    },
+  },
+  category: "free",
+  authType: "none",
+  noAuth: true,
+  authModes: ["none"],
+  transport: {
+    baseUrl: "devin://acp/stdio",
+    format: "openai",
+  },
+  models: [
+    { id: "swe-1.6-fast", name: "SWE-1.6 Fast" },
+    { id: "swe-1.6", name: "SWE-1.6" },
+    { id: "swe-1.5-fast", name: "SWE-1.5 Fast" },
+    { id: "swe-1.5", name: "SWE-1.5" },
+    { id: "claude-opus-4.7-max", name: "Claude Opus 4.7 Max", contextLength: 200000 },
+    { id: "claude-opus-4.7-high", name: "Claude Opus 4.7 High", contextLength: 200000 },
+    { id: "claude-opus-4.7-medium", name: "Claude Opus 4.7 Medium", contextLength: 200000 },
+    { id: "claude-opus-4.7-low", name: "Claude Opus 4.7 Low", contextLength: 200000 },
+    { id: "claude-sonnet-4.6-thinking-1m", name: "Claude Sonnet 4.6 Thinking 1M", contextLength: 1000000 },
+    { id: "claude-sonnet-4.6-thinking", name: "Claude Sonnet 4.6 Thinking", contextLength: 200000 },
+    { id: "claude-sonnet-4.6", name: "Claude Sonnet 4.6", contextLength: 200000 },
+    { id: "claude-opus-4.6-thinking", name: "Claude Opus 4.6 Thinking", contextLength: 200000 },
+    { id: "claude-opus-4.6", name: "Claude Opus 4.6", contextLength: 200000 },
+    { id: "claude-sonnet-4.5", name: "Claude Sonnet 4.5", contextLength: 200000 },
+    { id: "claude-haiku-4.5", name: "Claude Haiku 4.5", contextLength: 200000 },
+    { id: "gpt-5.5-xhigh", name: "GPT-5.5 XHigh", contextLength: 200000 },
+    { id: "gpt-5.5-high", name: "GPT-5.5 High", contextLength: 200000 },
+    { id: "gpt-5.5-medium", name: "GPT-5.5 Medium", contextLength: 200000 },
+    { id: "gpt-5.5-low", name: "GPT-5.5 Low", contextLength: 200000 },
+    { id: "gpt-5.4-high", name: "GPT-5.4 High", contextLength: 200000 },
+    { id: "gpt-5.4-medium", name: "GPT-5.4 Medium", contextLength: 200000 },
+    { id: "gpt-5.4-low", name: "GPT-5.4 Low", contextLength: 200000 },
+    { id: "gpt-5.3-codex-high", name: "GPT-5.3 Codex High", contextLength: 200000 },
+    { id: "gpt-5.3-codex-medium", name: "GPT-5.3 Codex Medium", contextLength: 200000 },
+    { id: "gpt-5.3-codex-low", name: "GPT-5.3 Codex Low", contextLength: 200000 },
+    { id: "gpt-5.2-high", name: "GPT-5.2 High", contextLength: 200000 },
+    { id: "gpt-5.2-medium", name: "GPT-5.2 Medium", contextLength: 200000 },
+    { id: "gpt-5.2-low", name: "GPT-5.2 Low", contextLength: 200000 },
+    { id: "gemini-3.1-pro-high", name: "Gemini 3.1 Pro High", contextLength: 1000000 },
+    { id: "gemini-3.1-pro-low", name: "Gemini 3.1 Pro Low", contextLength: 1000000 },
+    { id: "gemini-3.0-flash-high", name: "Gemini 3 Flash High", contextLength: 1000000 },
+    { id: "gemini-2.5-pro", name: "Gemini 2.5 Pro", contextLength: 1000000 },
+    { id: "deepseek-v4", name: "DeepSeek V4", contextLength: 1048576 },
+    { id: "kimi-k2.6", name: "Kimi K2.6", contextLength: 262144 },
+    { id: "glm-5.1", name: "GLM-5.1", contextLength: 204800 },
+  ],
+};
--- a/open-sse/providers/registry/featherless.js
+++ b/open-sse/providers/registry/featherless.js
@@ -0,0 +1,34 @@
+export default {
+  id: "featherless",
+  priority: 65,
+  alias: "featherless",
+  aliases: [
+    "fl",
+  ],
+  uiAlias: "fl",
+  display: {
+    name: "Featherless",
+    icon: "flutter_dash",
+    color: "#111827",
+    textIcon: "FL",
+    website: "https://featherless.ai",
+    notice: {
+      apiKeyUrl: "https://featherless.ai/account/api-keys",
+    },
+  },
+  category: "apikey",
+  authType: "apikey",
+  transport: {
+    baseUrl: "https://api.featherless.ai/v1/chat/completions",
+    validateUrl: "https://api.featherless.ai/v1/models",
+  },
+  models: [
+    { id: "deepseek-ai/DeepSeek-V4-Pro", name: "DeepSeek V4 Pro" },
+    { id: "deepseek-ai/DeepSeek-V4-Flash", name: "DeepSeek V4 Flash" },
+    { id: "zai-org/GLM-5.2", name: "GLM 5.2" },
+    { id: "zai-org/GLM-5.1", name: "GLM 5.1" },
+    { id: "moonshotai/Kimi-K2.7-Code", name: "Kimi K2.7 Code" },
+    { id: "moonshotai/Kimi-K2.6", name: "Kimi K2.6" },
+    { id: "moonshotai/Kimi-K2.5", name: "Kimi K2.5" },
+  ],
+};
--- a/open-sse/providers/registry/gemini.js
+++ b/open-sse/providers/registry/gemini.js
@@ -16,6 +16,8 @@ export default {
    },
  },
  category: "freeTier",
+  authType: "apikey",
+  authModes: ["apikey"],
  mediaPriority: 1,
  transport: {
    baseUrl: "https://generativelanguage.googleapis.com/v1beta/models",
@@ -34,6 +36,8 @@ export default {
    },
  },
  models: [
+    { id: "gemini-3.6-flash", name: "Gemini 3.6 Flash" },
+    { id: "gemini-3.5-flash-lite", name: "Gemini 3.5 Flash Lite" },
    { id: "gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview" },
    { id: "gemini-3.1-flash-lite-preview", name: "Gemini 3.1 Flash Lite Preview" },
    { id: "gemini-3-flash-preview", name: "Gemini 3 Flash Preview" },
--- a/open-sse/providers/registry/github.js
+++ b/open-sse/providers/registry/github.js
@@ -18,6 +18,7 @@ export default {
  transport: {
    baseUrl: "https://api.githubcopilot.com/chat/completions",
    responsesUrl: "https://api.githubcopilot.com/responses",
+    messagesUrl: "https://api.githubcopilot.com/v1/messages",
    headers: {
      "copilot-integration-id": "vscode-chat",
      "editor-version": "vscode/1.110.0",
@@ -46,6 +47,14 @@ export default {
    { id: "gpt-5.3-codex", name: "GPT-5.3 Codex" },
    { id: "gpt-5.4", name: "GPT-5.4" },
    { id: "gpt-5.4-mini", name: "GPT-5.4 Mini" },
+    // Note: routing to Copilot's Anthropic-native /v1/messages shim (see
+    // executors/github.js) is decided by model-NAME pattern at request time, not by
+    // a static targetFormat field here — Copilot's live model catalog (see
+    // services/copilotModels.js) regularly exposes claude-* models this static list
+    // hasn't caught up with yet (e.g. claude-opus-4.8), and a static per-entry
+    // targetFormat would silently miss those while also double-translating requests
+    // for models that ARE listed here (chatCore.js would pre-translate to Claude
+    // shape, then the executor would translate again). Keep these as plain entries.
    { id: "claude-haiku-4.5", name: "Claude Haiku 4.5" },
    { id: "claude-opus-4.5", name: "Claude Opus 4.5" },
    { id: "claude-sonnet-4.5", name: "Claude Sonnet 4.5" },
--- a/open-sse/providers/registry/grok-cli.js
+++ b/open-sse/providers/registry/grok-cli.js
@@ -0,0 +1,96 @@
+/**
+ * Grok CLI / Grok Build (cli-chat-proxy.grok.com)
+ *
+ * Source of truth: wire capture of official @xai-official/grok 0.2.99
+ * talking to https://cli-chat-proxy.grok.com (OpenAI Responses API).
+ *
+ * Distinct from:
+ *  - `xai`      → api.x.ai (API key / xAI API OAuth PKCE)
+ *  - `grok-web` → grok.com web SSO cookie
+ */
+import {
+  GROK_CLI_BASE_URL,
+  GROK_CLI_CLIENT_IDENTIFIER,
+  GROK_CLI_MODEL,
+  GROK_CLI_USER_AGENT,
+  GROK_CLI_VERSION,
+} from "../../config/grokCli.js";
+
+export default {
+  id: "grok-cli",
+  priority: 275,
+  alias: "gcli",
+  aliases: ["grok-build", "gb"],
+  uiAlias: "gcli",
+  display: {
+    name: "Grok CLI (Grok Build)",
+    icon: "auto_awesome",
+    color: "#1DA1F2",
+    textIcon: "GC",
+    website: "https://x.ai",
+    notice: {
+      text: "Sign in with your xAI / Grok account via device code. Uses Grok Build subscription credits (cli-chat-proxy.grok.com).",
+      signupUrl: "https://grok.com/supergrok",
+    },
+  },
+  category: "oauth",
+  authModes: ["oauth"],
+  hasOAuth: true,
+  thinkingConfig: {
+    options: ["low", "medium", "high", "xhigh"],
+    defaultMode: "high",
+  },
+  transport: {
+    baseUrl: `${GROK_CLI_BASE_URL}/responses`,
+    format: "openai-responses",
+    forceStream: true,
+    modelsUrl: `${GROK_CLI_BASE_URL}/models`,
+    userUrl: `${GROK_CLI_BASE_URL}/user`,
+    billingUrl: `${GROK_CLI_BASE_URL}/billing`,
+    clientVersion: GROK_CLI_VERSION,
+    clientIdentifier: GROK_CLI_CLIENT_IDENTIFIER,
+    tokenAuth: "xai-grok-cli",
+    headers: {
+      "User-Agent": GROK_CLI_USER_AGENT,
+      "x-grok-client-identifier": GROK_CLI_CLIENT_IDENTIFIER,
+      "x-grok-client-version": GROK_CLI_VERSION,
+    },
+    // Quota tracker: official CLI polls billing?format=credits + user?include=subscription
+    usage: {
+      url: `${GROK_CLI_BASE_URL}/billing?format=credits`,
+      userUrl: `${GROK_CLI_BASE_URL}/user?include=subscription`,
+    },
+    retry: {
+      429: { attempts: 2, delayMs: 2000 },
+      502: { attempts: 2, delayMs: 1500 },
+      503: { attempts: 2, delayMs: 1500 },
+    },
+  },
+  models: [
+    {
+      id: GROK_CLI_MODEL,
+      name: "Grok Build",
+      contextLength: 500000,
+      maxOutputTokens: 64000,
+    },
+    { id: "grok-4.5", name: "Grok 4.5" },
+    { id: "grok-4.5-high", name: "Grok 4.5 (High)", upstreamModelId: "grok-4.5" },
+    { id: "grok-4.5-medium", name: "Grok 4.5 (Medium)", upstreamModelId: "grok-4.5" },
+    { id: "grok-4.5-low", name: "Grok 4.5 (Low)", upstreamModelId: "grok-4.5" },
+  ],
+  features: {
+    usage: true,
+  },
+  oauth: {
+    // Same public client_id as Grok CLI / existing xai OAuth
+    clientId: "b1a00492-073a-47ea-816f-4c329264a828",
+    deviceCodeUrl: "https://auth.x.ai/oauth2/device/code",
+    tokenUrl: "https://auth.x.ai/oauth2/token",
+    refreshUrl: "https://auth.x.ai/oauth2/token",
+    // HAR scope includes conversations read/write beyond the api-only xai scope
+    scope:
+      "openid profile email offline_access grok-cli:access api:access conversations:read conversations:write",
+    referrer: "grok-build",
+    refreshLeadMs: 5 * 60 * 1000,
+  },
+};
--- a/open-sse/providers/registry/index.js
+++ b/open-sse/providers/registry/index.js
@@ -1,4 +1,4 @@
-// Auto-generated: static imports of all registry entries
+// Auto-generated: static imports for all registry entries
 import p0 from "./alicode-intl.js";
 import p1 from "./alicode.js";
 import p2 from "./anthropic.js";
@@ -15,85 +15,107 @@ import p12 from "./cerebras.js";
 import p13 from "./chutes.js";
 import p14 from "./claude.js";
 import p15 from "./cline.js";
-import p16 from "./cloudflare-ai.js";
-import p17 from "./codebuddy-cn.js";
-import p18 from "./codex.js";
-import p19 from "./cohere.js";
-import p20 from "./comfyui.js";
-import p21 from "./commandcode.js";
-import p22 from "./coqui.js";
-import p23 from "./cursor.js";
-import p24 from "./deepgram.js";
-import p25 from "./deepseek.js";
-import p26 from "./edge-tts.js";
-import p27 from "./elevenlabs.js";
-import p28 from "./exa.js";
-import p29 from "./fal-ai.js";
-import p30 from "./firecrawl.js";
-import p31 from "./fireworks.js";
-import p32 from "./gemini-cli.js";
-import p33 from "./gemini.js";
-import p34 from "./github.js";
-import p35 from "./gitlab.js";
-import p36 from "./glm-cn.js";
-import p37 from "./glm.js";
-import p38 from "./google-pse.js";
-import p39 from "./google-tts.js";
-import p40 from "./grok-web.js";
-import p41 from "./groq.js";
-import p42 from "./huggingface.js";
-import p43 from "./hyperbolic.js";
-import p44 from "./iflow.js";
-import p45 from "./inworld.js";
-import p46 from "./jina-ai.js";
-import p47 from "./jina-reader.js";
-import p48 from "./kilocode.js";
-import p49 from "./kimi-coding.js";
-import p50 from "./kimi.js";
-import p51 from "./kiro.js";
-import p52 from "./linkup.js";
-import p53 from "./local-device.js";
-import p54 from "./mimo-free.js";
-import p55 from "./minimax-cn.js";
-import p56 from "./minimax.js";
-import p57 from "./mistral.js";
-import p58 from "./mmf.js";
-import p59 from "./nanobanana.js";
-import p60 from "./nebius.js";
-import p61 from "./nvidia.js";
-import p62 from "./ollama-local.js";
-import p63 from "./ollama.js";
-import p64 from "./openai.js";
-import p65 from "./opencode-go.js";
-import p66 from "./opencode.js";
-import p67 from "./openrouter.js";
-import p68 from "./perplexity-web.js";
-import p69 from "./perplexity.js";
-import p70 from "./playht.js";
-import p71 from "./qoder.js";
-import p72 from "./qwen.js";
-import p73 from "./recraft.js";
-import p74 from "./runwayml.js";
-import p75 from "./sdwebui.js";
-import p76 from "./searchapi.js";
-import p77 from "./searxng.js";
-import p78 from "./serper.js";
-import p79 from "./siliconflow.js";
-import p80 from "./stability-ai.js";
-import p81 from "./tavily.js";
-import p82 from "./together.js";
-import p83 from "./topaz.js";
-import p84 from "./tortoise.js";
-import p85 from "./venice.js";
-import p86 from "./vercel-ai-gateway.js";
-import p87 from "./vertex-partner.js";
-import p88 from "./vertex.js";
-import p89 from "./volcengine-ark.js";
-import p90 from "./voyage-ai.js";
-import p91 from "./xai.js";
-import p92 from "./xiaomi-mimo.js";
-import p93 from "./xiaomi-tokenplan.js";
-import p94 from "./youcom.js";
+import p16 from "./clinepass.js";
+import p17 from "./cloudflare-ai.js";
+import p18 from "./codebuddy-cn.js";
+import p19 from "./codex.js";
+import p20 from "./cohere.js";
+import p21 from "./comfyui.js";
+import p22 from "./commandcode.js";
+import p23 from "./coqui.js";
+import p24 from "./cursor.js";
+import p25 from "./deepgram.js";
+import p26 from "./deepseek.js";
+import p27 from "./edge-tts.js";
+import p28 from "./elevenlabs.js";
+import p29 from "./exa.js";
+import p30 from "./fal-ai.js";
+import p31 from "./featherless.js";
+import p32 from "./firecrawl.js";
+import p33 from "./fireworks.js";
+import p34 from "./gemini-cli.js";
+import p35 from "./gemini.js";
+import p36 from "./github.js";
+import p37 from "./gitlab.js";
+import p38 from "./glm-cn.js";
+import p39 from "./glm.js";
+import p40 from "./google-pse.js";
+import p41 from "./google-tts.js";
+import p42 from "./grok-cli.js";
+import p43 from "./grok-web.js";
+import p44 from "./groq.js";
+import p45 from "./huggingface.js";
+import p46 from "./hyperbolic.js";
+import p47 from "./iflow.js";
+import p48 from "./inworld.js";
+import p49 from "./jina-ai.js";
+import p50 from "./jina-reader.js";
+import p51 from "./kilocode.js";
+import p52 from "./kimchi.js";
+import p53 from "./kimi.js";
+import p54 from "./kiro.js";
+import p55 from "./linkup.js";
+import p56 from "./local-device.js";
+import p57 from "./mimo-free.js";
+import p58 from "./minimax-cn.js";
+import p59 from "./minimax.js";
+import p60 from "./mistral.js";
+import p61 from "./mmf.js";
+import p62 from "./nanobanana.js";
+import p63 from "./nebius.js";
+import p64 from "./nvidia.js";
+import p65 from "./ollama-local.js";
+import p66 from "./ollama.js";
+import p67 from "./openai.js";
+import p68 from "./opencode-go.js";
+import p69 from "./opencode.js";
+import p70 from "./openrouter.js";
+import p71 from "./perplexity-web.js";
+import p72 from "./perplexity.js";
+import p73 from "./perplexity-agent.js";
+import p74 from "./playht.js";
+import p75 from "./qoder.js";
+import p76 from "./qwen.js";
+import p77 from "./recraft.js";
+import p78 from "./runwayml.js";
+import p79 from "./sdwebui.js";
+import p80 from "./searchapi.js";
+import p81 from "./searxng.js";
+import p82 from "./serper.js";
+import p83 from "./siliconflow.js";
+import p84 from "./stability-ai.js";
+import p85 from "./tavily.js";
+import p86 from "./together.js";
+import p87 from "./topaz.js";
+import p88 from "./tortoise.js";
+import p89 from "./venice.js";
+import p90 from "./vercel-ai-gateway.js";
+import p91 from "./vertex-partner.js";
+import p92 from "./vertex.js";
+import p93 from "./volcengine-ark.js";
+import p94 from "./voyage-ai.js";
+import p95 from "./xai.js";
+import p96 from "./xiaomi-mimo.js";
+import p97 from "./xiaomi-tokenplan.js";
+import p98 from "./youcom.js";
+import p99 from "./alims-intl.js";
+import p100 from "./codebuddy-intl.js";
+// Temporarily hidden — no tool calling support (trae SOLO agent / windsurf gRPC skip ToolCallChunk).
+// Re-enable by uncommenting both the import and the array entry below.
+// import p102 from "./trae.js";
+import p103 from "./zed.js";
+import p105 from "./api-airforce.js";
+import p106 from "./baidu.js";
+import p107 from "./bazaarlink.js";
+import p108 from "./bluesminds.js";
+import p109 from "./kilo-gateway.js";
+import p110 from "./llm7.js";
+import p111 from "./sambanova.js";
+import p112 from "./tencent.js";
+import p113 from "./morph.js";
+// import p114 from "./devin-cli.js";
+// import p104 from "./windsurf.js";
+import p115 from "./poolside.js";

 export default [
  p0,
@@ -190,5 +212,25 @@ export default [
  p91,
  p92,
  p93,
-  p94
+  p94,
+  p95,
+  p96,
+  p97,
+  p98,
+  p99,
+  p100,
+  // p102, // trae — hidden, no tool calling
+  p103,
+  p105,
+  p106,
+  p107,
+  p108,
+  p109,
+  p110,
+  p111,
+  p112,
+  p113,
+  // p114, // devin-cli — hidden, spawns local agent with shell/fs access
+  // p104, // windsurf — hidden, no tool calling
+  p115,
 ];
--- a/open-sse/providers/registry/kilo-gateway.js
+++ b/open-sse/providers/registry/kilo-gateway.js
@@ -0,0 +1,34 @@
+export default {
+  id: "kilo-gateway",
+  alias: "kgw",
+  aliases: [
+    "kilo-gateway",
+    "kilogateway",
+  ],
+  uiAlias: "kgw",
+  category: "freeTier",
+  display: {
+    name: "Kilo Gateway",
+    icon: "login",
+    color: "#8B5CF6",
+    textIcon: "KG",
+    website: "https://kilo.ai",
+    notice: {
+      apiKeyUrl: "https://kilo.ai/dashboard?tab=apiKeys",
+    },
+  },
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://api.kilo.ai/api/gateway/chat/completions",
+    validateUrl: "https://api.kilo.ai/api/gateway/models",
+  },
+  models: [
+    { id: "kilo-auto/free", name: "Kilo Auto Free", contextLength: 256000 },
+    { id: "nvidia/nemotron-3-super-120b-a12b:free", name: "Nemotron 3 Super 120B (Free)", contextLength: 262144 },
+    { id: "nvidia/nemotron-3-ultra-550b-a55b:free", name: "Nemotron 3 Ultra 550B (Free)", contextLength: 1000000 },
+    { id: "kwaipilot/kat-coder-pro-v2.5:free", name: "Kat Coder Pro v2.5 (Free)", contextLength: 256000 },
+    { id: "kilo-auto/frontier", name: "Kilo Auto Frontier", contextLength: 1000000 },
+    { id: "kilo-auto/balanced", name: "Kilo Auto Balanced", contextLength: 1000000 },
+  ],
+};
--- a/open-sse/providers/registry/kilocode.js
+++ b/open-sse/providers/registry/kilocode.js
@@ -36,6 +36,13 @@ export default {
    { id: "deepseek/deepseek-chat", name: "DeepSeek Chat" },
    { id: "deepseek/deepseek-reasoner", name: "DeepSeek Reasoner" },
  ],
+  // Kilo Code proxies the OpenRouter catalog (334 models at time of writing),
+  // so the hardcoded list above is only a fallback. Surfacing the full catalog
+  // requires a fetcher + passthroughModels, matching how openrouter.js is set up.
+  // Without these, only the 8 hardcoded models appear in the combo model picker,
+  // hiding dynamic models like cohere/north-mini-code:free and poolside/laguna-m.1:free.
+  modelsFetcher: { url: "https://api.kilo.ai/api/gateway/models", type: "openrouter-free" },
+  passthroughModels: true,
  oauth: {
    apiBaseUrl: "https://api.kilo.ai",
    initiateUrl: "https://api.kilo.ai/api/device-auth/codes",
--- a/open-sse/providers/registry/kimchi.js
+++ b/open-sse/providers/registry/kimchi.js
@@ -0,0 +1,49 @@
+export default {
+  id: "kimchi",
+  priority: 95,
+  alias: "kimchi",
+  uiAlias: "kimchi",
+  display: {
+    name: "Kimchi",
+    icon: "restaurant",
+    color: "#FF521D",
+    textIcon: "KC",
+    website: "https://kimchi.dev",
+    notice: {
+      signupUrl: "https://app.kimchi.dev",
+    },
+  },
+  category: "freeTier",
+  authModes: ["oauth"],
+  hasOAuth: true,
+  transport: {
+    baseUrl: "https://llm.kimchi.dev/openai/v1/chat/completions",
+    format: "openai",
+    headers: {
+      "User-Agent": "kimchi/0.1.50",
+    },
+    auth: {
+      combined: true,
+      header: "Authorization",
+      scheme: "bearer",
+    },
+  },
+  models: [
+    { id: "minimax-m3", name: "MiniMax-M3" },
+    { id: "kimi-k2.7", name: "Kimi-K2.7" },
+    { id: "kimi-k2.6", name: "Kimi-K2.6" },
+    { id: "kimi-k2.5", name: "Kimi-K2.5" },
+    { id: "nemotron-3-ultra-fp4", name: "Nemotron 3 Ultra FP4" },
+    { id: "minimax-m2.7", name: "MiniMax-M2.7" },
+    { id: "claude-opus-4-6", name: "Claude Opus 4.6" },
+    { id: "claude-sonnet-4-6", name: "Claude Sonnet 4.6" },
+  ],
+  serviceKinds: ["llm", "imageToText"],
+  oauth: {
+    webAppUrl: "https://app.kimchi.dev",
+    validationUrl: "https://api.cast.ai/v1/llm/openai/supported-providers",
+    userInfoUrl: "https://app.kimchi.dev/api/v1/me",
+    modelsUrl: "https://llm.kimchi.dev/v1/models/metadata?include_in_cli=true",
+  },
+  passthroughModels: true,
+};
--- a/open-sse/providers/registry/kimi-coding.js
+++ b/open-sse/providers/registry/kimi-coding.js
@@ -1,65 +0,0 @@
-import { CLAUDE_API_HEADERS, KIMI_CODING_BASE_URL } from "../shared.js";
-
-export default {
-  id: "kimi-coding",
-  hidden: true,
-  priority: 120,
-  alias: "kmc",
-  display: {
-    name: "Kimi Coding",
-    icon: "psychology",
-    color: "#1E40AF",
-    textIcon: "KC",
-    website: "https://kimi.moonshot.cn",
-    notice: {
-      signupUrl: "https://kimi.moonshot.cn",
-    },
-  },
-  category: "oauth",
-  transport: {
-    baseUrl: "https://api.kimi.com/coding/v1/messages",
-    format: "claude",
-    urlSuffix: "?beta=true",
-    headers: { ...CLAUDE_API_HEADERS },
-    clientId: "17e5f671-d194-4dfb-9706-5516cb48c098",
-    tokenUrl: "https://auth.kimi.com/api/oauth/token",
-    refreshUrl: "https://auth.kimi.com/api/oauth/token",
-    auth: {
-      combined: true,
-      header: "x-api-key",
-      scheme: "raw",
-      hooks: [
-        "kimiHeaders",
-      ],
-    },
-  },
-  // Multi-endpoint: pick the transport matching client sourceFormat to skip translation.
-  transports: [
-    {
-      format: "openai",
-      baseUrl: "https://api.kimi.com/coding/v1/chat/completions",
-      auth: { combined: true, header: "Authorization", scheme: "bearer", hooks: ["kimiHeaders"] },
-    },
-    {
-      format: "claude",
-      baseUrl: "https://api.kimi.com/coding/v1/messages",
-      urlSuffix: "?beta=true",
-      headers: { ...CLAUDE_API_HEADERS },
-      auth: { combined: true, header: "x-api-key", scheme: "raw", hooks: ["kimiHeaders"] },
-    },
-  ],
-  models: [
-    { id: "kimi-k2.6", name: "Kimi K2.6" },
-    { id: "kimi-k2.5", name: "Kimi K2.5" },
-    { id: "kimi-k2.5-thinking", name: "Kimi K2.5 Thinking" },
-    { id: "kimi-latest", name: "Kimi Latest" },
-  ],
-  oauth: {
-    deviceCodeUrl: "https://auth.kimi.com/api/oauth/device_authorization",
-    tokenUrl: "https://auth.kimi.com/api/oauth/token",
-    refreshLeadMs: 300000,
-  },
-  features: {
-    usage: true,
-  },
-};
--- a/open-sse/providers/registry/kimi.js
+++ b/open-sse/providers/registry/kimi.js
@@ -1,9 +1,14 @@
-import { CLAUDE_API_HEADERS, KIMI_CODING_BASE_URL } from "../shared.js";
+import { CLAUDE_API_HEADERS } from "../shared.js";

+// Dual auth (same pattern as xai): OAuth = Kimi Code subscription (device code),
+// API key = platform.moonshot / api.kimi.com. Transport is shared.
+// CLIProxyAPI parity: client_id, auth.kimi.com device+token, X-Msh-* headers, device_id.
 export default {
  id: "kimi",
  priority: 170,
  alias: "kimi",
+  // Legacy id + short alias from former kimi-coding registry entry
+  aliases: ["kimi-coding", "kmc"],
  display: {
    name: "Kimi",
    icon: "psychology",
@@ -12,18 +17,25 @@ export default {
    website: "https://kimi.moonshot.cn",
    notice: {
      apiKeyUrl: "https://platform.moonshot.ai/console/api-keys",
+      signupUrl: "https://www.kimi.com/code",
    },
  },
-  category: "apikey",
+  category: "oauth",
+  authModes: ["oauth", "apikey"],
+  hasOAuth: true,
  transport: {
    baseUrl: "https://api.kimi.com/coding/v1/messages",
    format: "claude",
    urlSuffix: "?beta=true",
    headers: { ...CLAUDE_API_HEADERS },
+    clientId: "17e5f671-d194-4dfb-9706-5516cb48c098",
+    tokenUrl: "https://auth.kimi.com/api/oauth/token",
+    refreshUrl: "https://auth.kimi.com/api/oauth/token",
    auth: {
      combined: true,
      header: "x-api-key",
      scheme: "raw",
+      hooks: ["kimiHeaders"],
    },
  },
  // Multi-endpoint: pick the transport matching client sourceFormat to skip translation.
@@ -31,26 +43,50 @@ export default {
    {
      format: "openai",
      baseUrl: "https://api.kimi.com/coding/v1/chat/completions",
-      auth: { combined: true, header: "Authorization", scheme: "bearer" },
+      auth: { combined: true, header: "Authorization", scheme: "bearer", hooks: ["kimiHeaders"] },
    },
    {
      format: "claude",
      baseUrl: "https://api.kimi.com/coding/v1/messages",
      urlSuffix: "?beta=true",
      headers: { ...CLAUDE_API_HEADERS },
-      auth: { combined: true, header: "x-api-key", scheme: "raw" },
+      auth: { combined: true, header: "x-api-key", scheme: "raw", hooks: ["kimiHeaders"] },
    },
  ],
  models: [
+    // Flagship K3 — platform.kimi.ai id `kimi-k3`, Kimi Code OAuth id `k3` (up to 1M)
+    { id: "kimi-k3", name: "Kimi K3" },
+    { id: "k3", name: "Kimi K3 (Code)" },
+    // Kimi Code subscription stable ids (map to K2.7 Code backend)
+    { id: "kimi-for-coding", name: "Kimi for Coding" },
+    { id: "kimi-for-coding-highspeed", name: "Kimi for Coding Highspeed" },
+    // Pay-as-you-go platform ids
+    { id: "kimi-k2.7-code", name: "Kimi K2.7 Code" },
+    { id: "kimi-k2.7-code-highspeed", name: "Kimi K2.7 Code Highspeed" },
    { id: "kimi-k2.6", name: "Kimi K2.6" },
    { id: "kimi-k2.5", name: "Kimi K2.5" },
    { id: "kimi-k2.5-thinking", name: "Kimi K2.5 Thinking" },
    { id: "kimi-latest", name: "Kimi Latest" },
  ],
-  serviceKinds: ["llm","webSearch"],
+  serviceKinds: ["llm", "webSearch"],
  searchViaChat: {
-    defaultModel: "kimi-k2.5",
+    defaultModel: "kimi-k3",
    endpoint: "https://api.moonshot.cn/v1/chat/completions",
-    pricingUrl: "https://platform.moonshot.ai/docs/pricing/chat",
+    pricingUrl: "https://platform.kimi.ai/docs/pricing/chat",
+  },
+  oauth: {
+    clientId: "17e5f671-d194-4dfb-9706-5516cb48c098",
+    deviceCodeUrl: "https://auth.kimi.com/api/oauth/device_authorization",
+    tokenUrl: "https://auth.kimi.com/api/oauth/token",
+    refreshUrl: "https://auth.kimi.com/api/oauth/token",
+    // CLIProxyAPI refreshThresholdSeconds = 300
+    refreshLeadMs: 300000,
+    authorizeDeviceUrl: "https://www.kimi.com/code/authorize_device",
+  },
+  features: {
+    usage: true,
+    // API-key connections also hit /v1/usages (x-api-key) — need usageApikey
+    // so isUsageEligible + /api/usage allow non-oauth authType.
+    usageApikey: true,
  },
 };
--- a/open-sse/providers/registry/kiro.js
+++ b/open-sse/providers/registry/kiro.js
@@ -29,7 +29,6 @@ export default {
    headers: {
      "Content-Type": "application/json",
      Accept: "application/vnd.amazon.eventstream",
-      "X-Amz-Target": "AmazonCodeWhispererStreamingService.GenerateAssistantResponse",
      "User-Agent": "AWS-SDK-JS/3.0.0 kiro-ide/1.0.0",
      "X-Amz-User-Agent": "aws-sdk-js/3.0.0 kiro-ide/1.0.0",
    },
@@ -42,18 +41,57 @@ export default {
    },
  },
  models: [
+    // Opus (added per kiro.dev/changelog/models and kiro.dev/docs/models)
+    { id: "claude-opus-5", name: "Claude Opus 5" },
+    { id: "claude-opus-5-thinking", name: "Claude Opus 5 (Thinking)" },
+    { id: "claude-opus-5-agentic", name: "Claude Opus 5 (Agentic)" },
+    { id: "claude-opus-5-thinking-agentic", name: "Claude Opus 5 (Thinking + Agentic)" },
+    { id: "claude-opus-4.8", name: "Claude Opus 4.8" },
+    { id: "claude-opus-4.8-thinking", name: "Claude Opus 4.8 (Thinking)" },
+    { id: "claude-opus-4.8-agentic", name: "Claude Opus 4.8 (Agentic)" },
+    { id: "claude-opus-4.8-thinking-agentic", name: "Claude Opus 4.8 (Thinking + Agentic)" },
+    { id: "claude-opus-4.7", name: "Claude Opus 4.7" },
+    { id: "claude-opus-4.7-thinking", name: "Claude Opus 4.7 (Thinking)" },
+    { id: "claude-opus-4.7-agentic", name: "Claude Opus 4.7 (Agentic)" },
+    { id: "claude-opus-4.7-thinking-agentic", name: "Claude Opus 4.7 (Thinking + Agentic)" },
+    { id: "claude-opus-4.5", name: "Claude Opus 4.5" },
+    { id: "claude-opus-4.5-thinking", name: "Claude Opus 4.5 (Thinking)" },
+    { id: "claude-opus-4.5-agentic", name: "Claude Opus 4.5 (Agentic)" },
+    { id: "claude-opus-4.5-thinking-agentic", name: "Claude Opus 4.5 (Thinking + Agentic)" },
+    // Sonnet
+    { id: "claude-sonnet-5", name: "Claude Sonnet 5" },
    { id: "claude-sonnet-4.5", name: "Claude Sonnet 4.5" },
+    // Haiku
    { id: "claude-haiku-4.5", name: "Claude Haiku 4.5" },
+    // Non-Anthropic
    { id: "deepseek-3.2", name: "DeepSeek 3.2", strip: ["image","audio"] },
    { id: "qwen3-coder-next", name: "Qwen3 Coder Next", strip: ["image","audio"] },
    { id: "glm-5", name: "GLM 5" },
    { id: "MiniMax-M2.5", name: "MiniMax M2.5" },
+    { id: "gpt-5.6-sol", name: "GPT 5.6 Sol", contextLength: 272000, rateMultiplier: 2.4, upstreamModelId: "gpt-5.6-sol", description: "Experimental preview of OpenAI GPT 5.6 Sol with 272k context window" },
+    { id: "gpt-5.6-terra", name: "GPT 5.6 Terra", contextLength: 272000, rateMultiplier: 1.2, upstreamModelId: "gpt-5.6-terra", description: "Experimental preview of OpenAI GPT 5.6 Terra with 272k context window" },
+    { id: "gpt-5.6-luna", name: "GPT 5.6 Luna", contextLength: 272000, rateMultiplier: 0.6, upstreamModelId: "gpt-5.6-luna", description: "Experimental preview of OpenAI GPT 5.6 Luna with 272k context window" },
+    // Thinking variants
+    { id: "claude-sonnet-5-thinking", name: "Claude Sonnet 5 (Thinking)" },
    { id: "claude-sonnet-4.5-thinking", name: "Claude Sonnet 4.5 (Thinking)" },
    { id: "claude-haiku-4.5-thinking", name: "Claude Haiku 4.5 (Thinking)" },
+    { id: "gpt-5.6-sol-thinking", name: "GPT 5.6 Sol (Thinking)", contextLength: 272000, rateMultiplier: 2.4, upstreamModelId: "gpt-5.6-sol", description: "Experimental preview of OpenAI GPT 5.6 Sol with 272k context window" },
+    { id: "gpt-5.6-terra-thinking", name: "GPT 5.6 Terra (Thinking)", contextLength: 272000, rateMultiplier: 1.2, upstreamModelId: "gpt-5.6-terra", description: "Experimental preview of OpenAI GPT 5.6 Terra with 272k context window" },
+    { id: "gpt-5.6-luna-thinking", name: "GPT 5.6 Luna (Thinking)", contextLength: 272000, rateMultiplier: 0.6, upstreamModelId: "gpt-5.6-luna", description: "Experimental preview of OpenAI GPT 5.6 Luna with 272k context window" },
+    // Agentic variants
+    { id: "claude-sonnet-5-agentic", name: "Claude Sonnet 5 (Agentic)" },
    { id: "claude-sonnet-4.5-agentic", name: "Claude Sonnet 4.5 (Agentic)" },
    { id: "claude-haiku-4.5-agentic", name: "Claude Haiku 4.5 (Agentic)" },
+    { id: "gpt-5.6-sol-agentic", name: "GPT 5.6 Sol (Agentic)", contextLength: 272000, rateMultiplier: 2.4, upstreamModelId: "gpt-5.6-sol", description: "Experimental preview of OpenAI GPT 5.6 Sol with 272k context window" },
+    { id: "gpt-5.6-terra-agentic", name: "GPT 5.6 Terra (Agentic)", contextLength: 272000, rateMultiplier: 1.2, upstreamModelId: "gpt-5.6-terra", description: "Experimental preview of OpenAI GPT 5.6 Terra with 272k context window" },
+    { id: "gpt-5.6-luna-agentic", name: "GPT 5.6 Luna (Agentic)", contextLength: 272000, rateMultiplier: 0.6, upstreamModelId: "gpt-5.6-luna", description: "Experimental preview of OpenAI GPT 5.6 Luna with 272k context window" },
+    // Thinking + Agentic variants
+    { id: "claude-sonnet-5-thinking-agentic", name: "Claude Sonnet 5 (Thinking + Agentic)" },
    { id: "claude-sonnet-4.5-thinking-agentic", name: "Claude Sonnet 4.5 (Thinking + Agentic)" },
    { id: "claude-haiku-4.5-thinking-agentic", name: "Claude Haiku 4.5 (Thinking + Agentic)" },
+    { id: "gpt-5.6-sol-thinking-agentic", name: "GPT 5.6 Sol (Thinking + Agentic)", contextLength: 272000, rateMultiplier: 2.4, upstreamModelId: "gpt-5.6-sol", description: "Experimental preview of OpenAI GPT 5.6 Sol with 272k context window" },
+    { id: "gpt-5.6-terra-thinking-agentic", name: "GPT 5.6 Terra (Thinking + Agentic)", contextLength: 272000, rateMultiplier: 1.2, upstreamModelId: "gpt-5.6-terra", description: "Experimental preview of OpenAI GPT 5.6 Terra with 272k context window" },
+    { id: "gpt-5.6-luna-thinking-agentic", name: "GPT 5.6 Luna (Thinking + Agentic)", contextLength: 272000, rateMultiplier: 0.6, upstreamModelId: "gpt-5.6-luna", description: "Experimental preview of OpenAI GPT 5.6 Luna with 272k context window" },
  ],
  oauth: {
    ssoOidcEndpoint: "https://oidc.us-east-1.amazonaws.com",
--- a/open-sse/providers/registry/llm7.js
+++ b/open-sse/providers/registry/llm7.js
@@ -0,0 +1,35 @@
+export default {
+  id: "llm7",
+  alias: "llm7",
+  aliases: [
+    "llm-7",
+  ],
+  uiAlias: "llm7",
+  display: {
+    name: "LLM7",
+    icon: "pool",
+    color: "#7C3AED",
+    textIcon: "L7",
+    website: "https://llm7.io",
+    notice: {
+      apiKeyUrl: "https://llm7.io",
+    },
+  },
+  category: "apikey",
+  authType: "apikey",
+  authModes: [
+    "apikey",
+  ],
+  transport: {
+    baseUrl: "https://api.llm7.io/v1/chat/completions",
+    validateUrl: "https://api.llm7.io/v1/models",
+  },
+  models: [
+    { id: "gpt-5.5", name: "GPT-5.5 (LLM7)", contextLength: 1050000 },
+    { id: "claude-opus-5", name: "Claude Opus 5 (LLM7)", contextLength: 1000000 },
+    { id: "deepseek-v4-flash", name: "DeepSeek V4 Flash (LLM7)", contextLength: 1000000 },
+    { id: "grok-4.5", name: "Grok 4.5 (LLM7)", contextLength: 500000 },
+    { id: "kimi-k3", name: "Kimi K3 (LLM7)", contextLength: 1000000 },
+  ],
+  passthroughModels: true,
+};
--- a/open-sse/providers/registry/mimo-free.js
+++ b/open-sse/providers/registry/mimo-free.js
@@ -1,5 +1,8 @@
+// Xiaomi ended the free MiMo channel ("MiMo free API service has ended").
+// Hidden until/unless a replacement (OAuth MiMo Platform) is wired.
 export default {
  id: "mimo-free",
+  hidden: true,
  priority: 50,
  hasFree: true,
  alias: "mmf",
--- a/open-sse/providers/registry/morph.js
+++ b/open-sse/providers/registry/morph.js
@@ -0,0 +1,29 @@
+export default {
+  id: "morph",
+  alias: "morph",
+  aliases: ["morphllm"],
+  uiAlias: "morph",
+  display: {
+    name: "Morph",
+    icon: "change_history",
+    color: "#14B8A6",
+    textIcon: "MP",
+    website: "https://morphllm.com",
+    notice: { apiKeyUrl: "https://morphllm.com" },
+  },
+  category: "apikey",
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://api.morphllm.com/v1/chat/completions",
+    validateUrl: "https://api.morphllm.com/v1/models",
+  },
+  models: [
+    { id: "morph-v3-large", name: "Morph v3 Large" },
+    { id: "morph-v3-fast", name: "Morph v3 Fast" },
+    { id: "morph-qwen35-397b", name: "Qwen 3.5 397B (Morph)", contextLength: 262144 },
+    { id: "morph-minimax27-230b", name: "MiniMax M2.7 (Morph)", contextLength: 200704 },
+    { id: "morph-qwen36-27b", name: "Qwen 3.6 27B (Morph)", contextLength: 262144 },
+    { id: "morph-dsv4flash", name: "DeepSeek V4 Flash (Morph)", contextLength: 1048576 },
+  ],
+};
--- a/open-sse/providers/registry/nvidia.js
+++ b/open-sse/providers/registry/nvidia.js
@@ -15,13 +15,20 @@ export default {
    },
  },
  category: "freeTier",
+  authType: "apikey",
+  authModes: ["apikey"],
  transport: {
    baseUrl: "https://integrate.api.nvidia.com/v1/chat/completions",
    validateUrl: "https://integrate.api.nvidia.com/v1/models",
  },
  models: [
-    { id: "minimaxai/minimax-m2.7", name: "Minimax M2.7" },
-    { id: "z-ai/glm4.7", name: "GLM 4.7" },
+    { id: "minimaxai/minimax-m2.7", name: "MiniMax M2.7" },
+    { id: "minimaxai/minimax-m3", name: "MiniMax M3" },
+    { id: "z-ai/glm-5.2", name: "GLM 5.2" },
+    { id: "deepseek-ai/deepseek-v4-pro", name: "DeepSeek V4 Pro" },
+    { id: "deepseek-ai/deepseek-v4-flash", name: "DeepSeek V4 Flash" },
+    { id: "moonshotai/kimi-k2.6", name: "Kimi K2.6" },
+    { id: "nvidia/nemotron-3-ultra-550b-a55b", name: "Nemotron 3 Ultra" },
    { id: "nvidia/nv-embedqa-e5-v5", name: "NV EmbedQA E5 v5", kind: "embedding" },
    { id: "nvidia/parakeet-ctc-1.1b-asr", name: "Parakeet CTC 1.1B", params: ["language"], kind: "stt" },
    { id: "fastpitch", name: "FastPitch", kind: "tts" },
--- a/open-sse/providers/registry/openrouter.js
+++ b/open-sse/providers/registry/openrouter.js
@@ -15,6 +15,8 @@ export default {
    },
  },
  category: "freeTier",
+  authType: "apikey",
+  authModes: ["apikey"],
  transport: {
    baseUrl: "https://openrouter.ai/api/v1/chat/completions",
    thinkingFormat: "openai",
--- a/open-sse/providers/registry/perplexity-agent.js
+++ b/open-sse/providers/registry/perplexity-agent.js
@@ -0,0 +1,49 @@
+export default {
+  id: "perplexity-agent",
+  priority: 181,
+  alias: "perplexity-agent",
+  aliases: [
+    "pplx-agent",
+    "pplx-responses",
+  ],
+  uiAlias: "pa",
+  display: {
+    name: "Perplexity Agent",
+    icon: "travel_explore",
+    color: "#20808D",
+    textIcon: "PA",
+    website: "https://www.perplexity.ai",
+    notice: {
+      text: "Perplexity Agent API exposes GPT, Claude, Gemini, Grok, GLM, Kimi, and Sonar models through one OpenAI-compatible Responses API.",
+      apiKeyUrl: "https://www.perplexity.ai/settings/api",
+    },
+  },
+  category: "apikey",
+  authType: "apikey",
+  transport: {
+    baseUrl: "https://api.perplexity.ai/v1/responses",
+    validateUrl: "https://api.perplexity.ai/v1/models",
+    format: "openai-responses",
+  },
+  models: [
+    { id: "perplexity/sonar", name: "Perplexity Sonar" },
+    { id: "openai/gpt-5.5", name: "GPT-5.5" },
+    { id: "openai/gpt-5.4", name: "GPT-5.4" },
+    { id: "openai/gpt-5.4-mini", name: "GPT-5.4 Mini" },
+    { id: "anthropic/claude-sonnet-4-6", name: "Claude Sonnet 4.6" },
+    { id: "anthropic/claude-opus-4-8", name: "Claude Opus 4.8" },
+    { id: "google/gemini-3.1-pro-preview", name: "Gemini 3.1 Pro" },
+    { id: "xai/grok-4.20-reasoning", name: "Grok 4.20 Reasoning" },
+    { id: "perplexity/glm-5.2", name: "GLM 5.2" },
+    { id: "perplexity/kimi-k2.7-code", name: "Kimi K2.7 Code" },
+    { id: "nvidia/nemotron-3-super-120b-a12b", name: "Nemotron 3 Super 120B" },
+  ],
+  serviceKinds: ["llm", "webSearch"],
+  searchViaChat: {
+    defaultModel: "perplexity/sonar",
+    endpoint: "https://api.perplexity.ai/v1/responses",
+    pricingUrl: "https://docs.perplexity.ai/docs/agent-api/models",
+  },
+  modelsFetcher: { url: "https://api.perplexity.ai/v1/models", type: "openai" },
+  passthroughModels: true,
+};
--- a/open-sse/providers/registry/poolside.js
+++ b/open-sse/providers/registry/poolside.js
@@ -0,0 +1,30 @@
+export default {
+  id: "poolside",
+  priority: 60,
+  alias: "poolside",
+  aliases: [
+    "ps",
+  ],
+  uiAlias: "ps",
+  display: {
+    name: "Poolside",
+    icon: "water_drop",
+    color: "#0EA5E9",
+    textIcon: "PS",
+    website: "https://poolside.ai",
+    notice: {
+      apiKeyUrl: "https://platform.poolside.ai/api-keys",
+    },
+  },
+  category: "freeTier",
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://inference.poolside.ai/v1/chat/completions",
+    validateUrl: "https://inference.poolside.ai/v1/models",
+  },
+  models: [
+    { id: "poolside/laguna-s-2.1", name: "Laguna S 2.1" },
+    { id: "poolside/laguna-xs-2.1", name: "Laguna XS 2.1" },
+  ],
+};
--- a/open-sse/providers/registry/qoder.js
+++ b/open-sse/providers/registry/qoder.js
@@ -11,10 +11,11 @@ export default {
    notice: {
      signupUrl: "https://qoder.com",
    },
-    deprecated: true,
-    deprecationNotice: "RISK_NOTICE",
  },
-  category: "free",
+  category: "oauth",
+  authModes: ["oauth", "apikey"],
+  hasOAuth: true,
+  authHint: "Personal Access Token (pt-...) từ https://qoder.com/account/integrations",
  transport: {
    baseUrl: "https://api3.qoder.sh/algo/api/v2/service/pro/sse/agent_chat_generation",
    headers: {},
@@ -25,18 +26,19 @@ export default {
    },
  },
  models: [
-    // { id: "auto", name: "Qoder Auto" },
-    // { id: "ultimate", name: "Qoder Ultimate" },
-    // { id: "performance", name: "Qoder Performance" },
-    // { id: "efficient", name: "Qoder Efficient" },
-    // { id: "lite", name: "Qoder Lite" },
-    // { id: "qmodel", name: "Qwen 3.6 Plus (Qoder)" },
-    { id: "qmodel_latest", name: "Qoder Qwen 3.7 Max" },
-    // { id: "dmodel", name: "DeepSeek V4 Pro (Qoder)" },
-    // { id: "dfmodel", name: "DeepSeek V4 Flash (Qoder)" },
-    // { id: "gm51model", name: "GLM 5.1 (Qoder)" },
-    // { id: "kmodel", name: "Kimi K2.6 (Qoder)" },
-    // { id: "mmodel", name: "MiniMax M2.7 (Qoder)" },
+    { id: "ultimate", name: "Ultimate" },
+    { id: "auto", name: "Auto" },
+    { id: "performance", name: "Performance" },
+    { id: "efficient", name: "Efficient" },
+    { id: "qmodel_preview", name: "Qwen3.8-Max-Preview" },
+    { id: "qmodel_latest", name: "Qwen3.7-Max" },
+    { id: "qmodel", name: "Qwen3.7-Plus" },
+    { id: "kmodel_latest", name: "Kimi-K3" },
+    { id: "kmodel", name: "Kimi-K2.7-Code" },
+    { id: "gm51model", name: "GLM-5.2" },
+    { id: "dmodel", name: "DeepSeek-V4-Pro" },
+    { id: "dfmodel", name: "DeepSeek-V4-Flash" },
+    { id: "mmodel", name: "MiniMax-M3" },
  ],
  oauth: {
    openApiBaseUrl: "https://openapi.qoder.sh",
--- a/open-sse/providers/registry/sambanova.js
+++ b/open-sse/providers/registry/sambanova.js
@@ -0,0 +1,27 @@
+export default {
+  id: "sambanova",
+  alias: "samba",
+  aliases: ["sambanova-ai"],
+  uiAlias: "samba",
+  hidden: true,
+  display: {
+    name: "SambaNova",
+    icon: "memory",
+    color: "#F97316",
+    textIcon: "SN",
+    website: "https://sambanova.ai",
+    notice: {
+      apiKeyUrl: "https://cloud.sambanova.ai/apis",
+    },
+  },
+  category: "apikey",
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://api.sambanova.ai/v1/chat/completions",
+    validateUrl: "https://api.sambanova.ai/v1/models",
+  },
+  models: [
+    { id: "MiniMax-M2.7", name: "MiniMax M2.7", contextLength: 196608 },
+  ],
+};
--- a/open-sse/providers/registry/searxng.js
+++ b/open-sse/providers/registry/searxng.js
@@ -1,3 +1,5 @@
+import { SEARXNG_URL } from "../../config/runtimeConfig.js";
+
 export default {
  id: "searxng",
  alias: "searxng",
@@ -15,7 +17,7 @@ export default {
  ],
  noAuth: true,
  searchConfig: {
-    baseUrl: "http://localhost:8888/search",
+    baseUrl: SEARXNG_URL,
    method: "GET",
    authType: "none",
    authHeader: "none",
--- a/open-sse/providers/registry/tencent.js
+++ b/open-sse/providers/registry/tencent.js
@@ -0,0 +1,27 @@
+export default {
+  id: "tencent",
+  alias: "hunyuan",
+  aliases: ["hunyuan", "tencent-hunyuan"],
+  uiAlias: "hunyuan",
+  display: {
+    name: "Tencent Hunyuan",
+    icon: "cloud",
+    color: "#0052D9",
+    textIcon: "HY",
+    website: "https://cloud.tencent.com/product/hunyuan",
+    notice: {
+      apiKeyUrl: "https://console.cloud.tencent.com/hunyuan/api-key",
+    },
+  },
+  category: "apikey",
+  authType: "apikey",
+  authModes: ["apikey"],
+  transport: {
+    baseUrl: "https://api.hunyuan.cloud.tencent.com/v1/chat/completions",
+    validateUrl: "https://api.hunyuan.cloud.tencent.com/v1/models",
+  },
+  models: [
+    { id: "hunyuan-turbos-latest", name: "Hunyuan TurboS Latest", contextLength: 200000 },
+    { id: "hunyuan-t1-latest", name: "Hunyuan T1 Latest", contextLength: 256000 },
+  ],
+};
--- a/open-sse/providers/registry/trae.js
+++ b/open-sse/providers/registry/trae.js
@@ -0,0 +1,76 @@
+// Trae (ByteDance marscode) provider registry entry.
+// Chat = SOLO remote agent API:
+//   POST {base}/chat_sessions → {data:{chat_session_id, message_id}}
+//   GET  {base}/chat_sessions/{id}/events?reply_to_message_id=... → SSE
+//   Auth: Authorization: Cloud-IDE-JWT <jwt>
+export default {
+  id: "trae",
+  alias: "tr",
+  uiAlias: "tr",
+  aliases: ["marscode"],
+  category: "oauth",
+  authType: "oauth",
+  hasOAuth: true,
+  authModes: ["oauth"],
+  display: {
+    name: "Trae",
+    icon: "bolt",
+    color: "#FF6A00",
+    textIcon: "TR",
+    website: "https://www.trae.ai",
+    notice: { signupUrl: "https://www.trae.ai" },
+  },
+  transport: {
+    // SOLO remote agent base — verified working chat endpoint.
+    baseUrl: "https://core-normal.trae.ai/api/remote/v1",
+    format: "openai",
+    headers: {
+      "X-Trae-Client-Type": "web",
+      "X-Preferenced-Language": "en",
+      "Referer": "https://solo.trae.ai/",
+    },
+    // Auth: Cloud-IDE-JWT scheme on Authorization — injected by executor buildHeaders.
+    auth: {
+      combined: true,
+      header: "Authorization",
+      scheme: "Cloud-IDE-JWT",
+    },
+    usage: {
+      url: "https://api.marscode.com/cloudide/api/v3/trae/GetUserInfo",
+    },
+    regions: {
+      cn: "https://api.marscode.com",
+      sg: "https://api.trae.ai",
+      us: "https://www.trae.ai",
+    },
+    defaultRegion: "cn",
+  },
+  oauth: {
+    clientId: "ono9krqynydwx5",
+    clientSecret: "-",
+    platform: "trae",
+    pollInterval: 1500,
+    // Login guidance returns LoginHost for browser open.
+    loginGuidanceUrl: "https://api.marscode.com/cloudide/api/v3/trae/GetLoginGuidance",
+    // ExchangeToken: refresh -> access (POST JSON, body below).
+    tokenUrl: "https://api.marscode.com/cloudide/api/v3/trae/oauth/ExchangeToken",
+    exchangeTokenUrl: "https://api.marscode.com/cloudide/api/v3/trae/oauth/ExchangeToken",
+    refreshUrl: "https://api.marscode.com/cloudide/api/v3/trae/oauth/ExchangeToken",
+    userInfoUrl: "https://api.marscode.com/cloudide/api/v3/trae/GetUserInfo",
+    // Trae refresh uses custom JSON body, not OAuth form — handled by refresh.js, not config-driven.
+    refresh: { encoding: "json" },
+  },
+  // Model catalog (IDE flow, core-normal.trae.ai).
+  models: [
+    { id: "auto", name: "Auto (Server Picks)" },
+    { id: "work", name: "Work (Fast)" },
+    { id: "gemini-3.1-pro", name: "Gemini 3.1 Pro" },
+    { id: "gemini-3-flash-solo", name: "Gemini 3 Flash" },
+    { id: "minimax-m3", name: "MiniMax M3" },
+    { id: "minimax-m2.7", name: "MiniMax M2.7" },
+    { id: "kimi-k2.5", name: "Kimi K2.5" },
+    { id: "gpt-5.4", name: "GPT 5.4" },
+    { id: "gpt-5.2", name: "GPT 5.2" },
+  ],
+  features: { usage: true },
+};
--- a/open-sse/providers/registry/windsurf.js
+++ b/open-sse/providers/registry/windsurf.js
@@ -0,0 +1,143 @@
+// Windsurf provider registry — Firebase+Codeium+Devin auth chain.
+// Chat = Codeium gRPC-web protobuf:
+//   POST {base}  Content-Type: application/grpc-web+proto
+//   Service: exa.language_server_pb.LanguageServerService / GetChatMessage
+export default {
+  id: "windsurf",
+  alias: "ws",
+  uiAlias: "ws",
+  display: {
+    name: "Windsurf",
+    icon: "surfing",
+    color: "#14B8A6",
+    website: "https://windsurf.com",
+    notice: { signupUrl: "https://windsurf.com" },
+  },
+  category: "oauth",
+  authType: "oauth",
+  hasOAuth: true,
+  authModes: ["oauth", "apikey"],
+
+  transport: {
+    baseUrl: "https://server.codeium.com/exa.language_server_pb.LanguageServerService/GetChatMessage",
+    format: "openai",
+    headers: {
+      "Content-Type": "application/grpc-web+proto",
+      "Accept": "application/grpc-web+proto",
+      "X-Grpc-Web": "1",
+    },
+    // apiKey (sk-ws-... or Firebase-derived) as Bearer + in protobuf Metadata.api_key.
+    auth: { combined: true, header: "Authorization", scheme: "Bearer" },
+  },
+
+  // Auth chain (4 terminal paths, all yield apiKey):
+  //  1) OAuth web → Firebase JWT → POST register.windsurf.com/.../RegisterUser {firebase_id_token} → {apiKey, apiServerUrl, name}
+  //  2) sk-ws-... direct API key (apiKey used as metadata.apiKey on GetUserStatus)
+  //  3) Firebase JWT (eyJ...) → same RegisterUser exchange as #1
+  //  4) Devin auth1_... → self-serve chain → ide_token used as apiKey on server.self-serve.windsurf.com
+  oauth: {
+    clientId: "3GUryQ7ldAeKEuD2obYnppsnmj58eP5u",
+    firebaseApiKey: "AIzaSyDsOl-1XpT5err0Tcn0TFFod1H8gVGIycY",
+    firebaseSignInUrl: "https://identitytoolkit.googleapis.com/v1/accounts:signInWithPassword",
+    registerUrl: "https://register.windsurf.com/exa.seat_management_pb.SeatManagementService/RegisterUser",
+    apiServerUrl: "https://server.codeium.com",
+    auth1ApiServerUrl: "https://server.self-serve.windsurf.com",
+    platform: "windsurf",
+    // Quota (Connect RPC, protobuf): POST windsurf.com/_backend/.../GetPlanStatus,
+    // headers Content-Type:application/proto + Connect-Protocol-Version:1 + X-Auth-Token:<session>,
+    // body = field1:session_token, field2:varint 1.
+    quotaUrl: "https://windsurf.com/_backend/exa.seat_management_pb.SeatManagementService/GetPlanStatus",
+  },
+
+  // Catalog verified against model_configs_v2.bin from Devin CLI (2026.5.x).
+  // Dot-notation ids; the executor MODEL_ALIAS_MAP maps these to Windsurf modelUid.
+  // contextLength dropped — 9router schema uses id+name only.
+  models: [
+    // Cognition / SWE
+    { id: "swe-1.6-fast", name: "SWE-1.6 Fast" },
+    { id: "swe-1.6", name: "SWE-1.6" },
+    { id: "swe-1.5-fast", name: "SWE-1.5 Fast" },
+    { id: "swe-1.5", name: "SWE-1.5" },
+    // Claude Opus 4.7 — effort-tiered
+    { id: "claude-opus-4.7-max", name: "Claude Opus 4.7 Max" },
+    { id: "claude-opus-4.7-xhigh", name: "Claude Opus 4.7 XHigh" },
+    { id: "claude-opus-4.7-high", name: "Claude Opus 4.7 High" },
+    { id: "claude-opus-4.7-medium", name: "Claude Opus 4.7 Medium" },
+    { id: "claude-opus-4.7-low", name: "Claude Opus 4.7 Low" },
+    { id: "claude-opus-4.7-review", name: "Claude Opus 4.7 Review" },
+    // Claude Sonnet/Opus 4.6
+    { id: "claude-sonnet-4.6-thinking-1m", name: "Claude Sonnet 4.6 Thinking 1M" },
+    { id: "claude-sonnet-4.6-1m", name: "Claude Sonnet 4.6 1M" },
+    { id: "claude-sonnet-4.6-thinking", name: "Claude Sonnet 4.6 Thinking" },
+    { id: "claude-sonnet-4.6", name: "Claude Sonnet 4.6" },
+    { id: "claude-opus-4.6-thinking", name: "Claude Opus 4.6 Thinking" },
+    { id: "claude-opus-4.6", name: "Claude Opus 4.6" },
+    // Claude 4.5
+    { id: "claude-opus-4.5-thinking", name: "Claude Opus 4.5 Thinking" },
+    { id: "claude-opus-4.5", name: "Claude Opus 4.5" },
+    { id: "claude-sonnet-4.5-thinking", name: "Claude Sonnet 4.5 Thinking" },
+    { id: "claude-sonnet-4.5", name: "Claude Sonnet 4.5" },
+    { id: "claude-haiku-4.5", name: "Claude Haiku 4.5" },
+    // GPT-5.5 — effort-tiered
+    { id: "gpt-5.5-xhigh-fast", name: "GPT-5.5 XHigh Fast" },
+    { id: "gpt-5.5-xhigh", name: "GPT-5.5 XHigh" },
+    { id: "gpt-5.5-high-fast", name: "GPT-5.5 High Fast" },
+    { id: "gpt-5.5-high", name: "GPT-5.5 High" },
+    { id: "gpt-5.5-medium-fast", name: "GPT-5.5 Medium Fast" },
+    { id: "gpt-5.5-medium", name: "GPT-5.5 Medium" },
+    { id: "gpt-5.5-low-fast", name: "GPT-5.5 Low Fast" },
+    { id: "gpt-5.5-low", name: "GPT-5.5 Low" },
+    { id: "gpt-5.5-none-fast", name: "GPT-5.5 None Fast" },
+    { id: "gpt-5.5-none", name: "GPT-5.5 None" },
+    // GPT-5.4 — effort-tiered
+    { id: "gpt-5.4-xhigh-fast", name: "GPT-5.4 XHigh Fast" },
+    { id: "gpt-5.4-xhigh", name: "GPT-5.4 XHigh" },
+    { id: "gpt-5.4-high-fast", name: "GPT-5.4 High Fast" },
+    { id: "gpt-5.4-high", name: "GPT-5.4 High" },
+    { id: "gpt-5.4-medium-fast", name: "GPT-5.4 Medium Fast" },
+    { id: "gpt-5.4-medium", name: "GPT-5.4 Medium" },
+    { id: "gpt-5.4-low-fast", name: "GPT-5.4 Low Fast" },
+    { id: "gpt-5.4-low", name: "GPT-5.4 Low" },
+    { id: "gpt-5.4-none-fast", name: "GPT-5.4 None Fast" },
+    { id: "gpt-5.4-none", name: "GPT-5.4 None" },
+    { id: "gpt-5.4-mini-xhigh", name: "GPT-5.4 Mini XHigh" },
+    { id: "gpt-5.4-mini-high", name: "GPT-5.4 Mini High" },
+    { id: "gpt-5.4-mini-medium", name: "GPT-5.4 Mini Medium" },
+    { id: "gpt-5.4-mini-low", name: "GPT-5.4 Mini Low" },
+    // GPT-5.3 Codex
+    { id: "gpt-5.3-codex-xhigh-fast", name: "GPT-5.3 Codex XHigh Fast" },
+    { id: "gpt-5.3-codex-xhigh", name: "GPT-5.3 Codex XHigh" },
+    { id: "gpt-5.3-codex-high-fast", name: "GPT-5.3 Codex High Fast" },
+    { id: "gpt-5.3-codex-high", name: "GPT-5.3 Codex High" },
+    { id: "gpt-5.3-codex-medium-fast", name: "GPT-5.3 Codex Medium Fast" },
+    { id: "gpt-5.3-codex-medium", name: "GPT-5.3 Codex Medium" },
+    { id: "gpt-5.3-codex-low-fast", name: "GPT-5.3 Codex Low Fast" },
+    { id: "gpt-5.3-codex-low", name: "GPT-5.3 Codex Low" },
+    // GPT-5.2 / 5
+    { id: "gpt-5.2-xhigh", name: "GPT-5.2 XHigh" },
+    { id: "gpt-5.2-high", name: "GPT-5.2 High" },
+    { id: "gpt-5.2-medium", name: "GPT-5.2 Medium" },
+    { id: "gpt-5.2-low", name: "GPT-5.2 Low" },
+    { id: "gpt-5.2-none", name: "GPT-5.2 None" },
+    { id: "gpt-5", name: "GPT-5" },
+    // GPT-4.1 / 4o
+    { id: "gpt-4.1", name: "GPT-4.1" },
+    { id: "gpt-4.1-mini", name: "GPT-4.1 Mini" },
+    { id: "gpt-4.1-nano", name: "GPT-4.1 Nano" },
+    { id: "gpt-4o", name: "GPT-4o" },
+    { id: "gpt-4o-mini", name: "GPT-4o Mini" },
+    // Gemini
+    { id: "gemini-3.1-pro-high", name: "Gemini 3.1 Pro High" },
+    { id: "gemini-3.1-pro-low", name: "Gemini 3.1 Pro Low" },
+    { id: "gemini-3.0-flash-high", name: "Gemini 3 Flash High" },
+    { id: "gemini-3.0-flash-medium", name: "Gemini 3 Flash Medium" },
+    { id: "gemini-3.0-flash-low", name: "Gemini 3 Flash Low" },
+    { id: "gemini-3.0-flash-minimal", name: "Gemini 3 Flash Minimal" },
+    { id: "gemini-2.5-pro", name: "Gemini 2.5 Pro" },
+    // Others
+    { id: "deepseek-v4", name: "DeepSeek V4" },
+    { id: "kimi-k2.6", name: "Kimi K2.6" },
+    { id: "kimi-k2.5", name: "Kimi K2.5" },
+    { id: "glm-5.1", name: "GLM-5.1" },
+  ],
+};
--- a/open-sse/providers/registry/xai.js
+++ b/open-sse/providers/registry/xai.js
@@ -25,19 +25,57 @@ export default {
    clientId: "b1a00492-073a-47ea-816f-4c329264a828",
    tokenUrl: "https://auth.x.ai/oauth2/token",
    refreshUrl: "https://auth.x.ai/oauth2/token",
+    // OAuth-only SuperGrok quota surfaces:
+    // - url: monthly API usage allotment (JSON)
+    // - creditsUrl: weekly SuperGrok limit (grpc-web)
+    // - settingsUrl: plan label (subscription_tier_display)
+    usage: {
+      url: "https://cli-chat-proxy.grok.com/v1/billing",
+      creditsUrl: "https://grok.com/grok_api_v2.GrokBuildBilling/GetGrokCreditsConfig",
+      settingsUrl: "https://cli-chat-proxy.grok.com/v1/settings",
+    },
  },
  models: [
    { id: "grok-4", name: "Grok 4" },
    { id: "grok-4-fast-reasoning", name: "Grok 4 Fast Reasoning" },
    { id: "grok-code-fast-1", name: "Grok Code Fast" },
    { id: "grok-3", name: "Grok 3" },
-    { id: "grok-2-image-1212", name: "Grok 2 Image", params: ["n","response_format"], kind: "image" },
+    {
+      id: "grok-imagine-image-quality",
+      name: "Grok Imagine Image Quality",
+      capabilities: ["text2img", "edit"],
+      params: ["n", "aspect_ratio", "resolution", "response_format", "size"],
+      kind: "image",
+    },
+    {
+      id: "grok-2-image-1212",
+      name: "Grok 2 Image",
+      capabilities: ["text2img", "edit"],
+      params: ["n", "aspect_ratio", "resolution", "response_format", "size"],
+      kind: "image",
+    },
+    {
+      id: "grok-imagine-video",
+      name: "Grok Imagine Video",
+      params: ["duration", "aspect_ratio", "resolution"],
+      kind: "video",
+    },
  ],
-  serviceKinds: ["llm","imageToText","webSearch","image"],
-  imageConfig: { baseUrl: "https://api.x.ai/v1/images/generations", bodyFields: ["model","prompt","n","response_format"] },
+  serviceKinds: ["llm", "imageToText", "webSearch", "image", "video"],
+  imageConfig: {
+    baseUrl: "https://api.x.ai/v1/images/generations",
+    editsUrl: "https://api.x.ai/v1/images/edits",
+    bodyFields: ["model", "prompt", "n", "response_format", "aspect_ratio", "resolution", "image", "images"],
+  },
+  // Async video jobs (POST returns { request_id }, GET polls until done/failed).
+  // Docs: https://docs.x.ai/developers/rest-api-reference/inference/videos
+  videoConfig: { baseUrl: "https://api.x.ai/v1/videos" },
  searchViaChat: {
    defaultModel: "grok-4.20-reasoning",
    endpoint: "https://api.x.ai/v1/responses",
    pricingUrl: "https://x.ai/api#pricing",
  },
+  features: {
+    usage: true,
+  },
 };
--- a/open-sse/providers/registry/xiaomi-tokenplan.js
+++ b/open-sse/providers/registry/xiaomi-tokenplan.js
@@ -21,6 +21,11 @@ export default {
  },
  category: "apikey",
  hasProviderSpecificData: true,
+  regions: [
+    { id: "sgp", label: "Singapore (新加坡)" },
+    { id: "cn", label: "China (中国大陆)" },
+    { id: "ams", label: "Amsterdam (阿姆斯特丹)" },
+  ],
  defaultRegion: "sgp",
  transport: {
    baseUrl: "https://token-plan-sgp.xiaomimimo.com/v1/chat/completions",
--- a/Show More
+++ b/Show More