luulam/9router
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(qoder): report usage to all clients and stop inlining large attachments

Browse Source 
- Coalesce Qoder's empty finish-in-delta frame with the later choices:[] usage
  frame so OpenAI and Claude clients receive prompt_tokens, completion_tokens
  and cache-hit tokens (the dashboard already saw them)
- Upload inlined images through /api/v2/image/upload like qodercli, and stub
  oversized non-image files instead of stuffing 30MB+ data URIs into
  agent_chat_generation
- Emit response.completed -> response.usage for chat-native upstreams so
  /v1/responses clients (Codex CLI, sub2api) no longer log 0/0/0
- Keep Claude message_delta.usage working when usage arrives without choices[0]
- Escalate to the smallest advertised Qoder context tier (200K/400K/1M) when
  the estimated prompt no longer fits max_input_tokens
- Pass apiKey for PAT connections and list hidden enable:false catalog keys
  from /v1/models
This commit is contained in:
LLL
2026-09-10 22:06:49 +07:00
committed by decolua
parent 832a34659e
commit 1f10f9e5c4
19 changed files with 1615 additions and 121 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
open-sse/shared/qoder/constants.js
View File

@@ -33,6 +33,39 @@ export const QODER_CHAT_SIG_PATH = "/api/v2/service/pro/sse/agent_chat_generatio
export const QODER_CHAT_URL = `${QODER_CHAT_BASE}/algo${QODER_CHAT_SIG_PATH}?FetchKeys=llm_model_result&AgentId=agent_common`;
export const QODER_CHAT_URL_ENCODED = `${QODER_CHAT_URL}&Encode=1`;
export const QODER_MODEL_LIST_URL = `${QODER_CHAT_BASE}/algo/api/v2/model/list`;
// Official qodercli uploads images here (COSY-signed PUT multipart, field "file")
// instead of inlining base64 into agent_chat_generation.
export const QODER_IMAGE_UPLOAD_SIG_PATH = "/api/v2/image/upload";
// Drop remaining inlined binaries if the Qoder JSON body would still exceed this.
// 30MB+ payloads are what blow past Claude-Code's ~200k context on the wire.
export const QODER_MAX_PAYLOAD_BYTES = 6 * 1024 * 1024;
// If OSS upload fails, keep tiny data-URIs; anything larger is stubbed.
export const QODER_INLINE_FALLBACK_MAX_BYTES = 512 * 1024;
// Context-window tier selection (see shared/qoder/contextTier.js). The IDE exposes the
// model's context_config tiers (200K/400K/1M); we auto-escalate when the estimated prompt
// (+ headroom, tokenizer variance) no longer fits the current max_input_tokens.
export const QODER_CONTEXT_TIER_HEADROOM = 0.15;
export const QODER_CONTEXT_TIER_ENV = "QODER_CONTEXT_TIER";
export const QODER_CONTEXT_TIER_MODES = Object.freeze({ AUTO: "auto", MAX: "max", DEFAULT: "default" });
/**
* Job-token (jt-...) traffic must hit api2.qoder.sh — api3 rejects jt- with
* "Login expired" (403). Device tokens (dt-...) stay on api3. PATs (pt-...)
* are exchanged for jt- before this is consulted.
*/
export function qoderInferenceBase(credentials) {
const raw = credentials?.apiKey || credentials?.accessToken;
if (
typeof raw === "string" &&
!raw.startsWith("pt-") &&
(raw.startsWith("jt-") || (credentials?.accessToken || "").startsWith("jt-"))
) {
return QODER_CHAT_BASE_ALT;
}
return QODER_CHAT_BASE;
}
// COSY header constants. These are not arbitrary — the upstream signature
// validation matches them against the values used at signing time.