Enhance chat handling and introduce Caveman feature

- Refactored handleChatCore to include Caveman functionality, allowing for terse-style system prompts to reduce output token usage. - Updated APIPageClient to manage Caveman settings, including enabling/disabling and selecting compression levels. - Adjusted AntigravityExecutor to consolidate function declarations for compatibility with Gemini. - Removed unnecessary console logs during translator initialization across multiple routes.
2026-04-30 18:00:38 +07:00
parent b3fb20e237
commit 936d65ae1c
14 changed files with 265 additions and 48 deletions
--- a/src/app/(dashboard)/dashboard/endpoint/EndpointPageClient.js
+++ b/src/app/(dashboard)/dashboard/endpoint/EndpointPageClient.js
@@ -14,6 +14,12 @@ const TUNNEL_BENEFITS = [

 const TUNNEL_PING_INTERVAL_MS = 2000;
 const TUNNEL_PING_MAX_MS = 300000;
+
+const CAVEMAN_LEVELS = [
+  { id: "lite", label: "Lite", desc: "Drop filler, keep grammar" },
+  { id: "full", label: "Full", desc: "Drop articles, fragments OK" },
+  { id: "ultra", label: "Ultra", desc: "Telegraphic, max compression" },
+];
 export default function APIPageClient({ machineId }) {
  const [keys, setKeys] = useState([]);
  const [loading, setLoading] = useState(true);
@@ -26,6 +32,8 @@ export default function APIPageClient({ machineId }) {
  const [hasPassword, setHasPassword] = useState(true);
  const [tunnelDashboardAccess, setTunnelDashboardAccess] = useState(false);
  const [rtkEnabled, setRtkEnabledState] = useState(true);
+  const [cavemanEnabled, setCavemanEnabled] = useState(false);
+  const [cavemanLevel, setCavemanLevel] = useState("full");

  // Cloudflare Tunnel state
  const [tunnelChecking, setTunnelChecking] = useState(true);
@@ -82,6 +90,8 @@ export default function APIPageClient({ machineId }) {
        setHasPassword(data.hasPassword || false);
        setTunnelDashboardAccess(data.tunnelDashboardAccess || false);
        setRtkEnabledState(data.rtkEnabled !== false);
+        setCavemanEnabled(!!data.cavemanEnabled);
+        setCavemanLevel(data.cavemanLevel || "full");
      }
      if (statusRes.ok) {
        const data = await statusRes.json();
@@ -182,6 +192,28 @@ export default function APIPageClient({ machineId }) {
    }
  };

+  const patchSetting = async (patch) => {
+    try {
+      await fetch("/api/settings", {
+        method: "PATCH",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(patch),
+      });
+    } catch (error) {
+      console.log("Error updating setting:", error);
+    }
+  };
+
+  const handleCavemanEnabled = (value) => {
+    setCavemanEnabled(value);
+    patchSetting({ cavemanEnabled: value });
+  };
+
+  const handleCavemanLevel = (level) => {
+    setCavemanLevel(level);
+    patchSetting({ cavemanLevel: level });
+  };
+
  const fetchData = async () => {
    try {
      const keysRes = await fetch("/api/keys");
@@ -813,16 +845,26 @@ export default function APIPageClient({ machineId }) {
        )}
      </Card>

-      {/* Token Saver (RTK) */}
+      {/* Token Saver (RTK + Caveman) */}
      <Card id="rtk">
        <div className="flex items-center justify-between mb-2">
          <h2 className="text-lg font-semibold">Token Saver</h2>
        </div>
-        <div className="flex items-center justify-between pt-2">
+        <div className="flex items-center justify-between pt-2 pb-4 border-b border-border">
          <div className="pr-4">
-            <p className="font-medium">Compress tool output</p>
+            <p className="font-medium">
+              Compress tool output{" "}
+              <a
+                href="https://github.com/rtk-ai/rtk"
+                target="_blank"
+                rel="noreferrer"
+                className="text-xs font-normal text-primary underline hover:opacity-80"
+              >
+                (RTK)
+              </a>
+            </p>
            <p className="text-sm text-text-muted">
-              Auto-compress tool output (git diff/grep/ls/tree/logs) before sending to LLM to save tokens. Disable if you see issues.
+              Auto-compress tool output (git diff/grep/ls/tree/logs) before sending to LLM (60-90% fewer input tokens on common dev commands). Disable if you see issues.
            </p>
          </div>
          <Toggle
@@ -830,6 +872,46 @@ export default function APIPageClient({ machineId }) {
            onChange={() => handleRtkEnabled(!rtkEnabled)}
          />
        </div>
+        <div className="flex items-center justify-between pt-4">
+          <div className="pr-4">
+            <p className="font-medium">
+              Compress LLM output{" "}
+              <a
+                href="https://github.com/JuliusBrussee/caveman"
+                target="_blank"
+                rel="noreferrer"
+                className="text-xs font-normal text-primary underline hover:opacity-80"
+              >
+                (Caveman)
+              </a>
+            </p>
+            <p className="text-sm text-text-muted">
+              Inject a terse-style instruction into the system prompt so the LLM replies shorter (~65% fewer output tokens on average, up to 87%). Code, errors and warnings stay exact.
+            </p>
+          </div>
+          <Toggle
+            checked={cavemanEnabled}
+            onChange={() => handleCavemanEnabled(!cavemanEnabled)}
+          />
+        </div>
+        {cavemanEnabled && (
+          <div className="mt-3 flex items-center gap-2">
+            {CAVEMAN_LEVELS.map((lvl) => (
+              <button
+                key={lvl.id}
+                onClick={() => handleCavemanLevel(lvl.id)}
+                className={`px-3 py-1.5 rounded text-xs font-medium border transition-colors ${
+                  cavemanLevel === lvl.id
+                    ? "bg-primary text-white border-primary"
+                    : "bg-transparent border-border text-text-muted hover:bg-black/5 dark:hover:bg-white/5"
+                }`}
+                title={lvl.desc}
+              >
+                {lvl.label}
+              </button>
+            ))}
+          </div>
+        )}
      </Card>

      {/* API Keys */}
--- a/src/app/api/v1/api/chat/route.js
+++ b/src/app/api/v1/api/chat/route.js
@@ -8,7 +8,6 @@ async function ensureInitialized() {
  if (!initialized) {
    await initTranslators();
    initialized = true;
-    console.log("[SSE] Translators initialized");
  }
 }

--- a/src/app/api/v1/chat/completions/route.js
+++ b/src/app/api/v1/chat/completions/route.js
@@ -11,7 +11,6 @@ async function ensureInitialized() {
  if (!initialized) {
    await initTranslators();
    initialized = true;
-    console.log("[SSE] Translators initialized");
  }
 }

--- a/src/app/api/v1/messages/route.js
+++ b/src/app/api/v1/messages/route.js
@@ -10,7 +10,6 @@ async function ensureInitialized() {
  if (!initialized) {
    await initTranslators();
    initialized = true;
-    console.log("[SSE] Translators initialized for /v1/messages");
  }
 }

--- a/src/app/api/v1/responses/route.js
+++ b/src/app/api/v1/responses/route.js
@@ -7,7 +7,6 @@ async function ensureInitialized() {
  if (!initialized) {
    await initTranslators();
    initialized = true;
-    console.log("[SSE] Translators initialized for /v1/responses");
  }
 }

--- a/src/app/api/v1beta/models/[...path]/route.js
+++ b/src/app/api/v1beta/models/[...path]/route.js
@@ -10,7 +10,6 @@ async function ensureInitialized() {
  if (!initialized) {
    await initTranslators();
    initialized = true;
-    console.log("[SSE] Translators initialized for /v1beta/models");
  }
 }

--- a/src/lib/localDb.js
+++ b/src/lib/localDb.js
@@ -37,6 +37,8 @@ const DEFAULT_SETTINGS = {
  outboundNoProxy: "",
  mitmRouterBaseUrl: DEFAULT_MITM_ROUTER_BASE,
  rtkEnabled: true,
+  cavemanEnabled: false,
+  cavemanLevel: "full",
 };

 function cloneDefaultData() {
--- a/src/mitm/server.js
+++ b/src/mitm/server.js
@@ -12,9 +12,15 @@ const { getCertForDomain } = require("./cert/generate");
 const DB_FILE = path.join(DATA_DIR, "db.json");
 const LOCAL_PORT = 443;
 const IS_WIN = process.platform === "win32";
-const ENABLE_FILE_LOG = false;
+const ENABLE_FILE_LOG = true;
 const INTERNAL_REQUEST_HEADER = { name: "x-request-source", value: "local" };

+// Host rewrite for upstream forward: PROD cloudcode-pa is rate-limited (429),
+// daily-cloudcode-pa (dev endpoint) accepts same body+token. Same trick as open-sse.
+const HOST_REWRITE = {
+  "cloudcode-pa.googleapis.com": "daily-cloudcode-pa.googleapis.com",
+};
+
 // Load handlers — dev/ overrides handlers/ for private implementations
 function loadHandler(name) {
  try { return require(`./dev/${name}`); } catch {}
@@ -43,7 +49,6 @@ function sniCallback(servername, cb) {
      cert: `${certData.cert}\n${rootCAPem}`
    });
    certCache.set(servername, ctx);
-    log(`🔐 Cert generated: ${servername}`);
    cb(null, ctx);
  } catch (e) {
    err(`SNI error for ${servername}: ${e.message}`);
@@ -123,7 +128,8 @@ function getMappedModel(tool, model) {
 * Also tees full stream into a dump file when ENABLE_FILE_LOG is on.
 */
 async function passthrough(req, res, bodyBuffer, onResponse) {
-  const targetHost = (req.headers.host || TARGET_HOSTS[0]).split(":")[0];
+  const originalHost = (req.headers.host || TARGET_HOSTS[0]).split(":")[0];
+  const targetHost = HOST_REWRITE[originalHost] || originalHost;
  const targetIP = await resolveTargetIP(targetHost);
  const dumper = ENABLE_FILE_LOG ? createResponseDumper(req, "passthrough") : null;

@@ -194,25 +200,18 @@ const server = https.createServer(sslOptions, async (req, res) => {
    const isChat = patterns.some(p => req.url.includes(p));
    if (!isChat) return passthrough(req, res, bodyBuffer);

-    log(`🔍 [${tool}] url=${req.url} | bodyLen=${bodyBuffer.length}`);
-
    // Cursor uses binary proto — model extraction not possible at this layer.
    // Delegate directly to handler which decodes proto internally.
    if (tool === "cursor") {
-      log(`⚡ intercept | cursor | proto`);
      return handlers[tool].intercept(req, res, bodyBuffer, null, passthrough);
    }

    const model = extractModel(req.url, bodyBuffer);
-    log(`🔍 [${tool}] model="${model}"`);
-
    const mappedModel = getMappedModel(tool, model);
    if (!mappedModel) {
-      log(`⏩ passthrough | no mapping | ${tool} | ${model || "unknown"}`);
      return passthrough(req, res, bodyBuffer);
    }

-    log(`⚡ intercept | ${tool} | ${model} → ${mappedModel}`);
    return handlers[tool].intercept(req, res, bodyBuffer, mappedModel, passthrough);
  } catch (e) {
    err(`Unhandled error: ${e.message}`);
--- a/src/sse/handlers/chat.js
+++ b/src/sse/handlers/chat.js
@@ -207,6 +207,8 @@ async function handleSingleModelChat(body, modelStr, clientRawRequest = null, re
      apiKey,
      ccFilterNaming: !!chatSettings.ccFilterNaming,
      rtkEnabled: !!chatSettings.rtkEnabled,
+      cavemanEnabled: !!chatSettings.cavemanEnabled,
+      cavemanLevel: chatSettings.cavemanLevel || "full",
      providerThinking,
      // Detect source format by endpoint + body
      sourceFormatOverride: request?.url ? detectFormatByEndpoint(new URL(request.url).pathname, body) : null,