n9router 0.4.37 → 0.4.40

package/app/open-sse/config/providers.js

@@ -194,7 +194,7 @@ export const PROVIDERS = {
   kiro: {
     baseUrl: "https://codewhisperer.us-east-1.amazonaws.com/generateAssistantResponse",
     format: "kiro",
-    retry: { 429: 2 },
+    retry: { 400: 3, 429: 3, 500: 3, 502: 3, 503: 3, 504: 2 },
     headers: {
       "Content-Type": "application/json",
       "Accept": "application/vnd.amazon.eventstream",

package/app/open-sse/config/runtimeConfig.js

@@ -58,6 +58,9 @@ export const DEFAULT_RETRY_CONFIG = {
 
 // Default 503 retry count for Antigravity executor (per-account before switching)
 export const DEFAULT_AG_503_RETRY_COUNT = 3;
+
+// Default retry count for Kiro executor (per-account before switching)
+export const DEFAULT_KIRO_RETRY_COUNT = 3;
 // Normalize a retry entry to { attempts, delayMs }
 export function resolveRetryEntry(entry) {
   if (entry == null) return { attempts: 0, delayMs: RETRY_CONFIG.delayMs };

package/app/open-sse/executors/kiro.js

@@ -3,7 +3,24 @@ import { PROVIDERS } from "../config/providers.js";
 import { v4 as uuidv4 } from "uuid";
 import { refreshKiroToken } from "../services/tokenRefresh.js";
 import { proxyAwareFetch } from "../utils/proxyFetch.js";
-import { HTTP_STATUS, RETRY_CONFIG, DEFAULT_RETRY_CONFIG, resolveRetryEntry } from "../config/runtimeConfig.js";
+import { HTTP_STATUS, DEFAULT_KIRO_RETRY_COUNT, FETCH_CONNECT_TIMEOUT_MS } from "../config/runtimeConfig.js";
+
+// SSE keepalive comment. Emitted when raw EventStream bytes arrive but don't yet
+// complete a parseable frame (partial-frame buffering during reasoning prefill).
+// Keeps the downstream stall watchdog (pipeWithDisconnect) aware of upstream
+// activity so it doesn't false-abort. Comment lines are dropped by the translate
+// parser and ignored by SSE clients in passthrough.
+const KIRO_KEEPALIVE = new TextEncoder().encode(": ka\n\n");
+
+// Exponential backoff config for Kiro retries (mirrors MITM approach)
+const KIRO_BACKOFF_BASE_MS = 2000;
+const KIRO_BACKOFF_CAP_MS = 15000;
+
+// Transient 400 patterns that should be retried (Kiro-specific false positives)
+const KIRO_RETRYABLE_400_PATTERNS = ["improperly formed request"];
+
+// Status codes eligible for in-executor retry with exponential backoff
+const KIRO_RETRYABLE_STATUSES = new Set([400, 429, 500, 502, 503, 504]);
 
 /**
  * KiroExecutor - Executor for Kiro AI (AWS CodeWhisperer)
@@ -33,33 +50,97 @@ export class KiroExecutor extends BaseExecutor {
   }
 
   /**
-   * Custom execute for Kiro - handles AWS EventStream binary response with retry support
+   * Check if a 400 error is a known transient Kiro issue that should be retried.
+   * Only retries specific patterns — genuine validation errors are not retried.
+   */
+  _isRetryable400(responseBody) {
+    if (!responseBody) return false;
+    const lower = responseBody.toLowerCase();
+    return KIRO_RETRYABLE_400_PATTERNS.some(pattern => lower.includes(pattern));
+  }
+
+  /**
+   * Custom execute for Kiro - handles AWS EventStream binary response with
+   * exponential backoff retry (MITM-style) for transient errors.
    */
-  async execute({ model, body, stream, credentials, signal, log, proxyOptions = null }) {
+  async execute({ model, body, stream, credentials, signal, log, proxyOptions = null, streamWatchdogEnabled = true }) {
     const url = this.buildUrl(model, stream, 0);
     const transformedBody = this.transformRequest(model, body, stream, credentials);
     
-    // Merge default retry config with provider-specific config
-    const retryConfig = { ...DEFAULT_RETRY_CONFIG, ...this.config.retry };
+    // Resolve max retry count: per-account → global setting → default
+    const maxRetries = credentials?.kiroRetryCount ?? DEFAULT_KIRO_RETRY_COUNT;
     let retryAttempts = 0;
 
     while (true) {
       const headers = this.buildHeaders(credentials, stream);
-      
-      const response = await proxyAwareFetch(url, {
-        method: "POST",
-        headers,
-        body: JSON.stringify(transformedBody),
-        signal
-      }, proxyOptions);
-
-      // Check if should retry based on status code
-      const { attempts: maxRetries, delayMs } = resolveRetryEntry(retryConfig[response.status]);
-      if (!response.ok && maxRetries > 0 && retryAttempts < maxRetries) {
-        retryAttempts++;
-        log?.debug?.("RETRY", `${response.status} retry ${retryAttempts}/${maxRetries} after ${delayMs / 1000}s`);
-        await new Promise(resolve => setTimeout(resolve, delayMs));
-        continue;
+
+      // Abort if upstream doesn't return response headers within FETCH_CONNECT_TIMEOUT_MS.
+      const connectCtrl = new AbortController();
+      const connectTimer = setTimeout(() => connectCtrl.abort(new Error("fetch connect timeout")), FETCH_CONNECT_TIMEOUT_MS);
+      const mergedSignal = signal ? AbortSignal.any([signal, connectCtrl.signal]) : connectCtrl.signal;
+
+      let response;
+      try {
+        response = await proxyAwareFetch(url, {
+          method: "POST",
+          headers,
+          body: JSON.stringify(transformedBody),
+          signal: mergedSignal
+        }, proxyOptions);
+        // Headers received — body streaming is now governed by the stall watchdog.
+        clearTimeout(connectTimer);
+      } catch (error) {
+        clearTimeout(connectTimer);
+        // Connect timeout is internal: convert to a retryable network error
+        const isConnectTimeout = connectCtrl.signal.aborted && error.name === "AbortError";
+        if (isConnectTimeout) {
+          if (maxRetries > 0 && retryAttempts < maxRetries) {
+            retryAttempts++;
+            const backoffMs = this._computeBackoff(retryAttempts);
+            log?.debug?.("RETRY", `connect timeout retry ${retryAttempts}/${maxRetries} after ${(backoffMs / 1000).toFixed(1)}s`);
+            await new Promise(resolve => setTimeout(resolve, backoffMs));
+            continue;
+          }
+          throw new Error(`Kiro upstream connect timeout after ${FETCH_CONNECT_TIMEOUT_MS}ms`);
+        }
+        throw error;
+      }
+
+      // Check if this status code is retryable
+      if (!response.ok && KIRO_RETRYABLE_STATUSES.has(response.status)) {
+        // For 400, only retry known transient patterns
+        let isTransient400 = false;
+        if (response.status === 400) {
+          let bodyText = "";
+          try { bodyText = await response.clone().text(); } catch { /* ignore */ }
+          
+          if (!this._isRetryable400(bodyText)) {
+            // Genuine validation error — don't retry
+            return { response, url, headers, transformedBody };
+          }
+          isTransient400 = true;
+        }
+
+        if (maxRetries > 0 && retryAttempts < maxRetries) {
+          retryAttempts++;
+          const backoffMs = this._computeBackoff(retryAttempts);
+          log?.debug?.("RETRY", `${response.status} retry ${retryAttempts}/${maxRetries} after ${(backoffMs / 1000).toFixed(1)}s`);
+          await new Promise(resolve => setTimeout(resolve, backoffMs));
+          continue;
+        }
+
+        // Retries exhausted — remap transient 400 to 429 so Claude Code CLI
+        // recognizes it as retryable (CLI only retries on 429/503, not 400)
+        if (isTransient400) {
+          const bodyText = await response.clone().text().catch(() => "");
+          const remappedResponse = new Response(bodyText, {
+            status: HTTP_STATUS.RATE_LIMITED,
+            statusText: "Too Many Requests",
+            headers: response.headers
+          });
+          log?.debug?.("RETRY", `400 → 429 remap after ${retryAttempts} retries (transient Kiro error)`);
+          return { response: remappedResponse, url, headers, transformedBody };
+        }
       }
 
       if (!response.ok) {
@@ -67,18 +148,26 @@ export class KiroExecutor extends BaseExecutor {
       }
 
       // Success - transform and return
-      // For Kiro, we need to transform the binary EventStream to SSE
-      // Create a TransformStream to convert binary to SSE text
-      const transformedResponse = this.transformEventStreamToSSE(response, model);
+      const transformedResponse = this.transformEventStreamToSSE(response, model, streamWatchdogEnabled);
       return { response: transformedResponse, url, headers, transformedBody };
     }
   }
 
+  /**
+   * Compute exponential backoff with full jitter (MITM-style).
+   * Formula: random in [0, min(cap, base * 2^(attempt-1))]
+   * Prevents thundering herd when multiple requests hit errors simultaneously.
+   */
+  _computeBackoff(attempt) {
+    const expCeiling = Math.min(KIRO_BACKOFF_CAP_MS, KIRO_BACKOFF_BASE_MS * Math.pow(2, attempt - 1));
+    return Math.floor(Math.random() * expCeiling) + KIRO_BACKOFF_BASE_MS / 2;
+  }
+
   /**
    * Transform AWS EventStream binary response to SSE text stream
    * Using TransformStream instead of ReadableStream.pull() to avoid Workers timeout
    */
-  transformEventStreamToSSE(response, model) {
+  transformEventStreamToSSE(response, model, streamWatchdogEnabled = true) {
     let buffer = new Uint8Array(0);
     let chunkIndex = 0;
     const responseId = `chatcmpl-${Date.now()}`;
@@ -94,7 +183,18 @@ export class KiroExecutor extends BaseExecutor {
     };
 
     const transformStream = new TransformStream({
-      async transform(chunk, controller) {
+      async transform(chunk, rawController) {
+        // Shadow the controller so we can tell whether this call emitted anything.
+        // Raw AWS EventStream frames can buffer for a long time (reasoning prefill)
+        // producing no SSE output; the downstream stall watchdog measures THIS
+        // stream's output, so a silent transform call looks like a stall and gets
+        // aborted. When a chunk arrived but yielded no frame, emit a keepalive.
+        let emitted = false;
+        const controller = {
+          enqueue: (d) => { emitted = true; rawController.enqueue(d); },
+          error: (e) => rawController.error(e),
+          terminate: () => rawController.terminate()
+        };
         // Append to buffer
         const newBuffer = new Uint8Array(buffer.length + chunk.length);
         newBuffer.set(buffer);
@@ -373,6 +473,14 @@ export class KiroExecutor extends BaseExecutor {
         if (iterations >= maxIterations) {
           console.warn("[Kiro] Max iterations reached in event parsing");
         }
+
+        // Bytes arrived but yielded no SSE frame yet (partial-frame buffering
+        // during reasoning prefill). Emit a keepalive so the downstream stall
+        // watchdog registers upstream activity and doesn't false-abort.
+        // Skipped when the watchdog is disabled (legacy v0.4.35 behavior).
+        if (!emitted && streamWatchdogEnabled) {
+          rawController.enqueue(KIRO_KEEPALIVE);
+        }
       },
 
       flush(controller) {

package/app/open-sse/handlers/chatCore/streamingHandler.js

@@ -39,11 +39,11 @@ function buildTransformStream({ provider, sourceFormat, targetFormat, userAgent,
 /**
  * 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, streamDetailId }) {
+export function handleStreamingResponse({ providerResponse, provider, model, sourceFormat, targetFormat, userAgent, body, stream, translatedBody, finalBody, requestStartTime, connectionId, apiKey, clientRawRequest, onRequestSuccess, reqLogger, toolNameMap, streamController, onStreamComplete, streamDetailId, streamWatchdogEnabled = true }) {
   if (onRequestSuccess) onRequestSuccess();
 
   const transformStream = buildTransformStream({ provider, sourceFormat, targetFormat, userAgent, reqLogger, toolNameMap, model, connectionId, body, onStreamComplete, apiKey });
-  const transformedBody = pipeWithDisconnect(providerResponse, transformStream, streamController);
+  const transformedBody = pipeWithDisconnect(providerResponse, transformStream, streamController, streamWatchdogEnabled);
 
   saveRequestDetail(buildRequestDetail({
     provider, model, connectionId, apiKey,

package/app/open-sse/handlers/chatCore.js

@@ -27,7 +27,7 @@ import { compressMessages, formatRtkLog } from "../rtk/index.js";
  * @param {object} options.credentials - Provider credentials
  * @param {string} options.sourceFormatOverride - Override detected source format (e.g. "openai-responses")
  */
-export async function handleChatCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, clientRawRequest, connectionId, userAgent, apiKey, ccFilterNaming, rtkEnabled, cavemanEnabled, cavemanLevel, sourceFormatOverride, providerThinking }) {
+export async function handleChatCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, clientRawRequest, connectionId, userAgent, apiKey, ccFilterNaming, rtkEnabled, cavemanEnabled, cavemanLevel, sourceFormatOverride, providerThinking, streamWatchdogEnabled = true }) {
   const { provider, model } = modelInfo;
   const requestStartTime = Date.now();
 
@@ -177,7 +177,7 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
   // Execute request
   let providerResponse, providerUrl, providerHeaders, finalBody;
   try {
-    const result = await executor.execute({ model, body: translatedBody, stream, credentials, signal: streamController.signal, log, proxyOptions });
+    const result = await executor.execute({ model, body: translatedBody, stream, credentials, signal: streamController.signal, log, proxyOptions, streamWatchdogEnabled });
     providerResponse = result.response;
     providerUrl = result.url;
     providerHeaders = result.headers;
@@ -216,7 +216,7 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
           try { await onCredentialsRefreshed(newCredentials); } catch (e) { log?.warn?.("TOKEN", `onCredentialsRefreshed failed: ${e.message}`); }
         }
         try {
-          const retryResult = await executor.execute({ model, body: translatedBody, stream, credentials, signal: streamController.signal, log, proxyOptions });
+          const retryResult = await executor.execute({ model, body: translatedBody, stream, credentials, signal: streamController.signal, log, proxyOptions, streamWatchdogEnabled });
           if (retryResult.response.ok) { providerResponse = retryResult.response; providerUrl = retryResult.url; }
         } catch { log?.warn?.("TOKEN", `${provider.toUpperCase()} | retry after refresh failed`); }
       } else {
@@ -267,7 +267,7 @@ export async function handleChatCore({ body, modelInfo, credentials, log, onCred
 
   // Streaming response
   const { onStreamComplete, streamDetailId } = buildOnStreamComplete({ ...sharedCtx });
-  return handleStreamingResponse({ ...sharedCtx, providerResponse, sourceFormat, targetFormat, userAgent, reqLogger, toolNameMap, streamController, onStreamComplete, streamDetailId });
+  return handleStreamingResponse({ ...sharedCtx, providerResponse, sourceFormat, targetFormat, userAgent, reqLogger, toolNameMap, streamController, onStreamComplete, streamDetailId, streamWatchdogEnabled });
 }
 
 export function isTokenExpiringSoon(expiresAt, bufferMs = 5 * 60 * 1000) {

package/app/open-sse/handlers/responsesHandler.js

@@ -21,7 +21,7 @@ import { convertResponsesStreamToJson } from "../transformer/streamToJsonConvert
  * @param {string} options.connectionId - Connection ID for usage tracking
  * @returns {Promise<{success: boolean, response?: Response, status?: number, error?: string}>}
  */
-export async function handleResponsesCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, connectionId }) {
+export async function handleResponsesCore({ body, modelInfo, credentials, log, onCredentialsRefreshed, onRequestSuccess, onDisconnect, connectionId, streamWatchdogEnabled = true }) {
   // Convert Responses API format to Chat Completions format
   const convertedBody = convertResponsesApiFormat(body);
 
@@ -42,6 +42,7 @@ export async function handleResponsesCore({ body, modelInfo, credentials, log, o
     onRequestSuccess,
     onDisconnect,
     connectionId,
+    streamWatchdogEnabled,
     sourceFormatOverride: "openai-responses"
   });
 

package/app/open-sse/utils/streamHandler.js

@@ -2,6 +2,12 @@
 import { STREAM_STALL_TIMEOUT_MS } from "../config/runtimeConfig.js";
 import { dbg, isDebugEnabled } from "./debugLog.js";
 
+// Terminal SSE sentinel. When a stream is aborted/errored mid-flight the
+// transform stream's flush() never runs, so the normal "data: [DONE]" is never
+// emitted and clients hang until their own timeout. We send it manually on the
+// graceful-close path so clients always see a clean end-of-stream.
+const DONE_SENTINEL = new TextEncoder().encode("data: [DONE]\n\n");
+
 // Get HH:MM:SS timestamp
 function getTimeString() {
   return new Date().toLocaleTimeString("en-US", { hour12: false, hour: "2-digit", minute: "2-digit", second: "2-digit" });
@@ -94,14 +100,36 @@ export function createStreamController({ onDisconnect, onError, log, provider, m
  * for long periods while raw bytes still flow (e.g. Kiro EventStream
  * binary frames buffering, Claude reasoning streams).
  */
-export function createDisconnectAwareStream(transformStream, streamController) {
+export function createDisconnectAwareStream(transformStream, streamController, streamWatchdogEnabled = true) {
   const reader = transformStream.readable.getReader();
   const writer = transformStream.writable.getWriter();
+  // Client-facing disconnect detection: when the upstream aborts/errors mid-flight,
+  // the transform flush() is skipped and "data: [DONE]" is never produced. Track
+  // whether we've terminated the client stream so we can inject the sentinel once.
+  let clientClosed = false;
+
+  // Emit "data: [DONE]" then close the client stream. Used on the abort/error path
+  // where the normal in-band sentinel from flush() will never arrive. When the watchdog
+  // is disabled (legacy v0.4.35 behavior) we close WITHOUT injecting the sentinel.
+  const closeWithSentinel = (controller) => {
+    if (clientClosed) return;
+    clientClosed = true;
+    if (streamWatchdogEnabled) {
+      try {
+        controller.enqueue(DONE_SENTINEL);
+      } catch (e) { /* downstream already gone */ }
+    }
+    try {
+      controller.close();
+    } catch (e) { /* already closed or cancelled */ }
+  };
 
   return new ReadableStream({
     async pull(controller) {
       if (!streamController.isConnected()) {
-        controller.close();
+        // Stream was aborted/stalled before this pull — terminate cleanly so the
+        // client sees an end-of-stream instead of hanging on a truncated response.
+        closeWithSentinel(controller);
         return;
       }
 
@@ -109,7 +137,9 @@ export function createDisconnectAwareStream(transformStream, streamController) {
         const { done, value } = await reader.read();
 
         if (done) {
+          // Graceful EOF: the transform flush() already emitted "data: [DONE]".
           streamController.handleComplete();
+          clientClosed = true;
           controller.close();
           return;
         }
@@ -136,11 +166,9 @@ export function createDisconnectAwareStream(transformStream, streamController) {
           code === "UND_ERR_SOCKET";
 
         if (!wasConnected || isNetworkClose) {
-          try {
-            controller.close();
-          } catch (e) {
-            // Stream might already be closed or cancelled
-          }
+          // Upstream stall/abort/reset: flush() was skipped, so emit the sentinel
+          // ourselves before closing so the client terminates cleanly.
+          closeWithSentinel(controller);
         } else {
           try {
             controller.error(error);
@@ -150,6 +178,8 @@ export function createDisconnectAwareStream(transformStream, streamController) {
     },
 
     cancel(reason) {
+      // Client went away — nothing to flush to it.
+      clientClosed = true;
       streamController.handleDisconnect(reason || "cancelled");
       reader.cancel();
       writer.abort();
@@ -173,7 +203,7 @@ export function createDisconnectAwareStream(transformStream, streamController) {
  * @param {TransformStream} transformStream - Transform stream for SSE
  * @param {object} streamController - Stream controller from createStreamController
  */
-export function pipeWithDisconnect(providerResponse, transformStream, streamController) {
+export function pipeWithDisconnect(providerResponse, transformStream, streamController, streamWatchdogEnabled = true) {
   let stallTimer = null;
   let chunkCount = 0;
   let totalBytes = 0;
@@ -184,6 +214,7 @@ export function pipeWithDisconnect(providerResponse, transformStream, streamCont
     if (stallTimer) { clearTimeout(stallTimer); stallTimer = null; }
   };
   const armStall = () => {
+    if (!streamWatchdogEnabled) return;   // OFF → never arm watchdog (v0.4.35 behavior)
     clearStall();
     stallTimer = setTimeout(() => {
       stallTimer = null;
@@ -207,7 +238,7 @@ export function pipeWithDisconnect(providerResponse, transformStream, streamCont
   };
 
   armStall();
-  dbg(tag, `pipe start | stallTimeout=${STREAM_STALL_TIMEOUT_MS}ms`);
+  dbg(tag, `pipe start | watchdog=${streamWatchdogEnabled ? `on stallTimeout=${STREAM_STALL_TIMEOUT_MS}ms` : "off (legacy)"}`);
 
   const upstreamTap = new TransformStream({
     transform(chunk, controller) {
@@ -232,7 +263,8 @@ export function pipeWithDisconnect(providerResponse, transformStream, streamCont
 
   return createDisconnectAwareStream(
     { readable: transformedBody, writable: { getWriter: () => ({ abort: () => Promise.resolve() }) } },
-    wrappedController
+    wrappedController,
+    streamWatchdogEnabled
   );
 }
 

package/app/package.json

@@ -1,6 +1,6 @@
 {
   "name": "n9router",
-  "version": "0.4.37",
+  "version": "0.4.40",
   "description": "Self-hosted AI routing gateway — local proxy for Claude, Gemini, OpenAI and 40+ providers",
   "keywords": [
     "ai",

package/app/src/mitm/server.js

@@ -24,6 +24,7 @@ const { createAntigravityDebugContext, extractBearerToken, maskToken } = require
 const { getCertForDomain } = require("./cert/generate");
 const { buildInputOnlyRequestDetail, createTokenSwapUsageObserver, generateDetailId } = require("./usageTracker");
 const { pushHealthEvent, getLastEventStatus, migrateToEmailKeys } = require("./healthStore");
+const { getTokenSwapRetryType, isRetryableTokenSwapStatus } = require("./tokenSwapRetry");
 
 const { applyRtkCompression } = require("./rtkCompressor");
 const { applyAntigravityIdeVersionOverride } = require("./antigravityIdeVersion");
@@ -203,7 +204,7 @@ async function passthrough(req, res, bodyBuffer, onResponse, debugContext = null
 
 // ── Token swap forward ────────────────────────────────────────
 // Unlike passthrough(), this checks upstream statusCode BEFORE
-// piping to client — enabling auto-retry on 403/429/503.
+// piping to client — enabling auto-retry on retryable auth/quota/server errors.
 
 async function tokenSwapForward(req, res, bodyBuffer, connections, model, strategy, provider, requestStartTime, debugContext = null) {
   let targetHost = (req.headers.host || TARGET_HOSTS[0]).split(":")[0];
@@ -286,16 +287,12 @@ async function tokenSwapForward(req, res, bodyBuffer, connections, model, strate
             servername: targetHost,
             rejectUnauthorized: false
           }, (forwardRes) => {
-            if (forwardRes.statusCode === 403 || forwardRes.statusCode === 429 || forwardRes.statusCode === 503 || forwardRes.statusCode === 401) {
+            if (isRetryableTokenSwapStatus(forwardRes.statusCode)) {
               const chunks = [];
               forwardRes.on("data", c => chunks.push(c));
               forwardRes.on("end", () => {
                 const body = Buffer.concat(chunks).toString();
-                const retryType = forwardRes.statusCode === 401
-                  ? "auth"
-                  : forwardRes.statusCode === 403
-                    ? "permission"
-                    : "quota";
+                const retryType = getTokenSwapRetryType(forwardRes.statusCode);
                 resolve({ retry: true, retryType, body, headers: forwardRes.headers, statusCode: forwardRes.statusCode });
               });
             } else {
@@ -307,9 +304,9 @@ async function tokenSwapForward(req, res, bodyBuffer, connections, model, strate
           forwardReq.end();
         });
 
-        if (result.retry && (result.retryType === "quota" || result.retryType === "permission")) {
-          // ── Unified retry: 403 IAM denial, 429 and 503 treated identically ──
-          // Both are false-positive-prone from Antigravity; retry same account
+        if (result.retry && (result.retryType === "quota" || result.retryType === "permission" || result.retryType === "server_error")) {
+          // ── Unified retry: 403 IAM denial, 429, 500 and 503 treated identically ──
+          // These can be false-positive-prone from Antigravity; retry same account
           // with exponential backoff before moving on.
           const maxRetries = getAntigravity503RetryCount(conn.antigravity503RetryCount);
           if (!conn._quotaRetryCount) conn._quotaRetryCount = 0;

package/app/src/mitm/tokenPool.js

@@ -430,20 +430,11 @@ function getConnectionLabel(connection) {
   return connection.name || connection.id.slice(0, 8);
 }
 
-function isTokenSwapEnabled(provider) {
+function isTokenSwapEnabled() {
   try {
     if (!fs.existsSync(DB_FILE)) return false;
     const db = JSON.parse(fs.readFileSync(DB_FILE, "utf-8"));
-    // Explicit toggle — must be enabled in settings
-    if (!db.settings?.tokenSwapEnabled) {
-      return false;
-    }
-    const active = getActiveConnections(provider);
-    if (active.length === 0) {
-      log(`⚙️ [token-pool] swap enabled in settings but no active ${provider} connections`);
-      return false;
-    }
-    return true;
+    return !!db.settings?.tokenSwapEnabled;
   } catch { return false; }
 }
 

package/app/src/mitm/tokenSwapRetry.js

@@ -0,0 +1,19 @@
+const RETRYABLE_TOKEN_SWAP_STATUSES = new Set([401, 403, 429, 500, 503]);
+
+function getTokenSwapRetryType(statusCode) {
+  const status = Number(statusCode);
+  if (status === 401) return "auth";
+  if (status === 403) return "permission";
+  if (status === 429 || status === 503) return "quota";
+  if (status === 500) return "server_error";
+  return null;
+}
+
+function isRetryableTokenSwapStatus(statusCode) {
+  return RETRYABLE_TOKEN_SWAP_STATUSES.has(Number(statusCode));
+}
+
+module.exports = {
+  getTokenSwapRetryType,
+  isRetryableTokenSwapStatus,
+};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "n9router",
-  "version": "0.4.37",
+  "version": "0.4.40",
   "description": "Self-hosted AI routing gateway — local proxy for Claude, Gemini, OpenAI and 40+ providers",
   "keywords": [
     "ai",

package/app/.next/static/YEhAtVGU-1J8HgdJmmpjP/_buildManifest.js

@@ -1 +0,0 @@
-self.__BUILD_MANIFEST=function(e,s,t){return{__rewrites:{afterFiles:[{has:e,source:"/v1/v1/:path*",destination:s},{has:e,source:"/v1/v1",destination:t},{has:e,source:"/codex/:path*",destination:"/api/v1/responses"},{has:e,source:"/v1/:path*",destination:s},{has:e,source:"/v1",destination:t}],beforeFiles:[],fallback:[]},__routerFilterStatic:{numItems:143,errorRate:1e-4,numBits:2742,numHashes:14,bitArray:[0,1,1,0,1,0,1,1,0,0,0,1,1,0,1,0,0,0,1,1,1,1,1,0,1,0,1,1,0,0,0,1,1,1,1,1,1,0,1,0,0,1,0,1,0,1,1,0,0,1,1,0,1,1,0,1,0,0,0,0,0,0,0,1,1,0,0,0,0,0,1,1,0,1,0,1,0,1,1,1,1,1,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,0,1,1,0,0,1,1,1,1,0,0,0,0,1,0,0,1,1,0,1,1,0,1,0,0,1,1,1,1,1,1,1,1,0,1,0,1,0,0,1,1,0,1,1,1,1,0,0,0,0,1,1,1,0,1,0,1,0,1,1,0,0,0,0,1,0,1,0,0,1,0,0,0,1,1,0,1,0,0,0,0,0,1,1,1,1,0,0,1,1,0,0,1,1,1,1,1,0,1,0,0,1,0,0,1,1,0,1,0,0,0,1,1,0,1,0,1,1,0,1,1,1,1,0,0,0,1,1,1,0,1,0,1,1,1,1,0,1,0,1,1,1,0,1,0,0,1,1,0,0,1,0,0,1,1,1,1,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,1,0,0,1,0,0,1,1,0,0,0,1,0,1,1,0,1,0,0,0,1,0,0,0,1,1,1,0,1,1,1,0,0,1,1,0,0,0,0,1,1,0,0,1,1,1,1,0,1,1,0,1,1,0,0,0,1,1,1,0,0,1,0,1,1,1,0,1,1,0,0,1,1,0,0,0,1,1,0,0,0,1,0,0,1,1,1,0,1,1,1,0,1,0,0,1,1,0,0,1,1,0,0,1,0,1,0,0,1,1,1,0,1,0,0,0,0,1,0,1,1,1,1,1,1,0,1,0,1,1,1,0,1,1,0,0,1,0,1,0,0,1,1,0,0,0,0,0,0,1,1,1,1,1,1,0,0,0,1,1,1,0,1,1,0,0,0,0,1,0,1,0,0,1,0,0,0,1,1,0,1,1,0,1,0,0,1,1,0,0,0,1,0,0,1,0,1,0,0,1,0,1,0,1,1,1,0,1,0,0,1,1,0,0,0,0,1,0,0,0,0,0,0,0,1,0,0,1,0,1,1,0,1,1,1,1,0,0,1,1,1,1,0,1,1,1,0,0,0,1,1,0,1,1,0,0,1,0,0,0,0,0,1,0,0,1,1,1,0,1,1,0,0,0,1,1,0,0,1,1,0,1,1,0,1,0,1,1,0,0,1,1,0,0,0,0,0,1,0,0,1,0,1,1,1,1,1,1,1,0,0,1,0,0,1,0,1,0,1,1,0,0,0,1,1,0,0,1,1,1,1,0,1,1,0,1,1,0,0,1,0,1,1,0,1,1,1,0,1,1,0,0,1,1,1,1,1,0,1,0,0,0,0,1,0,0,0,1,1,1,0,0,0,1,0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,1,1,1,0,1,1,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,0,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1,1,1,1,1,0,0,1,0,1,1,0,0,0,1,0,1,1,1,1,1,1,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,0,0,0,1,0,0,0,1,1,1,0,1,1,1,1,1,1,0,0,1,1,1,1,1,0,1,1,0,1,0,1,0,0,1,0,0,1,1,1,0,0,1,0,0,1,1,1,0,0,1,0,0,1,1,0,1,0,1,1,1,1,1,1,1,1,0,1,1,0,1,0,1,1,1,0,0,1,1,0,0,0,0,1,1,1,1,1,1,1,0,1,0,0,1,0,1,1,1,0,0,1,1,0,0,1,1,1,0,0,0,1,1,0,0,1,0,0,0,0,0,1,1,1,1,0,1,0,0,0,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,1,1,1,0,1,0,0,1,0,1,1,1,1,1,0,0,0,0,0,1,0,0,1,1,0,0,0,0,1,0,0,1,1,1,1,1,0,0,1,1,0,0,1,0,1,0,0,0,0,1,1,1,0,1,1,0,1,0,0,1,0,0,0,0,1,1,1,0,1,1,0,1,1,0,0,0,1,0,0,0,0,1,1,0,0,1,1,1,0,0,0,0,1,1,0,0,1,0,0,1,0,1,0,1,0,0,1,1,0,1,0,1,0,0,1,1,1,1,1,0,1,0,0,1,0,1,0,0,0,1,0,0,0,0,0,0,0,1,0,0,1,1,0,1,1,1,1,1,1,1,1,0,1,0,0,0,1,0,0,1,0,0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,1,1,1,1,1,1,1,1,0,1,0,0,0,1,0,1,0,0,0,0,1,1,0,1,1,1,1,0,1,1,1,1,0,1,0,0,0,0,1,0,0,0,1,1,0,0,0,1,1,0,1,1,1,1,0,1,0,1,0,0,0,1,1,1,1,1,1,0,1,0,1,1,1,0,1,0,0,0,1,1,0,1,0,0,1,1,0,0,0,1,1,0,0,1,0,1,0,1,1,1,0,0,1,0,1,1,1,0,1,1,1,0,1,1,1,1,1,1,0,0,1,0,0,1,0,1,0,0,0,1,1,1,1,1,1,1,1,0,1,0,0,1,0,1,0,1,1,0,0,0,1,0,1,0,0,1,1,1,0,1,0,0,0,1,0,1,1,1,1,1,1,0,0,1,0,0,0,0,1,0,0,1,1,1,0,1,1,0,0,1,1,0,1,0,1,0,1,1,0,0,1,0,1,1,0,0,0,0,0,0,0,0,1,1,1,1,0,0,1,1,1,0,1,1,0,1,0,1,1,1,0,0,1,1,0,1,1,1,1,1,0,1,1,1,0,1,1,1,1,1,0,0,1,1,0,0,0,1,0,0,1,0,1,1,1,0,1,1,1,0,0,1,0,1,1,1,1,1,0,0,0,0,1,0,0,1,0,1,0,0,1,1,1,1,1,1,1,1,1,1,0,0,1,1,0,0,1,0,0,0,1,0,1,0,1,1,1,1,1,0,1,1,1,1,1,0,1,1,1,1,1,0,1,1,1,1,0,0,1,1,0,0,1,0,0,1,1,1,1,1,0,0,1,1,0,1,0,0,0,0,0,1,1,1,0,1,1,1,0,1,1,1,0,0,0,0,0,1,0,1,1,1,1,0,0,1,1,0,1,1,1,1,1,0,1,0,0,1,1,0,1,1,0,1,1,1,1,0,0,1,0,0,0,0,0,1,0,1,0,1,0,0,0,0,1,1,0,0,1,0,1,1,1,1,1,0,0,0,1,0,1,1,0,1,1,1,1,1,1,0,1,0,0,0,1,0,1,1,0,0,0,0,1,0,0,1,0,0,0,1,1,1,0,0,1,1,1,0,0,0,0,0,1,0,0,0,1,0,1,0,1,1,0,0,1,0,1,0,1,1,0,1,1,1,1,1,1,1,1,1,0,1,0,0,1,1,1,0,0,1,1,1,1,0,1,0,1,0,0,1,0,1,1,0,1,1,1,0,0,0,1,1,1,1,1,1,1,1,0,1,0,1,1,1,1,1,1,0,0,0,1,0,1,1,0,1,1,0,0,1,0,1,0,1,1,1,0,0,0,1,1,0,1,1,0,1,0,0,1,1,1,1,0,1,0,0,0,1,0,0,0,0,1,1,0,1,0,0,1,1,0,1,1,1,1,1,0,0,0,1,0,0,0,1,1,1,0,1,1,0,1,1,0,1,1,0,1,1,1,1,1,1,1,0,0,0,0,0,1,0,0,0,0,0,0,1,1,1,1,1,1,1,0,0,0,0,0,0,0,1,1,1,1,0,1,0,1,1,1,1,0,0,0,1,0,1,0,0,0,0,1,0,1,1,1,1,1,0,1,0,1,0,1,0,1,1,1,0,1,1,1,1,1,1,0,1,0,1,1,1,1,0,0,0,1,0,0,1,0,1,1,0,0,1,1,1,1,0,0,0,0,1,0,0,1,1,1,1,0,0,0,0,1,0,1,0,0,1,1,1,1,0,1,1,0,0,1,0,1,0,0,0,1,0,1,1,1,0,0,0,0,0,1,1,1,0,1,1,0,0,1,1,0,0,0,0,1,1,0,1,1,1,0,0,1,1,1,1,1,0,0,1,1,1,0,1,0,0,0,1,0,0,1,0,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,1,1,1,0,0,0,0,0,1,0,1,1,1,0,1,1,0,1,0,1,0,0,1,1,0,1,1,1,0,1,0,1,1,0,0,0,0,1,1,1,0,1,1,1,0,0,1,0,0,1,0,0,0,0,1,1,1,1,0,0,0,1,1,1,0,0,1,0,1,0,1,1,0,1,0,1,1,0,0,1,1,1,1,0,1,1,0,0,0,1,1,0,0,1,1,1,0,0,1,0,0,0,1,0,1,1,0,0,1,1,0,1,1,1,1,0,0,1,0,1,0,1,0,0,1,1,1,0,1,1,0,1,1,1,0,1,1,0,0,0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,1,1,1,1,1,1,0,0,0,1,1,1,0,0,0,0,0,1,0,0,1,1,0,1,0,0,0,0,0,1,1,1,1,0,0,1,0,1,1,0,1,0,1,1,1,1,0,0,0,1,0,1,1,0,0,0,1,1,0,0,0,0,1,1,0,0,0,1,1,0,1,0,1,1,1,1,0,1,0,1,1,1,1,1,1,1,1,0,0,1,1,0,0,0,1,0,0,1,0,1,1,0,1,1,1,0,1,0,1,1,0,0,1,0,1,1,0,0,1,0,1,1,0,0,1,0,0,0,1,0,0,0,0,0,1,1,0,0,1,0,1,1,0,1,1,1,1,1,0,1,1,0,0,0,1,1,1,1,1,1,0,0,1,0,1,0,0,1,1,1,0,1,1,0,0,1,0,1,1,1,1,1,0,1,1,1,1,1,0,1,1,0,0,0,0,0,1,1,0,0,0,1,0,0,1,1,1,1,0,0,1,0,0,1,0,1,0,0,1,1,1,0,0,1,1,0,0,1,0,0,1,1,0,1,0,0,1,1,1,0,0,1,1,1,1,0,1,0,0,0,0,0,0,0,1,1,1,0,1,1,1,0,1,1,1,1,1,0,0,0,0,1,0,1,1,0,1,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,1,1,0,1,0,0,0,1,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,0,0,1,1,0,1,1,1,1,0,1,1,1,0,1,1,0,1,1,1,1,1,0,0,1,0,0,0,1,1,0,1,0,0,1,0,1,1,0,1,1,0,1,0,1,1,1,1,0,0,0,0,0,0,1,0,0,0,0,0,1,0,1,0,1,1,1,0,0,1,0,0,0,0,1,0,0,0,0,0,1,1,1,1,0,0,0,1,1,0,1,1,0,0,0,1,0,1,1,0,1,0,0,0,1,1,0,1,0,0,1,1,0,1,1,1,1,1,1,1,0,0,0,1,0,1,0,1,0,0,0,1,0,1,0,0,1,1,1,1,0,1,1,0,0,1,0,0,1,1,0,0,0,1,1,0,1,1,0,1,1,1,0,1,1,1,1,0,1,1,1,0,0,1,1,1,1,0,0,1,1,0,1,0,1,1,0,0,0,1,1,1,1,1,1,1,0,1,1,1,0,1,0,1,1,1,1,1,0,1,0,1,0,1,1,0,1,0,1,0,0,0,0,1,1,1,1,0,0,1,1,1,0,0,0,1,0,0,0,0,1,0,0,0,1,1,0,1,1,1,1,0,0,0,0,0,1,0,1,1,0,1,1,1,0,1,1,0,1,1,0,1,1,0,0,0,0,0,0,1,0,1,0,1,0,1,1,0,1,0,1,1,0,0,1,0,1,1,0,0,1,0,1,1,0,1,0,0,1,0,1,0,1,1,0,1,0,1,0,1,1,0,0,1,1,1,0,0,0,1,1,0,1,1,0,0,1,0,1,0,1,1,1,1,0,1,0,0,1,1,0,0,0,1,0,1,0,1,0,0,0,1,0,0,0,1]},__routerFilterDynamic:{numItems:14,errorRate:1e-4,numBits:269,numHashes:14,bitArray:[0,0,0,1,1,1,0,0,1,1,0,1,1,1,0,0,1,0,1,1,0,0,0,0,0,1,0,0,1,1,1,1,0,0,0,1,0,1,1,0,0,0,0,0,0,1,0,1,1,0,1,1,1,0,1,0,1,0,1,0,0,0,1,0,0,0,0,1,0,0,1,0,1,1,1,1,0,0,1,1,0,1,0,0,1,1,0,0,0,1,1,0,0,0,0,0,1,1,1,0,1,0,0,1,1,0,1,0,1,0,1,0,1,0,1,0,1,1,1,0,1,1,1,0,0,1,1,1,0,1,0,0,0,1,0,1,0,0,1,1,0,0,1,1,0,1,0,0,1,0,0,1,0,0,1,1,0,0,0,1,1,1,0,1,1,1,1,1,0,1,1,0,1,1,1,0,0,0,0,1,1,0,1,0,0,0,1,0,0,0,0,1,0,1,1,1,0,0,1,1,1,0,0,0,1,0,1,0,0,0,0,0,1,0,1,1,1,0,0,0,1,1,0,1,1,1,0,1,1,1,1,1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,0,1,0,0,1,1,1,1,0,1,0,1,1,0,1,0,1,0,1]},sortedPages:["/_app"]}}(void 0,"/api/v1/:path*","/api/v1"),self.__BUILD_MANIFEST_CB&&self.__BUILD_MANIFEST_CB();