fix(types): cast extracted usage to Record<string,number> in stream.ts to resolve TS property errors

Also fix syntax error in openai-to-claude-strip-empty.test.mjs (tool/assistant messages were incorrectly nested)

.gitignore

@@ -89,6 +89,7 @@ docs/*
 !docs/MCP-SERVER.md
 !docs/CLI-TOOLS.md
 
+
 # open-sse tests
 open-sse/test/*
 

CHANGELOG.md

@@ -6,9 +6,47 @@
 
 ---
 
-## [3.0.0-rc.5] — 2026-03-22 _(What's New vs v2.9.5 — will be released as v3.0.0)_
+## [3.0.0-rc.10] — 2026-03-23
 
-> **Upgrade from v2.9.5:** 16 issues resolved · 2 community PRs merged · 2 new providers · 7 new API endpoints · 3 new features · DB migration 008 · 832 tests passing.
+### 🔧 Bug Fixes
+
+- **#509 / #508** — Electron build regression: downgraded Next.js from `16.1.x` to `16.0.10` to eliminate Turbopack module-hashing instability that caused blank screens in the Electron desktop bundle.
+- **Unit test fixes** — Corrected two stale test assertions (`nanobanana-image-handler` aspect ratio/resolution, `thinking-budget` Gemini `thinkingConfig` field mapping) that had drifted after recent implementation changes.
+- **#541** — Responded to user feedback about installation complexity; no code changes required.
+
+---
+
+## [3.0.0-rc.9] — 2026-03-23
+
+### ✨ New Features
+
+- **T29** — Vertex AI SA JSON Executor: implemented using the `jose` library to handle JWT/Service Account auth, along with configurable regions in the UI and automatic partner model URL building.
+- **T42** — Image generation aspect ratio mapping: created `sizeMapper` logic for generic OpenAI formats (`size`), added native `imagen3` handling, and updated NanoBanana endpoints to utilize mapped aspect ratios automatically.
+- **T38** — Centralized model specifications: `modelSpecs.ts` created for limits and parameters per model.
+
+### 🔧 Improvements
+
+- **T40** — OpenCode CLI tools integration: native `opencode-zen` and `opencode-go` integration completed in earlier PR.
+
+---
+
+## [3.0.0-rc.8] — 2026-03-23
+
+### 🔧 Bug Fixes & Improvements (Fallback, Quota & Budget)
+
+- **T24** — `503` cooldown await fix + `406` mapping: mapped `406 Not Acceptable` to `503 Service Unavailable` with proper cooldown intervals.
+- **T25** — Provider validation fallback: graceful fallback to standard validation models when a specific `validationModelId` is not present.
+- **T36** — `403` vs `429` provider handling refinement: extracted into `errorClassifier.ts` to properly segregate hard permissions failures (`403`) from rate limits (`429`).
+- **T39** — Endpoint Fallback for `fetchAvailableModels`: implemented a tri-tier mechanism (`/models` -> `/v1/models` -> local generic catalog) + `list_models_catalog` MCP tool updates to reflect `source` and `warning`.
+- **T33** — Thinking level to budget conversion: translates qualitative thinking levels into precise budget allocations.
+- **T41** — Background task auto redirect: routes heavy background evaluation tasks to flash/efficient models automatically.
+- **T23** — Intelligent quota reset fallback: accurately extracts `x-ratelimit-reset` / `retry-after` header values or maps static cooldowns.
+
+---
+
+## [3.0.0-rc.7] — 2026-03-23 _(What's New vs v2.9.5 — will be released as v3.0.0)_
+
+> **Upgrade from v2.9.5:** 16 issues resolved · 2 community PRs merged · 2 new providers · 7 new API endpoints · 3 new features · DB migration 008+009 · 832 tests passing · 15 sub2api gap improvements (T01–T15 complete).
 
 ### 🆕 New Providers
 
@@ -111,6 +149,43 @@ OmniRoute now automatically refreshes model lists for connected providers every
 
 ---
 
+## [3.0.0-rc.7] - 2026-03-23
+
+### 🔧 Improvements (sub2api Gap Analysis — T05, T08, T09, T13, T14)
+
+- **T05** — Rate-limit DB persistence: `setConnectionRateLimitUntil()`, `isConnectionRateLimited()`, `getRateLimitedConnections()` in `providers.ts`. The existing `rate_limited_until` column is now exposed as a dedicated API — OAuth token refresh must NOT touch this field to prevent rate-limit loops.
+- **T08** — Per-API-key session limit: `max_sessions INTEGER DEFAULT 0` added to `api_keys` via auto-migration. `sessionManager.ts` gains `registerKeySession()`, `unregisterKeySession()`, `checkSessionLimit()`, and `getActiveSessionCountForKey()`. Callers in `chatCore.js` can enforce the limit and decrement on `req.close`.
+- **T09** — Codex vs Spark rate-limit scopes: `getCodexModelScope()` and `getCodexRateLimitKey()` in `codex.ts`. Standard models (`gpt-5.x-codex`, `codex-mini`) get scope `"codex"`; spark models (`codex-spark*`) get scope `"spark"`. Rate-limit keys should be `${accountId}:${scope}` so exhausting one pool doesn't block the other.
+- **T13** — Stale quota display fix: `getEffectiveQuotaUsage(used, resetAt)` returns `0` when the reset window has passed; `formatResetCountdown(resetAt)` returns a human-readable countdown string (e.g. `"2h 35m"`). Both exported from `providers.ts` + `localDb.ts` for dashboard consumption.
+- **T14** — Proxy fast-fail: new `src/lib/proxyHealth.ts` with `isProxyReachable(proxyUrl, timeoutMs=2000)` (TCP check, ≤2s instead of 30s timeout), `getCachedProxyHealth()`, `invalidateProxyHealth()`, and `getAllProxyHealthStatuses()`. Results cached 30s by default; configurable via `PROXY_FAST_FAIL_TIMEOUT_MS` / `PROXY_HEALTH_CACHE_TTL_MS`.
+
+### 🧪 Tests
+
+- Test suite: **832 tests, 0 failures**
+
+---
+
+## [3.0.0-rc.6] - 2026-03-23
+
+### 🔧 Bug Fixes & Improvements (sub2api Gap Analysis — T01–T15)
+
+- **T01** — `requested_model` column in `call_logs` (migration 009): track which model the client originally requested vs the actual routed model. Enables fallback rate analytics.
+- **T02** — Strip empty text blocks from nested `tool_result.content`: prevents Anthropic 400 errors (`text content blocks must be non-empty`) when Claude Code chains tool results.
+- **T03** — Parse `x-codex-5h-*` / `x-codex-7d-*` headers: `parseCodexQuotaHeaders()` + `getCodexResetTime()` extract Codex quota windows for precise cooldown scheduling instead of generic 5-min fallback.
+- **T04** — `X-Session-Id` header for external sticky routing: `extractExternalSessionId()` in `sessionManager.ts` reads `x-session-id` / `x-omniroute-session` headers with `ext:` prefix to avoid collision with internal SHA-256 session IDs. Nginx-compatible (hyphenated header).
+- **T06** — Account deactivated → permanent block: `isAccountDeactivated()` in `accountFallback.ts` detects 401 deactivation signals and applies a 1-year cooldown to prevent retrying permanently dead accounts.
+- **T07** — X-Forwarded-For IP validation: new `src/lib/ipUtils.ts` with `extractClientIp()` and `getClientIpFromRequest()` — skips `unknown`/non-IP entries in `X-Forwarded-For` chains (Nginx/proxy-forwarded requests).
+- **T10** — Credits exhausted → distinct fallback: `isCreditsExhausted()` in `accountFallback.ts` returns 1h cooldown with `creditsExhausted` flag, distinct from generic 429 rate limiting.
+- **T11** — `max` reasoning effort → 131072 budget tokens: `EFFORT_BUDGETS` and `THINKING_LEVEL_MAP` updated; reverse mapping now returns `"max"` for full-budget responses. Unit test updated.
+- **T12** — MiniMax M2.7 pricing entries added: `minimax-m2.7`, `MiniMax-M2.7`, `minimax-m2.7-highspeed` added to pricing table (sub2api PR #1120). M2.5/GLM-4.7/GLM-5/Kimi pricing already existed.
+- **T15** — Array content normalization: `normalizeContentToString()` helper in `openai-to-claude.ts` correctly collapses array-formatted system/tool messages to string before sending to Anthropic.
+
+### 🧪 Tests
+
+- Test suite: **832 tests, 0 failures** (unchanged from rc.5)
+
+---
+
 ## [3.0.0-rc.5] - 2026-03-22
 
 ### ✨ New Features

docs/openapi.yaml

@@ -1,7 +1,7 @@
 openapi: 3.1.0
 info:
   title: OmniRoute API
-  version: 3.0.0-rc.5
+  version: 3.0.0-rc.10
   description: |
     OmniRoute is a local-first AI API proxy router. It provides an OpenAI-compatible
     endpoint that routes requests to multiple AI providers with load balancing,

open-sse/config/providerRegistry.ts

@@ -6,6 +6,8 @@
  * is auto-generated from this registry.
  */
 
+import { platform, arch } from "os";
+
 // ── Types ─────────────────────────────────────────────────────────────────
 
 export interface RegistryModel {
@@ -96,6 +98,32 @@ const KIMI_CODING_SHARED = {
   ] as RegistryModel[],
 } as const;
 
+function mapStainlessOs() {
+  switch (platform()) {
+    case "darwin":
+      return "MacOS";
+    case "win32":
+      return "Windows";
+    case "linux":
+      return "Linux";
+    default:
+      return `Other::${platform()}`;
+  }
+}
+
+function mapStainlessArch() {
+  switch (arch()) {
+    case "x64":
+      return "x64";
+    case "arm64":
+      return "arm64";
+    case "ia32":
+      return "x86";
+    default:
+      return `other::${arch()}`;
+  }
+}
+
 // ── Registry ──────────────────────────────────────────────────────────────
 
 export const REGISTRY: Record<string, RegistryEntry> = {
@@ -112,19 +140,19 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     headers: {
       "Anthropic-Version": "2023-06-01",
       "Anthropic-Beta":
-        "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14,context-management-2025-06-27",
+        "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14,context-management-2025-06-27,prompt-caching-scope-2026-01-05",
       "Anthropic-Dangerous-Direct-Browser-Access": "true",
-      "User-Agent": "claude-cli/1.0.83 (external, cli)",
+      "User-Agent": "claude-cli/2.1.63 (external, cli)",
       "X-App": "cli",
       "X-Stainless-Helper-Method": "stream",
       "X-Stainless-Retry-Count": "0",
       "X-Stainless-Runtime-Version": "v24.3.0",
-      "X-Stainless-Package-Version": "0.55.1",
+      "X-Stainless-Package-Version": "0.74.0",
       "X-Stainless-Runtime": "node",
       "X-Stainless-Lang": "js",
-      "X-Stainless-Arch": "arm64",
-      "X-Stainless-Os": "MacOS",
-      "X-Stainless-Timeout": "60",
+      "X-Stainless-Arch": mapStainlessArch(),
+      "X-Stainless-Os": mapStainlessOs(),
+      "X-Stainless-Timeout": "600",
     },
     oauth: {
       clientIdEnv: "CLAUDE_OAUTH_CLIENT_ID",
@@ -159,6 +187,8 @@ export const REGISTRY: Record<string, RegistryEntry> = {
       clientSecretDefault: "",
     },
     models: [
+      { 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-3.1-pro", name: "Gemini 3.1 Pro" },
       { id: "gemini-3-1-pro", name: "Gemini 3.1 Pro (Alt ID)" },
       { id: "gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview" },
@@ -191,6 +221,8 @@ export const REGISTRY: Record<string, RegistryEntry> = {
       clientSecretDefault: "",
     },
     models: [
+      { 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-3.1-pro", name: "Gemini 3.1 Pro" },
       { id: "gemini-3-1-pro", name: "Gemini 3.1 Pro (Alt ID)" },
       { id: "gemini-3.1-pro-preview", name: "Gemini 3.1 Pro Preview" },
@@ -322,7 +354,11 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     alias: "ag",
     format: "antigravity",
     executor: "antigravity",
-    baseUrls: ["https://daily-cloudcode-pa.googleapis.com", "https://cloudcode-pa.googleapis.com"],
+    baseUrls: [
+      "https://daily-cloudcode-pa.googleapis.com",
+      "https://daily-cloudcode-pa.sandbox.googleapis.com",
+      "https://cloudcode-pa.googleapis.com",
+    ],
     urlBuilder: (base, model, stream) => {
       const path = stream
         ? "/v1internal:streamGenerateContent?alt=sse"
@@ -332,7 +368,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     authType: "oauth",
     authHeader: "bearer",
     headers: {
-      "User-Agent": "antigravity/1.104.0 darwin/arm64",
+      "User-Agent": `antigravity/1.107.0 ${platform()}/${arch()}`,
     },
     oauth: {
       clientIdEnv: "ANTIGRAVITY_OAUTH_CLIENT_ID",
@@ -361,9 +397,9 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     authHeader: "bearer",
     headers: {
       "copilot-integration-id": "vscode-chat",
-      "editor-version": "vscode/1.107.1",
-      "editor-plugin-version": "copilot-chat/0.26.7",
-      "user-agent": "GitHubCopilotChat/0.26.7",
+      "editor-version": "vscode/1.110.0",
+      "editor-plugin-version": "copilot-chat/0.38.0",
+      "user-agent": "GitHubCopilotChat/0.38.0",
       "openai-intent": "conversation-panel",
       "x-github-api-version": "2025-04-01",
       "x-vscode-user-agent-library-version": "electron-fetch",
@@ -1146,7 +1182,7 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     alias: "vertex",
     // Vertex AI uses Google's generateContent format (same as Gemini)
     format: "gemini",
-    executor: "default",
+    executor: "vertex",
     // URL uses {project_id} and {region} from providerSpecificData — handled by custom executor or fallback
     // Default to us-central1 / generic endpoint; users configure project via providerSpecificData
     baseUrl: "https://us-central1-aiplatform.googleapis.com/v1/projects",
@@ -1241,6 +1277,68 @@ export const REGISTRY: Record<string, RegistryEntry> = {
     ],
   },
 
+  puter: {
+    id: "puter",
+    alias: "pu",
+    format: "openai",
+    executor: "puter",
+    // OpenAI-compatible gateway with 500+ models (GPT, Claude, Gemini, Grok, DeepSeek, Qwen…)
+    // Auth: Bearer <puter_auth_token> from puter.com/dashboard → Copy Auth Token
+    // Model IDs use provider/model-name format for non-OpenAI models.
+    // Only chat completions (incl. streaming) are available via REST.
+    // Image gen, TTS, STT, video are puter.js SDK-only (browser).
+    baseUrl: "https://api.puter.com/puterai/openai/v1/chat/completions",
+    authType: "apikey",
+    authHeader: "bearer",
+    models: [
+      // OpenAI — use bare IDs
+      { id: "gpt-4o-mini", name: "GPT-4o Mini (🆓 Puter)" },
+      { id: "gpt-4o", name: "GPT-4o (Puter)" },
+      { id: "gpt-4.1", name: "GPT-4.1 (Puter)" },
+      { id: "gpt-4.1-mini", name: "GPT-4.1 Mini (Puter)" },
+      { id: "gpt-5-nano", name: "GPT-5 Nano (Puter)" },
+      { id: "gpt-5-mini", name: "GPT-5 Mini (Puter)" },
+      { id: "gpt-5", name: "GPT-5 (Puter)" },
+      { id: "o3-mini", name: "OpenAI o3-mini (Puter)" },
+      { id: "o3", name: "OpenAI o3 (Puter)" },
+      { id: "o4-mini", name: "OpenAI o4-mini (Puter)" },
+      // Anthropic Claude — use bare IDs (confirmed working)
+      { id: "claude-haiku-4-5", name: "Claude Haiku 4.5 (Puter)" },
+      { id: "claude-sonnet-4-5", name: "Claude Sonnet 4.5 (Puter)" },
+      { id: "claude-opus-4-5", name: "Claude Opus 4.5 (Puter)" },
+      { id: "claude-sonnet-4", name: "Claude Sonnet 4 (Puter)" },
+      { id: "claude-opus-4", name: "Claude Opus 4 (Puter)" },
+      // Google Gemini — use google/ prefix (confirmed working)
+      { id: "google/gemini-2.0-flash", name: "Gemini 2.0 Flash (Puter)" },
+      { id: "google/gemini-2.5-flash", name: "Gemini 2.5 Flash (Puter)" },
+      { id: "google/gemini-2.5-pro", name: "Gemini 2.5 Pro (Puter)" },
+      { id: "google/gemini-3-flash", name: "Gemini 3 Flash (Puter)" },
+      { id: "google/gemini-3-pro", name: "Gemini 3 Pro (Puter)" },
+      // DeepSeek — use deepseek/ prefix (confirmed working)
+      { id: "deepseek/deepseek-chat", name: "DeepSeek Chat (Puter)" },
+      { id: "deepseek/deepseek-r1", name: "DeepSeek R1 (Puter)" },
+      { id: "deepseek/deepseek-v3.2", name: "DeepSeek V3.2 (Puter)" },
+      // xAI Grok — use x-ai/ prefix
+      { id: "x-ai/grok-3", name: "Grok 3 (Puter)" },
+      { id: "x-ai/grok-3-mini", name: "Grok 3 Mini (Puter)" },
+      { id: "x-ai/grok-4", name: "Grok 4 (Puter)" },
+      { id: "x-ai/grok-4-fast", name: "Grok 4 Fast (Puter)" },
+      // Meta Llama — bare IDs (confirmed ✅)
+      { id: "llama-4-scout", name: "Llama 4 Scout (Puter)" },
+      { id: "llama-4-maverick", name: "Llama 4 Maverick (Puter)" },
+      { id: "llama-3.3-70b-instruct", name: "Llama 3.3 70B (Puter)" },
+      // Mistral — bare IDs (confirmed ✅)
+      { id: "mistral-small-latest", name: "Mistral Small (Puter)" },
+      { id: "mistral-medium-latest", name: "Mistral Medium (Puter)" },
+      { id: "open-mistral-nemo", name: "Mistral Nemo (Puter)" },
+      // Qwen — use qwen/ prefix (confirmed ✅)
+      { id: "qwen/qwen3-235b-a22b", name: "Qwen3 235B (Puter)" },
+      { id: "qwen/qwen3-32b", name: "Qwen3 32B (Puter)" },
+      { id: "qwen/qwen3-coder", name: "Qwen3 Coder 480B (Puter)" },
+    ],
+    passthroughModels: true, // 500+ models available — users can type any Puter model ID
+  },
+
   "cloudflare-ai": {
     id: "cloudflare-ai",
     alias: "cf",

open-sse/executors/codex.ts

@@ -3,6 +3,112 @@ import { CODEX_DEFAULT_INSTRUCTIONS } from "../config/codexInstructions.ts";
 import { PROVIDERS } from "../config/constants.ts";
 import { refreshCodexToken } from "../services/tokenRefresh.ts";
 
+// ─── T09: Codex vs Spark Scope-Aware Rate Limiting ────────────────────────
+// Codex has two independent quota pools: "codex" (standard) and "spark" (premium).
+// Exhausting one should NOT block requests to the other.
+// Ref: sub2api PR #1129 (feat(openai): split codex spark rate limiting from codex)
+
+/**
+ * Maps model name substrings to their rate-limit scope.
+ * Checked in order — first match wins.
+ */
+const CODEX_SCOPE_PATTERNS: Array<{ pattern: string; scope: "codex" | "spark" }> = [
+  { pattern: "codex-spark", scope: "spark" },
+  { pattern: "spark", scope: "spark" },
+  { pattern: "codex", scope: "codex" },
+  { pattern: "gpt-5", scope: "codex" }, // gpt-5.2-codex, gpt-5.3-codex, etc.
+];
+
+/**
+ * T09: Determine the rate-limit scope for a Codex model.
+ * Use this key as the suffix for per-scope rate limit state:
+ *   `${accountId}:${getModelScope(model)}`
+ *
+ * @param model - The Codex model ID (e.g. "gpt-5.3-codex", "codex-spark-mini")
+ * @returns "codex" | "spark"
+ */
+export function getCodexModelScope(model: string): "codex" | "spark" {
+  const lower = model.toLowerCase();
+  for (const { pattern, scope } of CODEX_SCOPE_PATTERNS) {
+    if (lower.includes(pattern)) return scope;
+  }
+  return "codex"; // default scope
+}
+
+/**
+ * T09: Get the scope-keyed rate limit identifier for an account+model combination.
+ * Use this as the key for rateLimitState maps to ensure scope isolation.
+ */
+export function getCodexRateLimitKey(accountId: string, model: string): string {
+  return `${accountId}:${getCodexModelScope(model)}`;
+}
+
+/**
+ * T03: Parsed quota snapshot from Codex response headers.
+ * Codex includes per-account usage windows that allow precise reset scheduling.
+ * Ref: sub2api PR #357 (feat(oauth): persist usage snapshots and window cooldown)
+ */
+export interface CodexQuotaSnapshot {
+  usage5h: number; // tokens used in 5h window
+  limit5h: number; // token limit for 5h window
+  resetAt5h: string | null; // ISO timestamp when 5h window resets
+  usage7d: number; // tokens used in 7d window
+  limit7d: number; // token limit for 7d window
+  resetAt7d: string | null; // ISO timestamp when 7d window resets
+}
+
+/**
+ * T03: Parse Codex-specific quota headers from a provider response.
+ * Returns null if none of the relevant headers are present.
+ *
+ * Extracts:
+ *   x-codex-5h-usage / x-codex-5h-limit / x-codex-5h-reset-at
+ *   x-codex-7d-usage / x-codex-7d-limit / x-codex-7d-reset-at
+ */
+export function parseCodexQuotaHeaders(headers: Headers): CodexQuotaSnapshot | null {
+  const usage5h = headers.get("x-codex-5h-usage");
+  const limit5h = headers.get("x-codex-5h-limit");
+  const resetAt5h = headers.get("x-codex-5h-reset-at");
+  const usage7d = headers.get("x-codex-7d-usage");
+  const limit7d = headers.get("x-codex-7d-limit");
+  const resetAt7d = headers.get("x-codex-7d-reset-at");
+
+  // Return null if none of the quota headers are present (not a quota-aware response)
+  if (!usage5h && !limit5h && !resetAt5h && !usage7d && !limit7d && !resetAt7d) {
+    return null;
+  }
+
+  return {
+    usage5h: usage5h ? parseFloat(usage5h) : 0,
+    limit5h: limit5h ? parseFloat(limit5h) : Infinity,
+    resetAt5h: resetAt5h ?? null,
+    usage7d: usage7d ? parseFloat(usage7d) : 0,
+    limit7d: limit7d ? parseFloat(limit7d) : Infinity,
+    resetAt7d: resetAt7d ?? null,
+  };
+}
+
+/**
+ * T03: Get the soonest quota reset time from a CodexQuotaSnapshot.
+ * 7d window takes priority (wider window, harder limit) but we use whichever
+ * is further in the future to avoid releasing the block too early.
+ *
+ * @returns Unix timestamp (ms) of the soonest effective reset, or null
+ */
+export function getCodexResetTime(quota: CodexQuotaSnapshot): number | null {
+  const times: number[] = [];
+  if (quota.resetAt7d) {
+    const t = new Date(quota.resetAt7d).getTime();
+    if (!isNaN(t) && t > Date.now()) times.push(t);
+  }
+  if (quota.resetAt5h) {
+    const t = new Date(quota.resetAt5h).getTime();
+    if (!isNaN(t) && t > Date.now()) times.push(t);
+  }
+  if (times.length === 0) return null;
+  return Math.max(...times); // Use furthest-out reset to avoid premature unblock
+}
+
 // Ordered list of effort levels from lowest to highest
 const EFFORT_ORDER = ["none", "low", "medium", "high", "xhigh"] as const;
 type EffortLevel = (typeof EFFORT_ORDER)[number];

open-sse/executors/github.ts

@@ -1,4 +1,4 @@
-import { BaseExecutor } from "./base.ts";
+import { BaseExecutor, ExecuteInput } from "./base.ts";
 import { PROVIDERS, OAUTH_ENDPOINTS } from "../config/constants.ts";
 import { getModelTargetFormat } from "../config/providerModels.ts";
 
@@ -19,15 +19,82 @@ export class GithubExecutor extends BaseExecutor {
     return this.config.baseUrl;
   }
 
+  injectResponseFormat(messages: any[], responseFormat: any) {
+    if (!responseFormat) return messages;
+
+    let formatInstruction = "";
+    if (responseFormat.type === "json_object") {
+      formatInstruction =
+        "Respond only with valid JSON. Do not include any text before or after the JSON object.";
+    } else if (responseFormat.type === "json_schema" && responseFormat.json_schema) {
+      formatInstruction = `Respond only with valid JSON matching this schema:\n${JSON.stringify(
+        responseFormat.json_schema.schema,
+        null,
+        2
+      )}\nDo not include any text before or after the JSON.`;
+    }
+
+    if (!formatInstruction) return messages;
+
+    const systemIdx = messages.findIndex((m: any) => m.role === "system");
+    if (systemIdx >= 0) {
+      return messages.map((m: any, i: number) =>
+        i === systemIdx ? { ...m, content: `${m.content}\n\n${formatInstruction}` } : m
+      );
+    }
+
+    return [{ role: "system", content: formatInstruction }, ...messages];
+  }
+
+  transformRequest(model: string, body: any, stream: boolean, credentials: any): any {
+    const modifiedBody = JSON.parse(JSON.stringify(body));
+    if (modifiedBody.response_format && model.toLowerCase().includes("claude")) {
+      modifiedBody.messages = this.injectResponseFormat(
+        modifiedBody.messages,
+        modifiedBody.response_format
+      );
+      delete modifiedBody.response_format;
+    }
+    return modifiedBody;
+  }
+
+  async execute(input: ExecuteInput) {
+    const result = await super.execute(input);
+    if (!result || !result.response?.body) return result;
+
+    const isStreaming = input.stream === true;
+    if (isStreaming) {
+      const decoder = new TextDecoder();
+      const transformStream = new TransformStream({
+        transform(chunk, controller) {
+          const text = decoder.decode(chunk, { stream: true });
+          if (text.includes("data: [DONE]")) {
+            return;
+          }
+          controller.enqueue(chunk);
+        },
+      });
+
+      const newResponse = new Response(result.response.body.pipeThrough(transformStream), {
+        status: result.response.status,
+        statusText: result.response.statusText,
+        headers: result.response.headers, // Headers class carries over correctly
+      });
+      result.response = newResponse;
+    }
+
+    return result;
+  }
+
   buildHeaders(credentials, stream = true) {
     const token = credentials.copilotToken || credentials.accessToken;
     return {
       Authorization: `Bearer ${token}`,
       "Content-Type": "application/json",
       "copilot-integration-id": "vscode-chat",
-      "editor-version": "vscode/1.107.1",
-      "editor-plugin-version": "copilot-chat/0.26.7",
-      "user-agent": "GitHubCopilotChat/0.26.7",
+      "editor-version": "vscode/1.110.0",
+      "editor-plugin-version": "copilot-chat/0.38.0",
+      "user-agent": "GitHubCopilotChat/0.38.0",
       "openai-intent": "conversation-panel",
       "x-github-api-version": "2025-04-01",
       "x-request-id":
@@ -44,7 +111,7 @@ export class GithubExecutor extends BaseExecutor {
         headers: {
           Authorization: `token ${githubAccessToken}`,
           "User-Agent": "GithubCopilot/1.0",
-          "Editor-Version": "vscode/1.100.0",
+          "Editor-Version": "vscode/1.110.0",
           "Editor-Plugin-Version": "copilot/1.300.0",
           Accept: "application/json",
         },

open-sse/executors/index.ts

@@ -9,6 +9,8 @@ import { DefaultExecutor } from "./default.ts";
 import { PollinationsExecutor } from "./pollinations.ts";
 import { CloudflareAIExecutor } from "./cloudflare-ai.ts";
 import { OpencodeExecutor } from "./opencode.ts";
+import { PuterExecutor } from "./puter.ts";
+import { VertexExecutor } from "./vertex.ts";
 
 const executors = {
   antigravity: new AntigravityExecutor(),
@@ -25,6 +27,9 @@ const executors = {
   cf: new CloudflareAIExecutor(), // Alias
   "opencode-zen": new OpencodeExecutor("opencode-zen"),
   "opencode-go": new OpencodeExecutor("opencode-go"),
+  puter: new PuterExecutor(),
+  pu: new PuterExecutor(), // Alias
+  vertex: new VertexExecutor(),
 };
 
 const defaultCache = new Map();
@@ -51,3 +56,5 @@ export { DefaultExecutor } from "./default.ts";
 export { PollinationsExecutor } from "./pollinations.ts";
 export { CloudflareAIExecutor } from "./cloudflare-ai.ts";
 export { OpencodeExecutor } from "./opencode.ts";
+export { PuterExecutor } from "./puter.ts";
+export { VertexExecutor } from "./vertex.ts";

open-sse/executors/puter.ts

@@ -0,0 +1,59 @@
+import { BaseExecutor } from "./base.ts";
+import { PROVIDERS } from "../config/constants.ts";
+
+/**
+ * PuterExecutor — OpenAI-compatible proxy for Puter AI.
+ *
+ * Puter exposes 500+ models (GPT, Claude, Gemini, Grok, DeepSeek, Qwen, Mistral...)
+ * through a single OpenAI-compatible REST endpoint.
+ *
+ * Endpoint: https://api.puter.com/puterai/openai/v1/chat/completions
+ * Auth:     Bearer <puter_auth_token>  (from puter.com/dashboard → Copy Auth Token)
+ * Docs:     https://docs.puter.com/AI/
+ *
+ * Model ID examples:
+ *   OpenAI:   "gpt-4o-mini", "gpt-4o", "gpt-4.1"
+ *   Claude:   "claude-sonnet-4-5", "claude-opus-4", "claude-haiku-4-5"
+ *   Gemini:   "google/gemini-2.0-flash", "google/gemini-2.5-pro"
+ *   DeepSeek: "deepseek/deepseek-chat", "deepseek/deepseek-r1"
+ *   Grok:     "x-ai/grok-3", "x-ai/grok-4"
+ *   Mistral:  "mistralai/mistral-small-3.2"
+ *   Meta:     "meta-llama/llama-3.3-70b-instruct"
+ *
+ * Note: Image generation, TTS, STT, and video are puter.js SDK-only features.
+ * Only text chat completions (with streaming SSE) are available via REST.
+ */
+export class PuterExecutor extends BaseExecutor {
+  constructor() {
+    super("puter", PROVIDERS["puter"] || { format: "openai" });
+  }
+
+  buildUrl(_model: string, _stream: boolean, _urlIndex = 0, _credentials = null): string {
+    return "https://api.puter.com/puterai/openai/v1/chat/completions";
+  }
+
+  buildHeaders(credentials: any, stream = true): Record<string, string> {
+    const headers: Record<string, string> = {
+      "Content-Type": "application/json",
+    };
+
+    const key = credentials?.apiKey || credentials?.accessToken;
+    if (key) {
+      headers["Authorization"] = `Bearer ${key}`;
+    }
+
+    if (stream) {
+      headers["Accept"] = "text/event-stream";
+    }
+
+    return headers;
+  }
+
+  transformRequest(model: string, body: any, _stream: boolean, _credentials: any): any {
+    // Puter accepts model IDs directly from its catalog.
+    // No transformation required — model string is passed as-is.
+    return body;
+  }
+}
+
+export default PuterExecutor;

open-sse/executors/vertex.ts

@@ -0,0 +1,150 @@
+import { SignJWT, importPKCS8 } from "jose";
+import { BaseExecutor, ExecuteInput } from "./base.ts";
+import { PROVIDERS } from "../config/constants.ts";
+
+interface ServiceAccount {
+  type: string;
+  project_id: string;
+  private_key_id: string;
+  private_key: string;
+  client_email: string;
+  [key: string]: unknown;
+}
+
+const TOKEN_CACHE = new Map<string, { token: string; expiresAt: number }>();
+
+function parseSAFromApiKey(apiKey: string): ServiceAccount {
+  try {
+    return JSON.parse(apiKey);
+  } catch {
+    throw new Error("Vertex AI requires a valid Service Account JSON as the API key");
+  }
+}
+
+async function getAccessToken(sa: ServiceAccount): Promise<string> {
+  if (!sa.client_email || !sa.private_key) {
+    throw new Error(
+      "Service Account JSON is missing required fields (client_email or private_key)"
+    );
+  }
+
+  const cacheKey = sa.client_email;
+  const cached = TOKEN_CACHE.get(cacheKey);
+
+  // Buffer of 60 seconds
+  if (cached && Date.now() < cached.expiresAt - 60_000) {
+    return cached.token;
+  }
+
+  const privateKey = await importPKCS8(sa.private_key, "RS256");
+  const now = Math.floor(Date.now() / 1000);
+
+  const jwt = await new SignJWT({
+    iss: sa.client_email,
+    sub: sa.client_email,
+    aud: "https://oauth2.googleapis.com/token",
+    iat: now,
+    exp: now + 3600,
+    scope: "https://www.googleapis.com/auth/cloud-platform",
+  })
+    .setProtectedHeader({ alg: "RS256", kid: sa.private_key_id })
+    .sign(privateKey);
+
+  const tokenRes = await fetch("https://oauth2.googleapis.com/token", {
+    method: "POST",
+    headers: { "Content-Type": "application/x-www-form-urlencoded" },
+    body: new URLSearchParams({
+      grant_type: "urn:ietf:params:oauth:grant-type:jwt-bearer",
+      assertion: jwt,
+    }),
+  });
+
+  if (!tokenRes.ok) {
+    const errorText = await tokenRes.text();
+    throw new Error(
+      `Failed to exchange JWT for Vertex access token: ${tokenRes.status} ${errorText}`
+    );
+  }
+
+  const tokenData = await tokenRes.json();
+  const accessToken = tokenData.access_token;
+
+  if (!accessToken) {
+    throw new Error("Vertex AI token exchange succeeded but no access_token found");
+  }
+
+  TOKEN_CACHE.set(cacheKey, {
+    token: accessToken,
+    expiresAt: (now + 3600) * 1000,
+  });
+
+  return accessToken;
+}
+
+const PARTNER_MODELS = new Set([
+  "claude-3-5-sonnet",
+  "claude-3-opus",
+  "claude-3-haiku",
+  "deepseek-v3",
+  "deepseek-v3.2",
+  "deepseek-deepseek-r1",
+  "qwen3-next-80b",
+  "llama-3.1",
+  "mistral-",
+  "glm-5",
+  "meta/llama",
+]);
+
+function isPartnerModel(model: string) {
+  return [...PARTNER_MODELS].some((prefix) => model.startsWith(prefix));
+}
+
+export class VertexExecutor extends BaseExecutor {
+  constructor() {
+    super("vertex", PROVIDERS.vertex);
+  }
+
+  async execute(input: ExecuteInput) {
+    const { credentials, log } = input;
+    if (credentials.apiKey && !credentials.accessToken) {
+      try {
+        const sa = parseSAFromApiKey(credentials.apiKey);
+        credentials.accessToken = await getAccessToken(sa);
+      } catch (err: any) {
+        log?.error?.("VERTEX", `Failed to generate JWT token: ${err.message}`);
+        throw err;
+      }
+    }
+    return super.execute(input);
+  }
+
+  buildUrl(model: string, stream: boolean, urlIndex = 0, credentials: any = null) {
+    const region = credentials?.providerSpecificData?.region || "us-central1";
+    let project = "unknown-project";
+
+    if (credentials?.apiKey) {
+      try {
+        const sa = parseSAFromApiKey(credentials.apiKey);
+        if (sa.project_id) project = sa.project_id;
+      } catch {
+        // Ignored, handled in execute
+      }
+    }
+
+    if (isPartnerModel(model)) {
+      return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/global/endpoints/openapi/chat/completions`;
+    }
+    return `https://aiplatform.googleapis.com/v1/projects/${project}/locations/${region}/publishers/google/models/${model}:${stream ? "streamGenerateContent?alt=sse" : "generateContent"}`;
+  }
+
+  buildHeaders(credentials: any, stream = true) {
+    const headers: Record<string, string> = { "Content-Type": "application/json" };
+    if (credentials.accessToken) {
+      headers["Authorization"] = `Bearer ${credentials.accessToken}`;
+    }
+    if (stream) {
+      headers["Accept"] = "text/event-stream";
+    }
+    return headers;
+  }
+}

open-sse/handlers/chatCore.ts

@@ -16,6 +16,8 @@ import { resolveModelAlias } from "../services/modelDeprecation.ts";
 import { getUnsupportedParams } from "../config/providerRegistry.ts";
 import { createErrorResult, parseUpstreamError, formatProviderError } from "../utils/error.ts";
 import { HTTP_STATUS } from "../config/constants.ts";
+import { classifyProviderError, PROVIDER_ERROR_TYPES } from "../services/errorClassifier.ts";
+import { updateProviderConnection } from "@/lib/db/providers";
 import { handleBypassRequest } from "../utils/bypassHandler.ts";
 import {
   saveRequestUsage,
@@ -25,6 +27,11 @@ import {
 } from "@/lib/usageDb";
 import { getModelNormalizeToolCallId, getModelPreserveOpenAIDeveloperRole } from "@/lib/localDb";
 import { getExecutor } from "../executors/index.ts";
+import {
+  parseCodexQuotaHeaders,
+  getCodexResetTime,
+  getCodexModelScope,
+} from "../executors/codex.ts";
 import { translateNonStreamingResponse } from "./responseTranslator.ts";
 import { extractUsageFromResponse } from "./usageExtractor.ts";
 import { parseSSEToOpenAIResponse, parseSSEToResponsesOutput } from "./sseParser.ts";
@@ -44,6 +51,11 @@ import { getIdempotencyKey, checkIdempotency, saveIdempotency } from "@/lib/idem
 import { createProgressTransform, wantsProgress } from "../utils/progressTracker.ts";
 import { isModelUnavailableError, getNextFamilyFallback } from "../services/modelFamilyFallback.ts";
 import { computeRequestHash, deduplicate, shouldDeduplicate } from "../services/requestDedup.ts";
+import {
+  isBackgroundTask,
+  getDegradedModel,
+  getBackgroundDegradationConfig,
+} from "../services/backgroundTaskDetector.ts";
 import {
   shouldUseFallback,
   isFallbackDecision,
@@ -93,7 +105,9 @@ export async function handleChatCore({
   userAgent,
   comboName,
 }) {
-  const { provider, model, extendedContext } = modelInfo;
+  let { provider, model, extendedContext } = modelInfo;
+  const requestedModel =
+    typeof body?.model === "string" && body.model.trim().length > 0 ? body.model : model;
   const startTime = Date.now();
   const persistFailureUsage = (statusCode: number, errorCode?: string | null) => {
     saveRequestUsage({
@@ -112,6 +126,67 @@ export async function handleChatCore({
     }).catch(() => {});
   };
 
+  const persistCodexQuotaState = async (
+    headers: Headers | Record<string, string> | null,
+    status = 0
+  ) => {
+    if (provider !== "codex" || !connectionId || !headers) return;
+
+    try {
+      const quota = parseCodexQuotaHeaders(headers as Headers);
+      if (!quota) return;
+
+      const existingProviderData =
+        credentials?.providerSpecificData && typeof credentials.providerSpecificData === "object"
+          ? credentials.providerSpecificData
+          : {};
+      const scope = getCodexModelScope(model || requestedModel || "");
+      const quotaState = {
+        usage5h: quota.usage5h,
+        limit5h: quota.limit5h,
+        resetAt5h: quota.resetAt5h,
+        usage7d: quota.usage7d,
+        limit7d: quota.limit7d,
+        resetAt7d: quota.resetAt7d,
+        scope,
+        updatedAt: new Date().toISOString(),
+      };
+
+      const nextProviderData: Record<string, unknown> = {
+        ...existingProviderData,
+        codexQuotaState: quotaState,
+      };
+
+      // T03/T09: on 429, persist exact reset time per scope to avoid global over-blocking.
+      if (status === 429) {
+        const resetTimeMs = getCodexResetTime(quota);
+        if (resetTimeMs && resetTimeMs > Date.now()) {
+          const scopeUntil = new Date(resetTimeMs).toISOString();
+          const scopeMapRaw =
+            existingProviderData &&
+            typeof existingProviderData === "object" &&
+            existingProviderData.codexScopeRateLimitedUntil &&
+            typeof existingProviderData.codexScopeRateLimitedUntil === "object"
+              ? existingProviderData.codexScopeRateLimitedUntil
+              : {};
+
+          nextProviderData.codexScopeRateLimitedUntil = {
+            ...(scopeMapRaw as Record<string, unknown>),
+            [scope]: scopeUntil,
+          };
+        }
+      }
+
+      await updateProviderConnection(connectionId, {
+        providerSpecificData: nextProviderData,
+      });
+
+      credentials.providerSpecificData = nextProviderData;
+    } catch (err) {
+      log?.debug?.("CODEX", `Failed to persist codex quota state: ${err?.message || err}`);
+    }
+  };
+
   // ── Phase 9.2: Idempotency check ──
   const idempotencyKey = getIdempotencyKey(clientRawRequest?.headers);
   const cachedIdemp = checkIdempotency(idempotencyKey);
@@ -157,6 +232,22 @@ export async function handleChatCore({
   // Detect source format and get target format
   // Model-specific targetFormat takes priority over provider default
 
+  // ── Background Task Redirection (T41) ──
+  const bgConfig = getBackgroundDegradationConfig();
+  if (bgConfig.enabled && isBackgroundTask(body, clientRawRequest?.headers)) {
+    const degradedModel = getDegradedModel(model);
+    if (degradedModel !== model) {
+      log?.info?.(
+        "BACKGROUND",
+        `Background task detected: Redirecting ${model} → ${degradedModel}`
+      );
+      model = degradedModel;
+      if (body && typeof body === "object") {
+        body.model = model;
+      }
+    }
+  }
+
   // Apply custom model aliases (Settings → Model Aliases → Pattern→Target) before routing (#315, #472)
   // Custom aliases take priority over built-in and must be resolved here so the
   // downstream getModelTargetFormat() lookup AND the actual provider request use
@@ -173,7 +264,17 @@ export async function handleChatCore({
   const targetFormat = modelTargetFormat || getTargetFormat(provider);
 
   // Default to false unless client explicitly sets stream: true (OpenAI spec compliant)
-  const stream = body.stream === true;
+  const acceptHeader =
+    clientRawRequest?.headers && typeof clientRawRequest.headers.get === "function"
+      ? clientRawRequest.headers.get("accept") || clientRawRequest.headers.get("Accept")
+      : (clientRawRequest?.headers || {})["accept"] || (clientRawRequest?.headers || {})["Accept"];
+
+  const clientWantsJson =
+    typeof acceptHeader === "string" &&
+    acceptHeader.includes("application/json") &&
+    !acceptHeader.includes("text/event-stream");
+
+  const stream = body.stream === true && !clientWantsJson;
 
   // ── Phase 9.1: Semantic cache check (non-streaming, temp=0 only) ──
   if (isCacheable(body, clientRawRequest?.headers)) {
@@ -457,7 +558,7 @@ export async function handleChatCore({
       // Non-stream responses need cloning for shared dedup consumers.
       const status = rawResult.response.status;
       const statusText = rawResult.response.statusText;
-      const headers = Array.from(rawResult.response.headers.entries());
+      const headers = Array.from(rawResult.response.headers.entries()) as [string, string][];
       const payload = await rawResult.response.text();
 
       return {
@@ -532,6 +633,7 @@ export async function handleChatCore({
       path: clientRawRequest?.endpoint || "/v1/chat/completions",
       status: error.name === "AbortError" ? 499 : HTTP_STATUS.BAD_GATEWAY,
       model,
+      requestedModel,
       provider,
       connectionId,
       duration: Date.now() - startTime,
@@ -603,6 +705,8 @@ export async function handleChatCore({
     }
   }
 
+  await persistCodexQuotaState(providerResponse.headers, providerResponse.status);
+
   // Check provider response - return error info for fallback handling
   if (!providerResponse.ok) {
     trackPendingRequest(model, provider, connectionId, false);
@@ -610,6 +714,54 @@ export async function handleChatCore({
       providerResponse,
       provider
     );
+
+    // T06/T10/T36: classify provider errors and persist terminal account states.
+    const errorType = classifyProviderError(statusCode, message);
+    if (connectionId && errorType) {
+      try {
+        if (errorType === PROVIDER_ERROR_TYPES.FORBIDDEN) {
+          await updateProviderConnection(connectionId, {
+            isActive: false,
+            testStatus: "banned",
+            lastErrorType: errorType,
+            lastError: message,
+            errorCode: statusCode,
+          });
+          console.warn(
+            `[provider] Node ${connectionId} banned (${statusCode}) — disabling permanently`
+          );
+        } else if (errorType === PROVIDER_ERROR_TYPES.QUOTA_EXHAUSTED) {
+          await updateProviderConnection(connectionId, {
+            testStatus: "credits_exhausted",
+            lastErrorType: errorType,
+            lastError: message,
+            errorCode: statusCode,
+          });
+          console.warn(`[provider] Node ${connectionId} exhausted quota (${statusCode})`);
+        } else if (errorType === PROVIDER_ERROR_TYPES.ACCOUNT_DEACTIVATED) {
+          await updateProviderConnection(connectionId, {
+            isActive: false,
+            testStatus: "expired",
+            lastErrorType: errorType,
+            lastError: message,
+            errorCode: statusCode,
+          });
+          console.warn(
+            `[provider] Node ${connectionId} account deactivated (${statusCode}) — marked expired`
+          );
+        } else if (errorType === PROVIDER_ERROR_TYPES.UNAUTHORIZED) {
+          // Normal 401 (token/session auth issue): keep account active for refresh/re-auth.
+          await updateProviderConnection(connectionId, {
+            lastErrorType: errorType,
+            lastError: message,
+            errorCode: statusCode,
+          });
+        }
+      } catch {
+        // Best-effort state update; request flow should continue with fallback handling.
+      }
+    }
+
     appendRequestLog({ model, provider, connectionId, status: `FAILED ${statusCode}` }).catch(
       () => {}
     );
@@ -618,6 +770,7 @@ export async function handleChatCore({
       path: clientRawRequest?.endpoint || "/v1/chat/completions",
       status: statusCode,
       model,
+      requestedModel,
       provider,
       connectionId,
       duration: Date.now() - startTime,
@@ -808,6 +961,7 @@ export async function handleChatCore({
       path: clientRawRequest?.endpoint || "/v1/chat/completions",
       status: 200,
       model,
+      requestedModel,
       provider,
       connectionId,
       duration: Date.now() - startTime,
@@ -848,6 +1002,32 @@ export async function handleChatCore({
       ? translateNonStreamingResponse(responseBody, targetFormat, sourceFormat)
       : responseBody;
 
+    // T26: Strip markdown code blocks if provider format is Claude
+    if (sourceFormat === "claude" && !stream) {
+      if (translatedResponse?.choices?.[0]?.message?.content) {
+        const text = translatedResponse.choices[0].message.content;
+        const codeBlockRegex =
+          /^```(?:json|javascript|typescript|js|ts)?\s*\n?([\s\S]*?)\n?```\s*$/;
+        const match = text.trim().match(codeBlockRegex);
+        if (match) {
+          translatedResponse.choices[0].message.content = match[1].trim();
+        }
+      }
+    }
+
+    // T18: Normalize finish_reason to 'tool_calls' if tool calls are present
+    if (translatedResponse?.choices) {
+      for (const choice of translatedResponse.choices) {
+        if (
+          choice.message?.tool_calls &&
+          choice.message.tool_calls.length > 0 &&
+          choice.finish_reason !== "tool_calls"
+        ) {
+          choice.finish_reason = "tool_calls";
+        }
+      }
+    }
+
     // Sanitize response for OpenAI SDK compatibility
     // Strips non-standard fields (x_groq, usage_breakdown, service_tier, etc.)
     // Extracts <think> tags into reasoning_content
@@ -921,6 +1101,7 @@ export async function handleChatCore({
       path: clientRawRequest?.endpoint || "/v1/chat/completions",
       status: streamStatus || 200,
       model,
+      requestedModel,
       provider,
       connectionId,
       duration: Date.now() - startTime,

open-sse/handlers/imageGeneration.ts

@@ -16,6 +16,7 @@
  */
 
 import { getImageProvider, parseImageModel } from "../config/imageRegistry.ts";
+import { mapImageSize } from "../translator/image/sizeMapper.ts";
 import { saveCallLog } from "@/lib/usageDb";
 import {
   submitComfyWorkflow,
@@ -95,11 +96,21 @@ export async function handleImageGeneration({ body, credentials, log, resolvedPr
     });
   }
 
-  // Route to format-specific handler
   if (providerConfig.format === "gemini-image") {
     return handleGeminiImageGeneration({ model, providerConfig, body, credentials, log });
   }
 
+  if (providerConfig.format === "imagen3") {
+    return handleImagen3ImageGeneration({
+      model,
+      provider,
+      providerConfig,
+      body,
+      credentials,
+      log,
+    });
+  }
+
   if (providerConfig.format === "hyperbolic") {
     return handleHyperbolicImageGeneration({
       model,
@@ -539,7 +550,7 @@ async function handleNanoBananaImageGeneration({
       ? body.aspectRatio
       : typeof body.aspect_ratio === "string"
         ? body.aspect_ratio
-        : inferAspectRatioFromSize(body.size) || "1:1";
+        : mapImageSize(body.size);
 
   let resolution =
     typeof body.resolution === "string"
@@ -856,18 +867,6 @@ async function normalizeNanoBananaTaskResult(taskData, body, log) {
   return [];
 }
 
-function inferAspectRatioFromSize(size) {
-  if (typeof size !== "string") return null;
-  const [wRaw, hRaw] = size.split("x");
-  const width = Number(wRaw);
-  const height = Number(hRaw);
-  if (!Number.isFinite(width) || !Number.isFinite(height) || width <= 0 || height <= 0) return null;
-
-  const gcd = (a, b) => (b === 0 ? a : gcd(b, a % b));
-  const div = gcd(Math.round(width), Math.round(height));
-  return `${Math.round(width / div)}:${Math.round(height / div)}`;
-}
-
 function inferResolutionFromSize(size) {
   if (typeof size !== "string") return null;
   const [wRaw, hRaw] = size.split("x");
@@ -1081,3 +1080,113 @@ async function handleComfyUIImageGeneration({ model, provider, providerConfig, b
     return { success: false, status: 502, error: `Image provider error: ${err.message}` };
   }
 }
+
+/**
+ * Handle Imagen 3 image generation
+ */
+async function handleImagen3ImageGeneration({
+  model,
+  provider,
+  providerConfig,
+  body,
+  credentials,
+  log,
+}: any) {
+  const startTime = Date.now();
+  const token = credentials.apiKey || credentials.accessToken;
+  const aspectRatio = mapImageSize(body.size);
+
+  const upstreamBody = {
+    prompt: body.prompt,
+    aspect_ratio: aspectRatio,
+    number_of_images: body.n ?? 1,
+  };
+
+  if (log) {
+    const promptPreview = String(body.prompt ?? "").slice(0, 60);
+    log.info(
+      "IMAGE",
+      `${provider}/${model} (imagen3) | prompt: "${promptPreview}..." | aspect_ratio: ${aspectRatio}`
+    );
+  }
+
+  try {
+    const response = await fetch(providerConfig.baseUrl, {
+      method: "POST",
+      headers: {
+        "Content-Type": "application/json",
+        Authorization: `Bearer ${token}`,
+      },
+      body: JSON.stringify(upstreamBody),
+    });
+
+    if (!response.ok) {
+      const errorText = await response.text();
+      if (log)
+        log.error("IMAGE", `${provider} error ${response.status}: ${errorText.slice(0, 200)}`);
+
+      saveCallLog({
+        method: "POST",
+        path: "/v1/images/generations",
+        status: response.status,
+        model: `${provider}/${model}`,
+        provider,
+        duration: Date.now() - startTime,
+        error: errorText.slice(0, 500),
+        requestBody: upstreamBody,
+      }).catch(() => {});
+
+      return { success: false, status: response.status, error: errorText };
+    }
+
+    const data = await response.json();
+
+    // Normalize response to OpenAI format
+    const images: any[] = [];
+    if (Array.isArray(data.images)) {
+      images.push(
+        ...data.images.map((img: any) => ({
+          b64_json: img.image || img.b64_json || img.url || img,
+          revised_prompt: body.prompt,
+        }))
+      );
+    } else if (Array.isArray(data.data)) {
+      images.push(...data.data);
+    } else if (data.url || data.b64_json || data.image) {
+      images.push({
+        b64_json: data.image || data.b64_json || data.url,
+        url: data.url,
+        revised_prompt: body.prompt,
+      });
+    }
+
+    saveCallLog({
+      method: "POST",
+      path: "/v1/images/generations",
+      status: 200,
+      model: `${provider}/${model}`,
+      provider,
+      duration: Date.now() - startTime,
+      responseBody: { images_count: images.length },
+    }).catch(() => {});
+
+    return {
+      success: true,
+      data: { created: data.created || Math.floor(Date.now() / 1000), data: images },
+    };
+  } catch (err: any) {
+    if (log) log.error("IMAGE", `${provider} fetch error: ${err.message}`);
+
+    saveCallLog({
+      method: "POST",
+      path: "/v1/images/generations",
+      status: 502,
+      model: `${provider}/${model}`,
+      provider,
+      duration: Date.now() - startTime,
+      error: err.message,
+    }).catch(() => {});
+
+    return { success: false, status: 502, error: `Image provider error: ${err.message}` };
+  }
+}

open-sse/handlers/responseTranslator.ts

@@ -103,6 +103,25 @@ export function translateNonStreamingResponse(
       message.content = "";
     }
 
+    if (process.env.DEBUG_RESPONSES_SSE_TO_JSON === "true") {
+      const msgItems = output.filter((i) => toRecord(i).type === "message");
+      console.log(`[ResponsesSSE] ${output.length} output items, ${msgItems.length} message items`);
+      msgItems.forEach((item, idx) => {
+        const itemObj = toRecord(item);
+        let textLen = 0;
+        if (Array.isArray(itemObj.content)) {
+          for (const part of itemObj.content) {
+            const partObj = toRecord(part);
+            if (partObj.type === "output_text" && typeof partObj.text === "string") {
+              textLen += partObj.text.length;
+            }
+          }
+        }
+        console.log(`  [${idx}] text length: ${textLen}`);
+      });
+      console.log(`  → Final text content length: ${textContent.length}`);
+    }
+
     const createdAt = toNumber(response.created_at, Math.floor(Date.now() / 1000));
     const model = toString(response.model || responseRoot.model, "openai-responses");
     const finishReason = toolCalls.length > 0 ? "tool_calls" : "stop";

open-sse/handlers/sseParser.ts

@@ -23,9 +23,18 @@ export function parseSSEToOpenAIResponse(rawSSE, fallbackModel) {
   const first = chunks[0];
   const contentParts = [];
   const reasoningParts = [];
+  const accumulatedToolCalls = new Map<string, any>();
+  let unknownToolCallSeq = 0;
   let finishReason = "stop";
   let usage = null;
 
+  const getToolCallKey = (toolCall: any) => {
+    if (toolCall?.id) return `id:${toolCall.id}`;
+    if (Number.isInteger(toolCall?.index)) return `idx:${toolCall.index}`;
+    unknownToolCallSeq += 1;
+    return `seq:${unknownToolCallSeq}`;
+  };
+
   for (const chunk of chunks) {
     const choice = chunk?.choices?.[0];
     const delta = choice?.delta || {};
@@ -36,6 +45,40 @@ export function parseSSEToOpenAIResponse(rawSSE, fallbackModel) {
     if (typeof delta.reasoning_content === "string" && delta.reasoning_content.length > 0) {
       reasoningParts.push(delta.reasoning_content);
     }
+
+    // T18: Accumulate tool calls correctly across streamed chunks
+    if (delta.tool_calls) {
+      for (const tc of delta.tool_calls) {
+        const key = getToolCallKey(tc);
+        const existing = accumulatedToolCalls.get(key);
+        const deltaArgs = typeof tc?.function?.arguments === "string" ? tc.function.arguments : "";
+
+        if (!existing) {
+          accumulatedToolCalls.set(key, {
+            id: tc?.id ?? null,
+            index: Number.isInteger(tc?.index) ? tc.index : accumulatedToolCalls.size,
+            type: tc?.type || "function",
+            function: {
+              name: tc?.function?.name || "unknown",
+              arguments: deltaArgs,
+            },
+          });
+        } else {
+          existing.id = existing.id || tc?.id || null;
+          if (!Number.isInteger(existing.index) && Number.isInteger(tc?.index)) {
+            existing.index = tc.index;
+          }
+          if (tc?.function?.name && !existing.function?.name) {
+            existing.function = existing.function || {};
+            existing.function.name = tc.function.name;
+          }
+          existing.function = existing.function || {};
+          existing.function.arguments = `${existing.function.arguments || ""}${deltaArgs}`;
+          accumulatedToolCalls.set(key, existing);
+        }
+      }
+    }
+
     if (choice?.finish_reason) {
       finishReason = choice.finish_reason;
     }
@@ -46,12 +89,22 @@ export function parseSSEToOpenAIResponse(rawSSE, fallbackModel) {
 
   const message: Record<string, unknown> = {
     role: "assistant",
-    content: contentParts.join(""),
+    content: contentParts.length > 0 ? contentParts.join("") : null,
   };
   if (reasoningParts.length > 0) {
     message.reasoning_content = reasoningParts.join("");
   }
 
+  const finalToolCalls = [...accumulatedToolCalls.values()].filter(Boolean).sort((a, b) => {
+    const ai = Number.isInteger(a?.index) ? a.index : 0;
+    const bi = Number.isInteger(b?.index) ? b.index : 0;
+    return ai - bi;
+  });
+  if (finalToolCalls.length > 0) {
+    finishReason = "tool_calls"; // T18 normalization
+    message.tool_calls = finalToolCalls;
+  }
+
   const result: Record<string, unknown> = {
     id: first.id || `chatcmpl-${Date.now()}`,
     object: "chat.completion",

open-sse/mcp-server/server.ts

@@ -433,23 +433,48 @@ async function handleListModelsCatalog(args: { provider?: string; capability?: s
   const start = Date.now();
   try {
     let path = "/v1/models";
-    const params = new URLSearchParams();
-    if (args.provider) params.set("provider", args.provider);
-    if (args.capability) params.set("capability", args.capability);
-    if (params.toString()) path += `?${params.toString()}`;
+    let isProviderSpecific = false;
+    let source = "local_catalog";
+    let warning = undefined;
+
+    if (args.provider && !args.capability) {
+      // Use direct provider fetch to get real-time API status
+      path = `/api/providers/${encodeURIComponent(args.provider)}/models`;
+      isProviderSpecific = true;
+    } else {
+      const params = new URLSearchParams();
+      if (args.provider) params.set("provider", args.provider);
+      if (args.capability) params.set("capability", args.capability);
+      if (params.toString()) path += `?${params.toString()}`;
+    }
 
     const raw = toRecord(await omniRouteFetch(path));
+
+    // If we used the direct provider endpoint
+    let rawModels = [];
+    if (isProviderSpecific) {
+      rawModels = Array.isArray(raw.models) ? raw.models : [];
+      source = typeof raw.source === "string" ? raw.source : "api";
+      if (raw.warning) warning = String(raw.warning);
+    } else {
+      rawModels = Array.isArray(raw.data) ? raw.data : [];
+      source = "local_catalog";
+      // OmniRoute's global /v1/models is always a cached/local catalog
+    }
+
     const result = {
-      models: toArray(raw.data).map((rawModel) => {
+      models: rawModels.map((rawModel) => {
         const model = toRecord(rawModel);
         return {
           id: toString(model.id, ""),
-          provider: toString(model.owned_by, toString(model.provider, "unknown")),
+          provider: toString(model.owned_by, toString(model.provider, args.provider || "unknown")),
           capabilities: toStringArray(model.capabilities, ["chat"]),
           status: toString(model.status, "available"),
           pricing: model.pricing,
         };
       }),
+      source,
+      ...(warning ? { warning } : {}),
     };
 
     await logToolCall(

open-sse/services/accountFallback.ts

@@ -8,6 +8,46 @@ import {
 } from "../config/constants.ts";
 import { getProviderCategory } from "../config/providerRegistry.ts";
 
+// T06 (sub2api PR #1037): Signals that indicate permanent account deactivation.
+// When a 401 body contains these strings, the account is permanently dead
+// and should NOT be retried after token refresh.
+export const ACCOUNT_DEACTIVATED_SIGNALS = [
+  "account_deactivated",
+  "account has been deactivated",
+  "account has been disabled",
+  "your account has been suspended",
+  "this account is deactivated",
+];
+
+// T10 (sub2api PR #1169): Signals that indicate billing credits are exhausted.
+// Distinct from rate-limit 429 — the account won't recover until credits are added.
+export const CREDITS_EXHAUSTED_SIGNALS = [
+  "insufficient_quota",
+  "billing_hard_limit_reached",
+  "exceeded your current quota",
+  "credit_balance_too_low",
+  "your credit balance is too low",
+  "credits exhausted",
+  "out of credits",
+  "payment required",
+];
+
+/**
+ * T06: Returns true if response body indicates the account is permanently deactivated.
+ */
+export function isAccountDeactivated(errorText: string): boolean {
+  const lower = String(errorText || "").toLowerCase();
+  return ACCOUNT_DEACTIVATED_SIGNALS.some((sig) => lower.includes(sig));
+}
+
+/**
+ * T10: Returns true if response body indicates credits/quota are permanently exhausted.
+ */
+export function isCreditsExhausted(errorText: string): boolean {
+  const lower = String(errorText || "").toLowerCase();
+  return CREDITS_EXHAUSTED_SIGNALS.some((sig) => lower.includes(sig));
+}
+
 // ─── Provider Profile Helper ────────────────────────────────────────────────
 
 /**
@@ -201,6 +241,14 @@ export function classifyErrorText(errorText) {
   ) {
     return RateLimitReason.QUOTA_EXHAUSTED;
   }
+  // T10: credits_exhausted signals
+  if (isCreditsExhausted(errorText)) {
+    return RateLimitReason.QUOTA_EXHAUSTED;
+  }
+  // T06: account_deactivated signals
+  if (isAccountDeactivated(errorText)) {
+    return RateLimitReason.AUTH_ERROR;
+  }
   if (
     lower.includes("rate limit") ||
     lower.includes("too many requests") ||
@@ -294,13 +342,67 @@ export function checkFallbackError(
   errorText,
   backoffLevel = 0,
   model = null,
-  provider = null
+  provider = null,
+  headers = null
 ) {
+  const errorStr = (errorText || "").toString();
+
+  function parseResetFromHeaders(headers, errorStr = "") {
+    if (!headers) return null;
+
+    // Retry-After header
+    const retryAfter =
+      typeof headers.get === "function"
+        ? headers.get("retry-after")
+        : headers["retry-after"] || headers["Retry-After"];
+
+    if (retryAfter) {
+      const seconds = parseInt(retryAfter, 10);
+      if (!isNaN(seconds) && String(seconds) === String(retryAfter).trim()) {
+        return Date.now() + seconds * 1000;
+      }
+      const date = new Date(retryAfter);
+      if (!isNaN(date.getTime())) return date.getTime();
+    }
+
+    // X-RateLimit-Reset
+    const rlReset =
+      typeof headers.get === "function"
+        ? headers.get("x-ratelimit-reset")
+        : headers["x-ratelimit-reset"] || headers["X-RateLimit-Reset"];
+
+    if (rlReset) {
+      const ts = parseInt(rlReset, 10);
+      if (!isNaN(ts)) {
+        return ts > 10000000000 ? ts : ts * 1000;
+      }
+    }
+    return null;
+  }
   // Check error message FIRST - specific patterns take priority over status codes
   if (errorText) {
-    const errorStr = typeof errorText === "string" ? errorText : JSON.stringify(errorText);
     const lowerError = errorStr.toLowerCase();
 
+    // T06 (sub2api #1037): Permanent account deactivation — do NOT retry, mark as permanent failure
+    if (isAccountDeactivated(errorStr)) {
+      return {
+        shouldFallback: true,
+        cooldownMs: 365 * 24 * 60 * 60 * 1000, // 1 year = effectively permanent
+        reason: RateLimitReason.AUTH_ERROR,
+        permanent: true,
+      };
+    }
+
+    // T10 (sub2api #1169): Credits/quota exhausted — long cooldown, distinct from rate limit
+    if (isCreditsExhausted(errorStr)) {
+      return {
+        shouldFallback: true,
+        cooldownMs: COOLDOWN_MS.paymentRequired ?? 3600 * 1000, // 1h cooldown
+        reason: RateLimitReason.QUOTA_EXHAUSTED,
+        creditsExhausted: true,
+      };
+    }
+
     if (lowerError.includes("no credentials")) {
       return {
         shouldFallback: true,
@@ -325,6 +427,18 @@ export function checkFallbackError(
       lowerError.includes("capacity") ||
       lowerError.includes("overloaded")
     ) {
+      const resetTime = parseResetFromHeaders(headers);
+      if (resetTime) {
+        const waitMs = resetTime - Date.now();
+        if (waitMs > 60_000) {
+          return {
+            shouldFallback: true,
+            cooldownMs: waitMs,
+            newBackoffLevel: 0,
+            reason: RateLimitReason.RATE_LIMIT_EXCEEDED,
+          };
+        }
+      }
       const newLevel = Math.min(backoffLevel + 1, BACKOFF_CONFIG.maxLevel);
       const reason = classifyErrorText(errorStr);
       return {
@@ -362,6 +476,19 @@ export function checkFallbackError(
 
   // 429 - Rate limit with exponential backoff
   if (status === HTTP_STATUS.RATE_LIMITED) {
+    const resetTime = parseResetFromHeaders(headers);
+    if (resetTime) {
+      const waitMs = resetTime - Date.now();
+      if (waitMs > 60_000) {
+        return {
+          shouldFallback: true,
+          cooldownMs: waitMs,
+          newBackoffLevel: 0,
+          reason: RateLimitReason.RATE_LIMIT_EXCEEDED,
+        };
+      }
+    }
+
     const newLevel = Math.min(backoffLevel + 1, BACKOFF_CONFIG.maxLevel);
     return {
       shouldFallback: true,

open-sse/services/combo.ts

@@ -841,7 +841,8 @@ export async function handleComboChat({
         errorText,
         0,
         null,
-        provider
+        provider,
+        result.headers
       );
 
       // Record failure in circuit breaker for transient errors
@@ -865,6 +866,12 @@ export async function handleComboChat({
       if (!lastStatus) lastStatus = result.status;
       if (i > 0) fallbackCount++;
       log.warn("COMBO", `Model ${modelStr} failed, trying next`, { status: result.status });
+
+      if ([502, 503, 504].includes(result.status) && cooldownMs > 0 && cooldownMs <= 5000) {
+        log.info("COMBO", `Waiting ${cooldownMs}ms before fallback to next model`);
+        await new Promise((r) => setTimeout(r, cooldownMs));
+      }
+
       break; // Move to next model
     }
   }
@@ -886,7 +893,20 @@ export async function handleComboChat({
     );
   }
 
-  const status = lastStatus || 406;
+  if (!lastStatus) {
+    return new Response(
+      JSON.stringify({
+        error: {
+          message: "Service temporarily unavailable: all upstream accounts are inactive",
+          type: "service_unavailable",
+          code: "ALL_ACCOUNTS_INACTIVE",
+        },
+      }),
+      { status: 503, headers: { "Content-Type": "application/json" } }
+    );
+  }
+
+  const status = lastStatus;
   const msg = lastError || "All combo models unavailable";
 
   if (earliestRetryAfter) {
@@ -941,7 +961,7 @@ async function handleRoundRobinCombo({
 
   const modelCount = orderedModels.length;
   if (modelCount === 0) {
-    return unavailableResponse(406, "Round-robin combo has no models");
+    return unavailableResponse(503, "Round-robin combo has no models");
   }
 
   // Get and increment atomic counter
@@ -1077,7 +1097,8 @@ async function handleRoundRobinCombo({
           errorText,
           0,
           null,
-          provider
+          provider,
+          result.headers
         );
 
         // Transient errors → mark in semaphore AND record circuit breaker failure
@@ -1106,6 +1127,12 @@ async function handleRoundRobinCombo({
         if (!lastStatus) lastStatus = result.status;
         if (offset > 0) fallbackCount++;
         log.warn("COMBO-RR", `${modelStr} failed, trying next model`, { status: result.status });
+
+        if ([502, 503, 504].includes(result.status) && cooldownMs > 0 && cooldownMs <= 5000) {
+          log.info("COMBO-RR", `Waiting ${cooldownMs}ms before fallback to next model`);
+          await new Promise((r) => setTimeout(r, cooldownMs));
+        }
+
         break;
       }
     } finally {
@@ -1136,7 +1163,20 @@ async function handleRoundRobinCombo({
     );
   }
 
-  const status = lastStatus || 406;
+  if (!lastStatus) {
+    return new Response(
+      JSON.stringify({
+        error: {
+          message: "Service temporarily unavailable: all upstream accounts are inactive",
+          type: "service_unavailable",
+          code: "ALL_ACCOUNTS_INACTIVE",
+        },
+      }),
+      { status: 503, headers: { "Content-Type": "application/json" } }
+    );
+  }
+
+  const status = lastStatus;
   const msg = lastError || "All round-robin combo models unavailable";
 
   if (earliestRetryAfter) {

open-sse/services/errorClassifier.ts

@@ -0,0 +1,53 @@
+import { isAccountDeactivated, isCreditsExhausted } from "./accountFallback.ts";
+
+export const PROVIDER_ERROR_TYPES = {
+  RATE_LIMITED: "rate_limited", // 429 — transient, retry with backoff
+  UNAUTHORIZED: "unauthorized", // 401 — token expired, refresh
+  ACCOUNT_DEACTIVATED: "account_deactivated", // 401 + deactivation signal
+  FORBIDDEN: "forbidden", // 403 — account banned/revoked, disable node
+  SERVER_ERROR: "server_error", // 500/502/503 — retry limited
+  QUOTA_EXHAUSTED: "quota_exhausted", // 402/429/400 + billing signals
+};
+
+function responseBodyToString(responseBody: unknown): string {
+  if (typeof responseBody === "string") return responseBody;
+  if (responseBody !== null && typeof responseBody === "object") {
+    try {
+      return JSON.stringify(responseBody);
+    } catch {
+      return "";
+    }
+  }
+  return "";
+}
+
+export function classifyProviderError(statusCode: number, responseBody: unknown): string | null {
+  const bodyStr = responseBodyToString(responseBody);
+  const creditsExhausted = isCreditsExhausted(bodyStr);
+  const accountDeactivated = isAccountDeactivated(bodyStr);
+
+  // T10: credits exhausted is terminal and can appear as 400/402/429 depending on provider.
+  if (
+    creditsExhausted &&
+    (statusCode === 400 || statusCode === 402 || statusCode === 429 || statusCode === 403)
+  ) {
+    return PROVIDER_ERROR_TYPES.QUOTA_EXHAUSTED;
+  }
+
+  if (statusCode === 429) {
+    return PROVIDER_ERROR_TYPES.RATE_LIMITED;
+  }
+
+  // T06: only deactivation-like 401s should be treated as permanent account expiry.
+  if (statusCode === 401) {
+    return accountDeactivated
+      ? PROVIDER_ERROR_TYPES.ACCOUNT_DEACTIVATED
+      : PROVIDER_ERROR_TYPES.UNAUTHORIZED;
+  }
+
+  if (statusCode === 402) return PROVIDER_ERROR_TYPES.QUOTA_EXHAUSTED;
+  if (statusCode === 403) return PROVIDER_ERROR_TYPES.FORBIDDEN;
+  if (statusCode >= 500) return PROVIDER_ERROR_TYPES.SERVER_ERROR;
+
+  return null;
+}

open-sse/services/ipFilter.ts

@@ -4,6 +4,8 @@
  * IP-based access control with blacklist, whitelist, priority modes, and temporary bans.
  */
 
+import { isIP } from "node:net";
+
 // In-memory IP lists
 let _config = {
   enabled: false,
@@ -161,10 +163,10 @@ export function createIPFilterMiddleware() {
  */
 export function checkRequestIP(request) {
   const ip =
-    request.headers?.get?.("x-forwarded-for")?.split(",")[0].trim() ||
-    request.headers?.get?.("x-real-ip") ||
-    request.headers?.get?.("cf-connecting-ip") ||
-    request.ip ||
+    pickFirstValidIp(request.headers?.get?.("cf-connecting-ip")) ||
+    pickFirstValidIp(request.headers?.get?.("x-forwarded-for")) ||
+    pickFirstValidIp(request.headers?.get?.("x-real-ip")) ||
+    normalizeIP(request.ip || "") ||
     "unknown";
   return checkIP(ip);
 }
@@ -177,6 +179,18 @@ function normalizeIP(ip) {
   return ip.replace(/^::ffff:/, "").trim();
 }
 
+function pickFirstValidIp(rawValue) {
+  if (typeof rawValue !== "string" || rawValue.trim().length === 0) return null;
+  const candidates = rawValue.split(",");
+  for (const candidate of candidates) {
+    const normalized = normalizeIP(candidate);
+    if (normalized && isIP(normalized) !== 0) {
+      return normalized;
+    }
+  }
+  return null;
+}
+
 function matchesAny(ip, ipSet) {
   // Direct match
   if (ipSet.has(ip)) return true;
@@ -225,12 +239,13 @@ function matchesWildcard(ip, pattern) {
 }
 
 function extractClientIP(req) {
+  const headers = req.headers || {};
   return (
-    req.headers?.["x-forwarded-for"]?.split(",")[0].trim() ||
-    req.headers?.["x-real-ip"] ||
-    req.headers?.["cf-connecting-ip"] ||
-    req.socket?.remoteAddress ||
-    req.ip ||
+    pickFirstValidIp(headers["cf-connecting-ip"]) ||
+    pickFirstValidIp(headers["x-forwarded-for"]) ||
+    pickFirstValidIp(headers["x-real-ip"]) ||
+    pickFirstValidIp(req.socket?.remoteAddress) ||
+    pickFirstValidIp(req.ip) ||
     "unknown"
   );
 }

open-sse/services/modelDeprecation.ts

@@ -18,6 +18,8 @@ const BUILT_IN_ALIASES: Record<string, string> = {
   "gemini-1.5-flash": "gemini-2.5-flash",
   "gemini-1.0-pro": "gemini-2.5-pro",
   "gemini-2.0-flash": "gemini-2.5-flash",
+  "gemini-3-pro-high": "gemini-3.1-pro-high",
+  "gemini-3-pro-low": "gemini-3.1-pro-low",
 
   // Claude legacy → current
   "claude-3-opus-20240229": "claude-opus-4-20250514",

open-sse/services/modelFamilyFallback.ts

@@ -101,6 +101,7 @@ const MODEL_UNAVAILABLE_FRAGMENTS = [
   "does not support",
   "not enabled for",
   "access to model",
+  "improperly formed request", // Kiro 400 (model unavailable)
 ];
 
 /**

open-sse/services/rateLimitManager.ts

@@ -12,6 +12,7 @@ import Bottleneck from "bottleneck";
 import { parseRetryAfterFromBody, lockModel } from "./accountFallback.ts";
 import { getProviderCategory } from "../config/providerRegistry.ts";
 import { DEFAULT_API_LIMITS } from "../config/constants.ts";
+import { getCodexRateLimitKey } from "../executors/codex.ts";
 
 interface LearnedLimitEntry {
   provider: string;
@@ -195,8 +196,15 @@ export function isRateLimitEnabled(connectionId) {
 /**
  * Get or create a limiter for a given provider+connection combination
  */
+function getLimiterKey(provider, connectionId, model = null) {
+  if (provider === "codex" && model) {
+    return `${provider}:${getCodexRateLimitKey(connectionId, model)}`;
+  }
+  return `${provider}:${connectionId}`;
+}
+
 function getLimiter(provider, connectionId, model = null) {
-  const key = model ? `${provider}:${connectionId}:${model}` : `${provider}:${connectionId}`;
+  const key = getLimiterKey(provider, connectionId, model);
 
   if (!limiters.has(key)) {
     const limiter = new Bottleneck({
@@ -235,7 +243,7 @@ export async function withRateLimit(provider, connectionId, model, fn) {
     return fn();
   }
 
-  const limiter = getLimiter(provider, connectionId, null);
+  const limiter = getLimiter(provider, connectionId, model);
   return limiter.schedule(fn);
 }
 
@@ -320,7 +328,7 @@ export function updateFromHeaders(provider, connectionId, headers, status, model
   if (!enabledConnections.has(connectionId)) return;
   if (!headers) return;
 
-  const limiter = getLimiter(provider, connectionId, null);
+  const limiter = getLimiter(provider, connectionId, model);
   const headerMap =
     provider === "claude" || provider === "anthropic" ? ANTHROPIC_HEADERS : STANDARD_HEADERS;
 
@@ -340,7 +348,7 @@ export function updateFromHeaders(provider, connectionId, headers, status, model
   if (status === 429) {
     const retryAfterMs = parseResetTime(retryAfterStr) || 60000; // Default 60s
     const counts = limiter.counts();
-    const limiterKey = `${provider}:${connectionId}`;
+    const limiterKey = getLimiterKey(provider, connectionId, model);
     console.log(
       `🚫 [RATE-LIMIT] ${provider}:${connectionId.slice(0, 8)} — 429 received, pausing for ${Math.ceil(retryAfterMs / 1000)}s, dropping ${counts.QUEUED} queued request(s)`
     );
@@ -397,7 +405,12 @@ export function updateFromHeaders(provider, connectionId, headers, status, model
     limiter.updateSettings(updates);
 
     // Persist learned limits (debounced)
-    recordLearnedLimit(provider, connectionId, { limit, remaining, minTime: updates.minTime });
+    recordLearnedLimit(
+      provider,
+      connectionId,
+      { limit, remaining, minTime: updates.minTime },
+      model
+    );
   }
 }
 
@@ -459,9 +472,10 @@ export function getLearnedLimits() {
 function recordLearnedLimit(
   provider: string,
   connectionId: string,
-  limits: Partial<Omit<LearnedLimitEntry, "provider" | "connectionId" | "lastUpdated">>
+  limits: Partial<Omit<LearnedLimitEntry, "provider" | "connectionId" | "lastUpdated">>,
+  model: string | null = null
 ) {
-  const key = `${provider}:${connectionId}`;
+  const key = getLimiterKey(provider, connectionId, model);
   learnedLimits[key] = {
     ...limits,
     provider,

open-sse/services/sessionManager.ts

@@ -41,7 +41,13 @@ const SESSION_TTL_MS = 30 * 60 * 1000;
 const _cleanupTimer = setInterval(() => {
   const now = Date.now();
   for (const [key, entry] of sessions) {
-    if (now - entry.lastActive > SESSION_TTL_MS) sessions.delete(key);
+    if (now - entry.lastActive > SESSION_TTL_MS) {
+      sessions.delete(key);
+      for (const [apiKeyId, sessionSet] of activeSessionsByKey) {
+        sessionSet.delete(key);
+        if (sessionSet.size === 0) activeSessionsByKey.delete(apiKeyId);
+      }
+    }
   }
 }, 60_000);
 _cleanupTimer.unref();
@@ -173,6 +179,103 @@ export function getActiveSessions(): Array<SessionEntry & { sessionId: string; a
  */
 export function clearSessions(): void {
   sessions.clear();
+  activeSessionsByKey.clear();
+}
+
+// ─── T08: Per-API-Key Session Limit ─────────────────────────────────────────
+// Tracks concurrent sticky sessions per API key and enforces max_sessions limits.
+// Ref: sub2api PR #634 (fix: stabilize session hash + add user-level session limit)
+
+// Map: apiKeyId → Set<sessionId>
+const activeSessionsByKey = new Map<string, Set<string>>();
+
+/**
+ * T08: Get the number of currently active sessions for an API key.
+ * @param apiKeyId - The API key's UUID from the database
+ */
+export function getActiveSessionCountForKey(apiKeyId: string): number {
+  return activeSessionsByKey.get(apiKeyId)?.size ?? 0;
+}
+
+/**
+ * T08: Register a session as belonging to an API key.
+ * Call this after session creation is allowed (i.e., limit check passed).
+ */
+export function registerKeySession(apiKeyId: string, sessionId: string): void {
+  if (!activeSessionsByKey.has(apiKeyId)) {
+    activeSessionsByKey.set(apiKeyId, new Set());
+  }
+  activeSessionsByKey.get(apiKeyId)!.add(sessionId);
+}
+
+/**
+ * Check whether a given session is already registered for an API key.
+ */
+export function isSessionRegisteredForKey(apiKeyId: string, sessionId: string): boolean {
+  return activeSessionsByKey.get(apiKeyId)?.has(sessionId) === true;
+}
+
+/**
+ * T08: Unregister a session from an API key's active set.
+ * Call this when the request closes or the session TTL expires.
+ */
+export function unregisterKeySession(apiKeyId: string, sessionId: string): void {
+  activeSessionsByKey.get(apiKeyId)?.delete(sessionId);
+  // Clean up empty sets to avoid memory leaks
+  if (activeSessionsByKey.get(apiKeyId)?.size === 0) {
+    activeSessionsByKey.delete(apiKeyId);
+  }
+}
+
+/**
+ * T08: Check whether adding a new session would exceed the key's max_sessions limit.
+ * Returns null if allowed, or an error object to return as a 429 response.
+ *
+ * @param apiKeyId - The API key's UUID
+ * @param maxSessions - The limit from the DB (0 = unlimited)
+ */
+export function checkSessionLimit(
+  apiKeyId: string,
+  maxSessions: number
+): { code: "SESSION_LIMIT_EXCEEDED"; message: string; limit: number; current: number } | null {
+  if (!maxSessions || maxSessions <= 0) return null; // unlimited
+  const current = getActiveSessionCountForKey(apiKeyId);
+  if (current < maxSessions) return null;
+  return {
+    code: "SESSION_LIMIT_EXCEEDED",
+    message:
+      `You have reached the maximum number of active sessions (${maxSessions}). ` +
+      `Please close unused sessions or wait for them to expire.`,
+    limit: maxSessions,
+    current,
+  };
+}
+
+/**
+ * T04: Extract an external session ID from request headers.
+ * Accepts both hyphenated and underscore forms for Nginx compatibility.
+ * Nginx drops headers with underscores by default — use `underscores_in_headers on`
+ * in nginx.conf, or use X-Session-Id (hyphenated) which passes cleanly.
+ *
+ * Ref: sub2api README + PR #634
+ *
+ * @param headers - Request headers (Headers object or plain object with .get())
+ * @returns External session ID with "ext:" prefix, or null
+ */
+export function extractExternalSessionId(
+  headers: Headers | { get?: (n: string) => string | null } | null | undefined
+): string | null {
+  if (!headers || typeof (headers as Headers).get !== "function") return null;
+  const h = headers as Headers;
+  const raw =
+    h.get("x-session-id") ?? // Preferred: hyphenated (passes through Nginx)
+    h.get("x_session_id") ?? // Underscore variant (direct HTTP / custom clients)
+    h.get("x-omniroute-session") ?? // OmniRoute-specific form
+    h.get("session-id") ?? // Bare session-id
+    null;
+  if (!raw || !raw.trim()) return null;
+  // Prefix "ext:" to ensure no collision with internal SHA-256 hash IDs
+  return `ext:${raw.trim().slice(0, 64)}`; // max 64 chars to avoid abuse
 }
 
 // ─── Internal Helpers ───────────────────────────────────────────────────────

open-sse/services/thinkingBudget.ts

@@ -13,12 +13,16 @@ export const ThinkingMode = {
   ADAPTIVE: "adaptive", // Scale based on request complexity
 };
 
+import { capThinkingBudget, getDefaultThinkingBudget } from "@/shared/constants/modelSpecs";
+
 // Effort → budget token mapping
 export const EFFORT_BUDGETS = {
   none: 0,
   low: 1024,
   medium: 10240,
-  high: 131072,
+  high: 131072, // Handled globally by capThinkingBudget later
+  max: 131072, // T11: Claude "max" / "xhigh" — full budget
+  xhigh: 131072, // T11: explicit alias used internally
 };
 
 // thinkingLevel string → budget token mapping
@@ -28,6 +32,8 @@ export const THINKING_LEVEL_MAP = {
   low: 1024,
   medium: 10240,
   high: 131072,
+  max: 131072, // T11: max = full Claude budget (sub2api: xhigh)
+  xhigh: 131072, // T11: explicit xhigh alias
 };
 
 // Default config (passthrough = backward compatible)
@@ -68,8 +74,8 @@ export function normalizeThinkingLevel(body) {
 
   // Handle top-level thinkingLevel or thinking_level string fields
   const levelStr = result.thinkingLevel || result.thinking_level;
-  if (typeof levelStr === "string" && THINKING_LEVEL_MAP[levelStr] !== undefined) {
-    const budget = THINKING_LEVEL_MAP[levelStr];
+  if (typeof levelStr === "string" && THINKING_LEVEL_MAP[levelStr.toLowerCase()] !== undefined) {
+    const budget = THINKING_LEVEL_MAP[levelStr.toLowerCase()];
     // Convert to Claude thinking format as canonical representation
     result.thinking = {
       type: budget > 0 ? "enabled" : "disabled",
@@ -83,15 +89,21 @@ export function normalizeThinkingLevel(body) {
   const geminiLevel =
     result.generationConfig?.thinkingConfig?.thinkingLevel ||
     result.generationConfig?.thinking_config?.thinkingLevel;
-  if (typeof geminiLevel === "string" && THINKING_LEVEL_MAP[geminiLevel] !== undefined) {
-    const budget = THINKING_LEVEL_MAP[geminiLevel];
+  if (
+    typeof geminiLevel === "string" &&
+    THINKING_LEVEL_MAP[geminiLevel.toLowerCase()] !== undefined
+  ) {
+    const budget = THINKING_LEVEL_MAP[geminiLevel.toLowerCase()];
     result.generationConfig = {
       ...result.generationConfig,
-      thinking_config: { thinking_budget: budget },
+      thinkingConfig: { ...result.generationConfig.thinkingConfig, thinkingBudget: budget },
     };
-    // Clean up camelCase variant if it was the source
+    // Clean up string variants
     if (result.generationConfig.thinkingConfig) {
-      delete result.generationConfig.thinkingConfig;
+      delete result.generationConfig.thinkingConfig.thinkingLevel;
+    }
+    if (result.generationConfig.thinking_config) {
+      delete result.generationConfig.thinking_config;
     }
   }
 
@@ -118,7 +130,7 @@ export function ensureThinkingConfig(body) {
   const result = { ...body };
   result.thinking = {
     type: "enabled",
-    budget_tokens: EFFORT_BUDGETS.medium, // 10240 default
+    budget_tokens: getDefaultThinkingBudget(model) || EFFORT_BUDGETS.medium,
   };
   return result;
 }
@@ -198,7 +210,7 @@ function setCustomBudget(body, budget) {
     };
   }
 
-  // OpenAI reasoning_effort mapping
+  // OpenAI reasoning_effort mapping (T11: add 'max' tier for full budget)
   if (result.reasoning_effort !== undefined || result.reasoning !== undefined) {
     if (budget <= 0) {
       delete result.reasoning_effort;
@@ -207,8 +219,10 @@ function setCustomBudget(body, budget) {
       result.reasoning_effort = "low";
     } else if (budget <= 10240) {
       result.reasoning_effort = "medium";
-    } else {
+    } else if (budget < 131072) {
       result.reasoning_effort = "high";
+    } else {
+      result.reasoning_effort = "max"; // T11: full budget → "max"
     }
   }
 
@@ -251,8 +265,11 @@ function applyAdaptiveBudget(body, cfg) {
   if (toolCount > 3) multiplier += 0.5;
   if (lastMsgLength > 2000) multiplier += 0.3;
 
-  const baseBudget = EFFORT_BUDGETS[cfg.effortLevel] || EFFORT_BUDGETS.medium;
-  const budget = Math.min(Math.ceil(baseBudget * multiplier), 131072);
+  const baseBudget =
+    EFFORT_BUDGETS[cfg.effortLevel] ||
+    getDefaultThinkingBudget(body.model || "") ||
+    EFFORT_BUDGETS.medium;
+  const budget = capThinkingBudget(body.model || "", Math.ceil(baseBudget * multiplier));
 
   return setCustomBudget(body, budget);
 }

open-sse/transformer/responsesTransformer.ts

@@ -1,5 +1,5 @@
-import * as fs from 'fs';
-import * as path from 'path';
+import * as fs from "fs";
+import * as path from "path";
 /**
  * Responses API Transformer
  * Converts OpenAI Chat Completions SSE to Codex Responses API SSE format
@@ -402,6 +402,16 @@ export function createResponsesApiTransformStream(logger = null) {
             const newCallId = tc.id;
             const funcName = tc.function?.name;
 
+            // T37: Prevent merging if a new tool_call uses the same index
+            if (state.funcCallIds[tcIdx] && newCallId && state.funcCallIds[tcIdx] !== newCallId) {
+              closeToolCall(controller, tcIdx);
+              delete state.funcCallIds[tcIdx];
+              delete state.funcNames[tcIdx];
+              delete state.funcArgsBuf[tcIdx];
+              delete state.funcArgsDone[tcIdx];
+              delete state.funcItemDone[tcIdx];
+            }
+
             if (funcName) state.funcNames[tcIdx] = funcName;
 
             if (!state.funcCallIds[tcIdx] && newCallId) {

open-sse/translator/helpers/geminiHelper.ts

@@ -172,6 +172,9 @@ function convertEnumValuesToStrings(obj) {
 
   if (obj.enum && Array.isArray(obj.enum)) {
     obj.enum = obj.enum.map((v) => String(v));
+    if (!obj.type) {
+      obj.type = "string";
+    }
   }
 
   for (const value of Object.values(obj)) {

open-sse/translator/image/sizeMapper.ts

@@ -0,0 +1,22 @@
+const OPENAI_SIZE_TO_ASPECT_RATIO: Record<string, string> = {
+  "256x256": "1:1",
+  "512x512": "1:1",
+  "1024x1024": "1:1",
+  "1792x1024": "16:9",
+  "1024x1792": "9:16",
+  "1536x1024": "3:2",
+  "1024x1536": "2:3",
+};
+
+// Supports direct aspect ratios (e.g. "16:9")
+const ASPECT_RATIO_PASSTHROUGH = /^\d+:\d+$/;
+
+export function mapImageSize(sizeParam?: string | null): string {
+  if (!sizeParam) return "1:1"; // default
+
+  // Native aspect ratio (e.g. "16:9") — pass-through
+  if (ASPECT_RATIO_PASSTHROUGH.test(sizeParam)) return sizeParam;
+
+  // Map OpenAI sizes to aspect ratios
+  return OPENAI_SIZE_TO_ASPECT_RATIO[sizeParam] ?? "1:1";
+}

open-sse/translator/request/claude-to-gemini.ts

@@ -1,6 +1,10 @@
 import { register } from "../registry.ts";
 import { FORMATS } from "../formats.ts";
-import { DEFAULT_SAFETY_SETTINGS, tryParseJSON } from "../helpers/geminiHelper.ts";
+import {
+  DEFAULT_SAFETY_SETTINGS,
+  tryParseJSON,
+  cleanJSONSchemaForAntigravity,
+} from "../helpers/geminiHelper.ts";
 import { DEFAULT_THINKING_GEMINI_SIGNATURE } from "../../config/defaultThinkingSignature.ts";
 
 /**
@@ -154,7 +158,9 @@ export function claudeToGeminiRequest(model, body, stream) {
         functionDeclarations.push({
           name: tool.name,
           description: tool.description || "",
-          parameters: tool.input_schema || { type: "object", properties: {} },
+          parameters: cleanJSONSchemaForAntigravity(
+            tool.input_schema || { type: "object", properties: {} }
+          ),
         });
       }
     }

open-sse/translator/request/openai-to-claude.ts

@@ -27,6 +27,60 @@ type ClaudeTool = {
   defer_loading?: boolean;
 };
 
+/**
+ * T02: Recursively strips empty text blocks from content arrays.
+ * Anthropic returns 400 "text content blocks must be non-empty" if any
+ * text block has text: "". Must also recurse into nested tool_result.content.
+ * Ref: sub2api PR #1212
+ */
+export function stripEmptyTextBlocks(content: unknown[] | undefined): unknown[] {
+  if (!Array.isArray(content)) return content ?? [];
+  return content
+    .filter((block: unknown) => {
+      if (
+        block &&
+        typeof block === "object" &&
+        (block as Record<string, unknown>).type === "text"
+      ) {
+        const text = (block as Record<string, unknown>).text;
+        if (text === "" || text == null) return false;
+      }
+      return true;
+    })
+    .map((block: unknown) => {
+      if (
+        block &&
+        typeof block === "object" &&
+        (block as Record<string, unknown>).type === "tool_result" &&
+        Array.isArray((block as Record<string, unknown>).content)
+      ) {
+        // Recurse into nested tool_result.content
+        return {
+          ...(block as Record<string, unknown>),
+          content: stripEmptyTextBlocks((block as Record<string, unknown>).content as unknown[]),
+        };
+      }
+      return block;
+    });
+}
+
+/**
+ * T15: Normalize content to string form.
+ * Handles both string and array-of-blocks forms (Cursor, Codex 2.x, etc.).
+ * Ref: sub2api PR #1197
+ */
+export function normalizeContentToString(content: string | unknown[] | null | undefined): string {
+  if (!content) return "";
+  if (typeof content === "string") return content;
+  if (Array.isArray(content)) {
+    return (content as Array<Record<string, unknown>>)
+      .filter((b) => b.type === "text")
+      .map((b) => String(b.text ?? ""))
+      .join("\n");
+  }
+  return "";
+}
+
 // Convert OpenAI request to Claude format
 export function openaiToClaudeRequest(model, body, stream) {
   // Check if tool prefix should be disabled (configured per-provider or global)
@@ -61,11 +115,11 @@ export function openaiToClaudeRequest(model, body, stream) {
   const systemParts = [];
 
   if (body.messages && Array.isArray(body.messages)) {
-    // Extract system messages
+    // Extract system messages (T15: handle both string and array content)
     for (const msg of body.messages) {
       if (msg.role === "system") {
         systemParts.push(
-          typeof msg.content === "string" ? msg.content : extractTextContent(msg.content)
+          typeof msg.content === "string" ? msg.content : normalizeContentToString(msg.content)
         );
       }
     }
@@ -270,10 +324,14 @@ function getContentBlocksFromMessage(msg, toolNameMap = new Map(), disableToolPr
   const blocks = [];
 
   if (msg.role === "tool") {
+    // T02: Strip empty text blocks from nested tool_result content to avoid Anthropic 400
+    const toolContent = Array.isArray(msg.content)
+      ? stripEmptyTextBlocks(msg.content)
+      : msg.content;
     blocks.push({
       type: "tool_result",
       tool_use_id: msg.tool_call_id,
-      content: msg.content,
+      content: toolContent,
     });
   } else if (msg.role === "user") {
     if (typeof msg.content === "string") {
@@ -287,10 +345,14 @@ function getContentBlocksFromMessage(msg, toolNameMap = new Map(), disableToolPr
         } else if (part.type === "tool_result") {
           // Skip tool_result with no tool_use_id (would be useless and may cause errors)
           if (!part.tool_use_id) continue;
+          // T02: strip empty text blocks from nested content before passing to Anthropic
+          const resultContent = Array.isArray(part.content)
+            ? stripEmptyTextBlocks(part.content)
+            : part.content;
           blocks.push({
             type: "tool_result",
             tool_use_id: part.tool_use_id,
-            content: part.content,
+            content: resultContent,
             ...(part.is_error && { is_error: part.is_error }),
           });
         } else if (part.type === "image_url") {

open-sse/translator/request/openai-to-gemini.ts

@@ -3,6 +3,11 @@ import { FORMATS } from "../formats.ts";
 import { DEFAULT_THINKING_GEMINI_SIGNATURE } from "../../config/defaultThinkingSignature.ts";
 import { ANTIGRAVITY_DEFAULT_SYSTEM } from "../../config/constants.ts";
 import { openaiToClaudeRequestForAntigravity } from "./openai-to-claude.ts";
+import {
+  capMaxOutputTokens,
+  capThinkingBudget,
+  getDefaultThinkingBudget,
+} from "../../../src/shared/constants/modelSpecs.ts";
 
 function generateUUID() {
   return crypto.randomUUID();
@@ -88,7 +93,9 @@ function openaiToGeminiBase(model, body, stream) {
     result.generationConfig.topK = body.top_k;
   }
   if (body.max_tokens !== undefined) {
-    result.generationConfig.maxOutputTokens = body.max_tokens;
+    result.generationConfig.maxOutputTokens = capMaxOutputTokens(model, body.max_tokens);
+  } else {
+    result.generationConfig.maxOutputTokens = capMaxOutputTokens(model);
   }
 
   // Build tool_call_id -> name map
@@ -283,8 +290,12 @@ export function openaiToGeminiCLIRequest(model, body, stream) {
 
   // Add thinking config for CLI
   if (body.reasoning_effort) {
-    const budgetMap = { low: 1024, medium: 8192, high: 32768 };
-    const budget = budgetMap[body.reasoning_effort] || 8192;
+    const budgetMap = {
+      low: 1024,
+      medium: getDefaultThinkingBudget(model) || 8192,
+      high: capThinkingBudget(model, 32768),
+    };
+    const budget = budgetMap[body.reasoning_effort] || getDefaultThinkingBudget(model) || 8192;
     gemini.generationConfig.thinkingConfig = {
       thinkingBudget: budget,
       include_thoughts: true,

open-sse/translator/response/openai-responses.ts

@@ -257,6 +257,17 @@ function emitToolCall(state, emit, tc) {
   const newCallId = tc.id;
   const funcName = tc.function?.name;
 
+  // T37: If we already have a tool call at this index but the ID changed,
+  // we must close the current one and start a new one to prevent merging.
+  if (state.funcCallIds[tcIdx] && newCallId && state.funcCallIds[tcIdx] !== newCallId) {
+    closeToolCall(state, emit, tcIdx);
+    delete state.funcCallIds[tcIdx];
+    delete state.funcNames[tcIdx];
+    delete state.funcArgsBuf[tcIdx];
+    delete state.funcArgsDone[tcIdx];
+    delete state.funcItemDone[tcIdx];
+  }
+
   if (funcName) state.funcNames[tcIdx] = funcName;
 
   if (!state.funcCallIds[tcIdx] && newCallId) {

open-sse/utils/ollamaTransform.ts

@@ -12,6 +12,7 @@ type PendingToolCall = {
 export function transformToOllama(response, model) {
   let buffer = "";
   let pendingToolCalls: Record<number, PendingToolCall> = {};
+  const completedToolCalls: PendingToolCall[] = [];
 
   const transform = new TransformStream({
     transform(chunk, controller) {
@@ -41,6 +42,13 @@ export function transformToOllama(response, model) {
           if (toolCalls) {
             for (const tc of toolCalls) {
               const idx = tc.index;
+
+              // T37: Prevent merging tool_calls on same index if ID changes
+              if (pendingToolCalls[idx] && tc.id && pendingToolCalls[idx].id !== tc.id) {
+                completedToolCalls.push(pendingToolCalls[idx]);
+                delete pendingToolCalls[idx];
+              }
+
               if (!pendingToolCalls[idx]) {
                 pendingToolCalls[idx] = { id: tc.id, function: { name: "", arguments: "" } };
               }
@@ -59,7 +67,7 @@ export function transformToOllama(response, model) {
 
           const finishReason = parsed.choices?.[0]?.finish_reason;
           if (finishReason === "tool_calls" || finishReason === "stop") {
-            const toolCallsArr = Object.values(pendingToolCalls);
+            const toolCallsArr = [...completedToolCalls, ...Object.values(pendingToolCalls)];
             if (toolCallsArr.length > 0) {
               const formattedCalls = toolCallsArr.map((tc) => ({
                 function: {

open-sse/utils/stream.ts

@@ -222,16 +222,17 @@ export function createSSEStream(options: StreamOptions = {}) {
                   const extracted = extractUsage(parsed);
                   if (extracted) {
                     // Non-destructive merge: never overwrite a positive value with 0
-                    // message_start carries input_tokens, message_delta carries output_tokens
-                    if (!usage) usage = {};
-                    if (extracted.prompt_tokens > 0) usage.prompt_tokens = extracted.prompt_tokens;
-                    if (extracted.completion_tokens > 0)
-                      usage.completion_tokens = extracted.completion_tokens;
-                    if (extracted.total_tokens > 0) usage.total_tokens = extracted.total_tokens;
-                    if (extracted.cache_read_input_tokens)
-                      usage.cache_read_input_tokens = extracted.cache_read_input_tokens;
-                    if (extracted.cache_creation_input_tokens)
-                      usage.cache_creation_input_tokens = extracted.cache_creation_input_tokens;
+                    // message_start carries input_tokens, message_delta carries output_tokens;
+                    if (!usage) usage = {} as any;
+                    const u = usage as Record<string, number>;
+                    const eu = extracted as Record<string, number>;
+                    if (eu.prompt_tokens > 0) u.prompt_tokens = eu.prompt_tokens;
+                    if (eu.completion_tokens > 0) u.completion_tokens = eu.completion_tokens;
+                    if (eu.total_tokens > 0) u.total_tokens = eu.total_tokens;
+                    if (eu.cache_read_input_tokens)
+                      u.cache_read_input_tokens = eu.cache_read_input_tokens;
+                    if (eu.cache_creation_input_tokens)
+                      u.cache_creation_input_tokens = eu.cache_creation_input_tokens;
                   }
                   // Track content length and accumulate from Claude format
                   if (parsed.delta?.text) {
@@ -263,6 +264,11 @@ export function createSSEStream(options: StreamOptions = {}) {
                     }
                   }
 
+                  // T18: Track if we saw tool calls
+                  if (delta?.tool_calls && delta.tool_calls.length > 0) {
+                    (state as any).passthroughHasToolCalls = true;
+                  }
+
                   const content = delta?.content || delta?.reasoning_content;
                   if (content && typeof content === "string") {
                     totalContentLength += content.length;
@@ -278,6 +284,20 @@ export function createSSEStream(options: StreamOptions = {}) {
                   }
 
                   const isFinishChunk = parsed.choices?.[0]?.finish_reason;
+
+                  // T18: Normalize finish_reason to 'tool_calls' if tool calls were used
+                  if (
+                    isFinishChunk &&
+                    (state as any).passthroughHasToolCalls &&
+                    parsed.choices[0].finish_reason !== "tool_calls"
+                  ) {
+                    parsed.choices[0].finish_reason = "tool_calls";
+                    // If we modify it, we must output the modified object
+                    if (!injectedUsage && hasValidUsage(parsed.usage)) {
+                      output = `data: ${JSON.stringify(parsed)}\n`;
+                      injectedUsage = true;
+                    }
+                  }
                   if (isFinishChunk && !hasValidUsage(parsed.usage)) {
                     const estimated = estimateUsage(body, totalContentLength, FORMATS.OPENAI);
                     parsed.usage = filterUsageForFormat(estimated, FORMATS.OPENAI);
@@ -529,19 +549,17 @@ export function createSSEStream(options: StreamOptions = {}) {
                 if (!state.usage) {
                   state.usage = extracted;
                 } else {
-                  if (extracted.prompt_tokens > 0)
-                    state.usage.prompt_tokens = extracted.prompt_tokens;
-                  if (extracted.completion_tokens > 0)
-                    state.usage.completion_tokens = extracted.completion_tokens;
-                  if (extracted.total_tokens > 0) state.usage.total_tokens = extracted.total_tokens;
-                  if (extracted.cache_read_input_tokens > 0)
-                    state.usage.cache_read_input_tokens = extracted.cache_read_input_tokens;
-                  if (extracted.cache_creation_input_tokens > 0)
-                    state.usage.cache_creation_input_tokens = extracted.cache_creation_input_tokens;
-                  if (extracted.cached_tokens > 0)
-                    state.usage.cached_tokens = extracted.cached_tokens;
-                  if (extracted.reasoning_tokens > 0)
-                    state.usage.reasoning_tokens = extracted.reasoning_tokens;
+                  const su = state.usage as Record<string, number>;
+                  const eu = extracted as Record<string, number>;
+                  if (eu.prompt_tokens > 0) su.prompt_tokens = eu.prompt_tokens;
+                  if (eu.completion_tokens > 0) su.completion_tokens = eu.completion_tokens;
+                  if (eu.total_tokens > 0) su.total_tokens = eu.total_tokens;
+                  if (eu.cache_read_input_tokens > 0)
+                    su.cache_read_input_tokens = eu.cache_read_input_tokens;
+                  if (eu.cache_creation_input_tokens > 0)
+                    su.cache_creation_input_tokens = eu.cache_creation_input_tokens;
+                  if (eu.cached_tokens > 0) su.cached_tokens = eu.cached_tokens;
+                  if (eu.reasoning_tokens > 0) su.reasoning_tokens = eu.reasoning_tokens;
                 }
               }
 

open-sse/utils/streamHandler.ts

@@ -134,7 +134,36 @@ export function createDisconnectAwareStream(transformStream, streamController) {
         controller.enqueue(value);
       } catch (error) {
         streamController.handleError(error);
-        controller.error(error);
+
+        // T35: Encapsulate mid-stream errors as SSE events instead of abruptly aborting
+        // This prevents TransferEncodingError on the client side
+        const errorMsg = error instanceof Error ? error.message : "Upstream stream error";
+        const statusCode =
+          typeof error === "object" && error !== null && "statusCode" in error
+            ? (error as any).statusCode
+            : 500;
+
+        const errorEvent = {
+          object: "chat.completion.chunk",
+          choices: [
+            {
+              index: 0,
+              delta: {},
+              finish_reason: "error",
+            },
+          ],
+          error: {
+            message: errorMsg,
+            type: "upstream_error",
+            code: statusCode,
+          },
+        };
+
+        const encoder = new TextEncoder();
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify(errorEvent)}\n\n`));
+        controller.enqueue(encoder.encode(`data: [DONE]\n\n`));
+
+        controller.close();
       }
     },
 

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "omniroute",
-  "version": "3.0.0-rc.5",
+  "version": "3.0.0-rc.10",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omniroute",
-      "version": "3.0.0-rc.5",
+      "version": "3.0.0-rc.10",
       "hasInstallScript": true,
       "license": "MIT",
       "workspaces": [
@@ -28,7 +28,7 @@
         "jose": "^6.1.3",
         "lowdb": "^7.0.1",
         "monaco-editor": "^0.55.1",
-        "next": "^16.1.6",
+        "next": "^16.0.10",
         "next-intl": "^4.8.3",
         "node-machine-id": "^1.1.12",
         "open": "^11.0.0",
@@ -61,7 +61,7 @@
         "concurrently": "^9.2.1",
         "cross-env": "^10.1.0",
         "eslint": "^9.39.2",
-        "eslint-config-next": "16.1.6",
+        "eslint-config-next": "^16.0.10",
         "husky": "^9.1.7",
         "lint-staged": "^16.2.7",
         "prettier": "^3.8.1",
@@ -2567,15 +2567,15 @@
       }
     },
     "node_modules/@next/env": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/env/-/env-16.1.7.tgz",
-      "integrity": "sha512-rJJbIdJB/RQr2F1nylZr/PJzamvNNhfr3brdKP6s/GW850jbtR70QlSfFselvIBbcPUOlQwBakexjFzqLzF6pg==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/env/-/env-16.0.10.tgz",
+      "integrity": "sha512-8tuaQkyDVgeONQ1MeT9Mkk8pQmZapMKFh5B+OrFUlG3rVmYTXcXlBetBgTurKXGaIZvkoqRT9JL5K3phXcgang==",
       "license": "MIT"
     },
     "node_modules/@next/eslint-plugin-next": {
-      "version": "16.1.6",
-      "resolved": "https://registry.npmjs.org/@next/eslint-plugin-next/-/eslint-plugin-next-16.1.6.tgz",
-      "integrity": "sha512-/Qq3PTagA6+nYVfryAtQ7/9FEr/6YVyvOtl6rZnGsbReGLf0jZU6gkpr1FuChAQpvV46a78p4cmHOVP8mbfSMQ==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/eslint-plugin-next/-/eslint-plugin-next-16.0.10.tgz",
+      "integrity": "sha512-b2NlWN70bbPLmfyoLvvidPKWENBYYIe017ZGUpElvQjDytCWgxPJx7L9juxHt0xHvNVA08ZHJdOyhGzon/KJuw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -2583,9 +2583,9 @@
       }
     },
     "node_modules/@next/swc-darwin-arm64": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-darwin-arm64/-/swc-darwin-arm64-16.1.7.tgz",
-      "integrity": "sha512-b2wWIE8sABdyafc4IM8r5Y/dS6kD80JRtOGrUiKTsACFQfWWgUQ2NwoUX1yjFMXVsAwcQeNpnucF2ZrujsBBPg==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-darwin-arm64/-/swc-darwin-arm64-16.0.10.tgz",
+      "integrity": "sha512-4XgdKtdVsaflErz+B5XeG0T5PeXKDdruDf3CRpnhN+8UebNa5N2H58+3GDgpn/9GBurrQ1uWW768FfscwYkJRg==",
       "cpu": [
         "arm64"
       ],
@@ -2599,9 +2599,9 @@
       }
     },
     "node_modules/@next/swc-darwin-x64": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-darwin-x64/-/swc-darwin-x64-16.1.7.tgz",
-      "integrity": "sha512-zcnVaaZulS1WL0Ss38R5Q6D2gz7MtBu8GZLPfK+73D/hp4GFMrC2sudLky1QibfV7h6RJBJs/gOFvYP0X7UVlQ==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-darwin-x64/-/swc-darwin-x64-16.0.10.tgz",
+      "integrity": "sha512-spbEObMvRKkQ3CkYVOME+ocPDFo5UqHb8EMTS78/0mQ+O1nqE8toHJVioZo4TvebATxgA8XMTHHrScPrn68OGw==",
       "cpu": [
         "x64"
       ],
@@ -2615,12 +2615,15 @@
       }
     },
     "node_modules/@next/swc-linux-arm64-gnu": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-gnu/-/swc-linux-arm64-gnu-16.1.7.tgz",
-      "integrity": "sha512-2ant89Lux/Q3VyC8vNVg7uBaFVP9SwoK2jJOOR0L8TQnX8CAYnh4uctAScy2Hwj2dgjVHqHLORQZJ2wH6VxhSQ==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-gnu/-/swc-linux-arm64-gnu-16.0.10.tgz",
+      "integrity": "sha512-uQtWE3X0iGB8apTIskOMi2w/MKONrPOUCi5yLO+v3O8Mb5c7K4Q5KD1jvTpTF5gJKa3VH/ijKjKUq9O9UhwOYw==",
       "cpu": [
         "arm64"
       ],
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -2631,12 +2634,15 @@
       }
     },
     "node_modules/@next/swc-linux-arm64-musl": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-musl/-/swc-linux-arm64-musl-16.1.7.tgz",
-      "integrity": "sha512-uufcze7LYv0FQg9GnNeZ3/whYfo+1Q3HnQpm16o6Uyi0OVzLlk2ZWoY7j07KADZFY8qwDbsmFnMQP3p3+Ftprw==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-arm64-musl/-/swc-linux-arm64-musl-16.0.10.tgz",
+      "integrity": "sha512-llA+hiDTrYvyWI21Z0L1GiXwjQaanPVQQwru5peOgtooeJ8qx3tlqRV2P7uH2pKQaUfHxI/WVarvI5oYgGxaTw==",
       "cpu": [
         "arm64"
       ],
+      "libc": [
+        "musl"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -2647,12 +2653,15 @@
       }
     },
     "node_modules/@next/swc-linux-x64-gnu": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-gnu/-/swc-linux-x64-gnu-16.1.7.tgz",
-      "integrity": "sha512-KWVf2gxYvHtvuT+c4MBOGxuse5TD7DsMFYSxVxRBnOzok/xryNeQSjXgxSv9QpIVlaGzEn/pIuI6Koosx8CGWA==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-gnu/-/swc-linux-x64-gnu-16.0.10.tgz",
+      "integrity": "sha512-AK2q5H0+a9nsXbeZ3FZdMtbtu9jxW4R/NgzZ6+lrTm3d6Zb7jYrWcgjcpM1k8uuqlSy4xIyPR2YiuUr+wXsavA==",
       "cpu": [
         "x64"
       ],
+      "libc": [
+        "glibc"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -2663,12 +2672,15 @@
       }
     },
     "node_modules/@next/swc-linux-x64-musl": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-musl/-/swc-linux-x64-musl-16.1.7.tgz",
-      "integrity": "sha512-HguhaGwsGr1YAGs68uRKc4aGWxLET+NevJskOcCAwXbwj0fYX0RgZW2gsOCzr9S11CSQPIkxmoSbuVaBp4Z3dA==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-linux-x64-musl/-/swc-linux-x64-musl-16.0.10.tgz",
+      "integrity": "sha512-1TDG9PDKivNw5550S111gsO4RGennLVl9cipPhtkXIFVwo31YZ73nEbLjNC8qG3SgTz/QZyYyaFYMeY4BKZR/g==",
       "cpu": [
         "x64"
       ],
+      "libc": [
+        "musl"
+      ],
       "license": "MIT",
       "optional": true,
       "os": [
@@ -2679,9 +2691,9 @@
       }
     },
     "node_modules/@next/swc-win32-arm64-msvc": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-win32-arm64-msvc/-/swc-win32-arm64-msvc-16.1.7.tgz",
-      "integrity": "sha512-S0n3KrDJokKTeFyM/vGGGR8+pCmXYrjNTk2ZozOL1C/JFdfUIL9O1ATaJOl5r2POe56iRChbsszrjMAdWSv7kQ==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-win32-arm64-msvc/-/swc-win32-arm64-msvc-16.0.10.tgz",
+      "integrity": "sha512-aEZIS4Hh32xdJQbHz121pyuVZniSNoqDVx1yIr2hy+ZwJGipeqnMZBJHyMxv2tiuAXGx6/xpTcQJ6btIiBjgmg==",
       "cpu": [
         "arm64"
       ],
@@ -2695,9 +2707,9 @@
       }
     },
     "node_modules/@next/swc-win32-x64-msvc": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/@next/swc-win32-x64-msvc/-/swc-win32-x64-msvc-16.1.7.tgz",
-      "integrity": "sha512-mwgtg8CNZGYm06LeEd+bNnOUfwOyNem/rOiP14Lsz+AnUY92Zq/LXwtebtUiaeVkhbroRCQ0c8GlR4UT1U+0yg==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/@next/swc-win32-x64-msvc/-/swc-win32-x64-msvc-16.0.10.tgz",
+      "integrity": "sha512-E+njfCoFLb01RAFEnGZn6ERoOqhK1Gl3Lfz1Kjnj0Ulfu7oJbuMyvBKNj/bw8XZnenHDASlygTjZICQW+rYW1Q==",
       "cpu": [
         "x64"
       ],
@@ -7522,6 +7534,7 @@
       "version": "2.9.19",
       "resolved": "https://registry.npmjs.org/baseline-browser-mapping/-/baseline-browser-mapping-2.9.19.tgz",
       "integrity": "sha512-ipDqC8FrAl/76p2SSWKSI+H9tFwm7vYqXQrItCuiVPt26Km0jS+NzSsBWAaBusvSbQcfJG+JitdMm+wZAgTYqg==",
+      "dev": true,
       "license": "Apache-2.0",
       "bin": {
         "baseline-browser-mapping": "dist/cli.js"
@@ -9705,13 +9718,13 @@
       }
     },
     "node_modules/eslint-config-next": {
-      "version": "16.1.6",
-      "resolved": "https://registry.npmjs.org/eslint-config-next/-/eslint-config-next-16.1.6.tgz",
-      "integrity": "sha512-vKq40io2B0XtkkNDYyleATwblNt8xuh3FWp8SpSz3pt7P01OkBFlKsJZ2mWt5WsCySlDQLckb1zMY9yE9Qy0LA==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/eslint-config-next/-/eslint-config-next-16.0.10.tgz",
+      "integrity": "sha512-BxouZUm0I45K4yjOOIzj24nTi0H2cGo0y7xUmk+Po/PYtJXFBYVDS1BguE7t28efXjKdcN0tmiLivxQy//SsZg==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
-        "@next/eslint-plugin-next": "16.1.6",
+        "@next/eslint-plugin-next": "16.0.10",
         "eslint-import-resolver-node": "^0.3.6",
         "eslint-import-resolver-typescript": "^3.5.2",
         "eslint-plugin-import": "^2.32.0",
@@ -14658,14 +14671,13 @@
       }
     },
     "node_modules/next": {
-      "version": "16.1.7",
-      "resolved": "https://registry.npmjs.org/next/-/next-16.1.7.tgz",
-      "integrity": "sha512-WM0L7WrSvKwoLegLYr6V+mz+RIofqQgVAfHhMp9a88ms0cFX8iX9ew+snpWlSBwpkURJOUdvCEt3uLl3NNzvWg==",
+      "version": "16.0.10",
+      "resolved": "https://registry.npmjs.org/next/-/next-16.0.10.tgz",
+      "integrity": "sha512-RtWh5PUgI+vxlV3HdR+IfWA1UUHu0+Ram/JBO4vWB54cVPentCD0e+lxyAYEsDTqGGMg7qpjhKh6dc6aW7W/sA==",
       "license": "MIT",
       "dependencies": {
-        "@next/env": "16.1.7",
+        "@next/env": "16.0.10",
         "@swc/helpers": "0.5.15",
-        "baseline-browser-mapping": "^2.9.19",
         "caniuse-lite": "^1.0.30001579",
         "postcss": "8.4.31",
         "styled-jsx": "5.1.6"
@@ -14677,14 +14689,14 @@
         "node": ">=20.9.0"
       },
       "optionalDependencies": {
-        "@next/swc-darwin-arm64": "16.1.7",
-        "@next/swc-darwin-x64": "16.1.7",
-        "@next/swc-linux-arm64-gnu": "16.1.7",
-        "@next/swc-linux-arm64-musl": "16.1.7",
-        "@next/swc-linux-x64-gnu": "16.1.7",
-        "@next/swc-linux-x64-musl": "16.1.7",
-        "@next/swc-win32-arm64-msvc": "16.1.7",
-        "@next/swc-win32-x64-msvc": "16.1.7",
+        "@next/swc-darwin-arm64": "16.0.10",
+        "@next/swc-darwin-x64": "16.0.10",
+        "@next/swc-linux-arm64-gnu": "16.0.10",
+        "@next/swc-linux-arm64-musl": "16.0.10",
+        "@next/swc-linux-x64-gnu": "16.0.10",
+        "@next/swc-linux-x64-musl": "16.0.10",
+        "@next/swc-win32-arm64-msvc": "16.0.10",
+        "@next/swc-win32-x64-msvc": "16.0.10",
         "sharp": "^0.34.4"
       },
       "peerDependencies": {

package.json

@@ -1,6 +1,6 @@
 {
   "name": "omniroute",
-  "version": "3.0.0-rc.5",
+  "version": "3.0.0-rc.10",
   "description": "Smart AI Router with auto fallback — route to FREE & cheap models, zero downtime. Works with Cursor, Cline, Claude Desktop, Codex, and any OpenAI-compatible tool.",
   "type": "module",
   "bin": {
@@ -96,7 +96,7 @@
     "jose": "^6.1.3",
     "lowdb": "^7.0.1",
     "monaco-editor": "^0.55.1",
-    "next": "^16.1.6",
+    "next": "^16.0.10",
     "next-intl": "^4.8.3",
     "node-machine-id": "^1.1.12",
     "open": "^11.0.0",
@@ -125,7 +125,7 @@
     "concurrently": "^9.2.1",
     "cross-env": "^10.1.0",
     "eslint": "^9.39.2",
-    "eslint-config-next": "16.1.6",
+    "eslint-config-next": "^16.0.10",
     "husky": "^9.1.7",
     "lint-staged": "^16.2.7",
     "prettier": "^3.8.1",

src/app/(dashboard)/dashboard/providers/[id]/page.tsx

@@ -3017,6 +3017,7 @@ CooldownTimer.propTypes = {
 const ERROR_TYPE_LABELS = {
   runtime_error: { labelKey: "errorTypeRuntime", variant: "warning" },
   upstream_auth_error: { labelKey: "errorTypeUpstreamAuth", variant: "error" },
+  account_deactivated: { labelKey: "Account Deactivated", variant: "error" },
   auth_missing: { labelKey: "errorTypeMissingCredential", variant: "warning" },
   token_refresh_failed: { labelKey: "errorTypeRefreshFailed", variant: "warning" },
   token_expired: { labelKey: "errorTypeTokenExpired", variant: "warning" },
@@ -3025,10 +3026,14 @@ const ERROR_TYPE_LABELS = {
   network_error: { labelKey: "errorTypeNetworkError", variant: "warning" },
   unsupported: { labelKey: "errorTypeTestUnsupported", variant: "default" },
   upstream_error: { labelKey: "errorTypeUpstreamError", variant: "error" },
+  banned: { labelKey: "403 Banned", variant: "error" },
+  credits_exhausted: { labelKey: "No Credits", variant: "warning" },
 };
 
 function inferErrorType(connection, isCooldown) {
   if (isCooldown) return "upstream_rate_limited";
+  if (connection.testStatus === "banned") return "banned";
+  if (connection.testStatus === "credits_exhausted") return "credits_exhausted";
   if (connection.lastErrorType) return connection.lastErrorType;
 
   const code = Number(connection.errorCode);
@@ -3108,6 +3113,16 @@ function getStatusPresentation(connection, effectiveStatus, isCooldown, t) {
     };
   }
 
+  if (errorType === "account_deactivated") {
+    return {
+      statusVariant: "error",
+      statusLabel: t("statusDeactivated", "Deactivated"),
+      errorType,
+      errorBadge,
+      errorTextClass: "text-red-600 font-bold",
+    };
+  }
+
   if (
     errorType === "upstream_auth_error" ||
     errorType === "auth_missing" ||
@@ -3153,6 +3168,26 @@ function getStatusPresentation(connection, effectiveStatus, isCooldown, t) {
     };
   }
 
+  if (errorType === "banned") {
+    return {
+      statusVariant: "error",
+      statusLabel: t("statusBanned", "Banned (403)"),
+      errorType,
+      errorBadge,
+      errorTextClass: "text-red-600 font-bold",
+    };
+  }
+
+  if (errorType === "credits_exhausted") {
+    return {
+      statusVariant: "warning",
+      statusLabel: t("statusCreditsExhausted", "Out of Credits"),
+      errorType,
+      errorBadge,
+      errorTextClass: "text-amber-500",
+    };
+  }
+
   const fallbackStatusMap = {
     unavailable: t("statusUnavailable"),
     failed: t("statusFailed"),
@@ -3520,12 +3555,16 @@ function AddApiKeyModal({
   const t = useTranslations("providers");
   const isBailian = provider === "bailian-coding-plan";
   const defaultBailianUrl = "https://coding-intl.dashscope.aliyuncs.com/apps/anthropic/v1";
+  const isVertex = provider === "vertex";
+  const defaultRegion = "us-central1";
 
   const [formData, setFormData] = useState({
     name: "",
     apiKey: "",
     priority: 1,
     baseUrl: isBailian ? defaultBailianUrl : "",
+    region: isVertex ? defaultRegion : "",
+    validationModelId: "",
   });
   const [validating, setValidating] = useState(false);
   const [validationResult, setValidationResult] = useState(null);
@@ -3539,7 +3578,11 @@ function AddApiKeyModal({
       const res = await fetch("/api/providers/validate", {
         method: "POST",
         headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ provider, apiKey: formData.apiKey }),
+        body: JSON.stringify({
+          provider,
+          apiKey: formData.apiKey,
+          validationModelId: formData.validationModelId || undefined,
+        }),
       });
       const data = await res.json();
       setValidationResult(data.valid ? "success" : "failed");
@@ -3573,7 +3616,11 @@ function AddApiKeyModal({
         const res = await fetch("/api/providers/validate", {
           method: "POST",
           headers: { "Content-Type": "application/json" },
-          body: JSON.stringify({ provider, apiKey: formData.apiKey }),
+          body: JSON.stringify({
+            provider,
+            apiKey: formData.apiKey,
+            validationModelId: formData.validationModelId || undefined,
+          }),
         });
         const data = await res.json();
         isValid = !!data.valid;
@@ -3602,6 +3649,10 @@ function AddApiKeyModal({
         payload.providerSpecificData = {
           baseUrl: validatedBailianBaseUrl,
         };
+      } else if (isVertex) {
+        payload.providerSpecificData = {
+          region: formData.region,
+        };
       }
 
       const error = await onSave(payload);
@@ -3635,6 +3686,7 @@ function AddApiKeyModal({
             value={formData.apiKey}
             onChange={(e) => setFormData({ ...formData, apiKey: e.target.value })}
             className="flex-1"
+            placeholder={isVertex ? "Cole o Service Account JSON aqui" : undefined}
           />
           <div className="pt-6">
             <Button
@@ -3667,6 +3719,13 @@ function AddApiKeyModal({
                 })}
           </p>
         )}
+        <Input
+          label="Model ID (opcional)"
+          placeholder="ex: grok-3 ou meta-llama/Llama-3.1-8B-Instruct"
+          value={formData.validationModelId}
+          onChange={(e) => setFormData({ ...formData, validationModelId: e.target.value })}
+          hint="Usado como fallback se a listagem de models não estiver disponível"
+        />
         <Input
           label={t("priorityLabel")}
           type="number"
@@ -3684,6 +3743,15 @@ function AddApiKeyModal({
             hint="Optional: Custom base URL for bailian-coding-plan provider"
           />
         )}
+        {isVertex && (
+          <Input
+            label="Região (Region)"
+            value={formData.region}
+            onChange={(e) => setFormData({ ...formData, region: e.target.value })}
+            placeholder={defaultRegion}
+            hint="ex: us-central1 ou europe-west4. Partner models usam a região global automaticamente."
+          />
+        )}
         <div className="flex gap-2">
           <Button
             onClick={handleSubmit}
@@ -3732,6 +3800,8 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
     apiKey: "",
     healthCheckInterval: 60,
     baseUrl: "",
+    region: "",
+    validationModelId: "",
   });
   const [testing, setTesting] = useState(false);
   const [testResult, setTestResult] = useState(null);
@@ -3744,17 +3814,23 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
 
   const isBailian = connection?.provider === "bailian-coding-plan";
   const defaultBailianUrl = "https://coding-intl.dashscope.aliyuncs.com/apps/anthropic/v1";
+  const isVertex = connection?.provider === "vertex";
+  const defaultRegion = "us-central1";
 
   useEffect(() => {
     if (connection) {
       const rawBaseUrl = connection.providerSpecificData?.baseUrl;
       const existingBaseUrl = typeof rawBaseUrl === "string" ? rawBaseUrl : "";
+      const rawRegion = connection.providerSpecificData?.region;
+      const existingRegion = typeof rawRegion === "string" ? rawRegion : "";
       setFormData({
         name: connection.name || "",
         priority: connection.priority || 1,
         apiKey: "",
         healthCheckInterval: connection.healthCheckInterval ?? 60,
         baseUrl: existingBaseUrl || (isBailian ? defaultBailianUrl : ""),
+        region: existingRegion || (isVertex ? defaultRegion : ""),
+        validationModelId: connection.providerSpecificData?.validationModelId || "",
       });
       // Load existing extra keys from providerSpecificData
       const existing = connection.providerSpecificData?.extraApiKeys;
@@ -3771,7 +3847,13 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
     setTesting(true);
     setTestResult(null);
     try {
-      const res = await fetch(`/api/providers/${connection.id}/test`, { method: "POST" });
+      const res = await fetch(`/api/providers/${connection.id}/test`, {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({
+          validationModelId: formData.validationModelId || undefined,
+        }),
+      });
       const data = await res.json();
       setTestResult({
         valid: !!data.valid,
@@ -3797,7 +3879,11 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
       const res = await fetch("/api/providers/validate", {
         method: "POST",
         headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ provider: connection.provider, apiKey: formData.apiKey }),
+        body: JSON.stringify({
+          provider: connection.provider,
+          apiKey: formData.apiKey,
+          validationModelId: formData.validationModelId || undefined,
+        }),
       });
       const data = await res.json();
       setValidationResult(data.valid ? "success" : "failed");
@@ -3838,7 +3924,11 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
             const res = await fetch("/api/providers/validate", {
               method: "POST",
               headers: { "Content-Type": "application/json" },
-              body: JSON.stringify({ provider: connection.provider, apiKey: formData.apiKey }),
+              body: JSON.stringify({
+                provider: connection.provider,
+                apiKey: formData.apiKey,
+                validationModelId: formData.validationModelId || undefined,
+              }),
             });
             const data = await res.json();
             isValid = !!data.valid;
@@ -3865,9 +3955,14 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
           ...(connection.providerSpecificData || {}),
           extraApiKeys: extraApiKeys.filter((k) => k.trim().length > 0),
         };
+        if (formData.validationModelId) {
+          updates.providerSpecificData.validationModelId = formData.validationModelId;
+        }
         // Update baseUrl for bailian-coding-plan
         if (isBailian) {
           updates.providerSpecificData.baseUrl = validatedBailianBaseUrl;
+        } else if (isVertex) {
+          updates.providerSpecificData.region = formData.region;
         }
       }
       const error = (await onSave(updates)) as void | unknown;
@@ -3935,7 +4030,7 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
                 type="password"
                 value={formData.apiKey}
                 onChange={(e) => setFormData({ ...formData, apiKey: e.target.value })}
-                placeholder={t("enterNewApiKey")}
+                placeholder={isVertex ? "Cole o Service Account JSON aqui" : t("enterNewApiKey")}
                 hint={t("leaveBlankKeepCurrentApiKey")}
                 className="flex-1"
               />
@@ -3959,6 +4054,13 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
                 {saveError}
               </div>
             )}
+            <Input
+              label="Model ID (opcional)"
+              placeholder="ex: grok-3 ou meta-llama/Llama-3.1-8B-Instruct"
+              value={formData.validationModelId}
+              onChange={(e) => setFormData({ ...formData, validationModelId: e.target.value })}
+              hint="Usado como fallback se a listagem de models não estiver disponível"
+            />
           </>
         )}
 
@@ -3972,6 +4074,16 @@ function EditConnectionModal({ isOpen, connection, onSave, onClose }: EditConnec
           />
         )}
 
+        {isVertex && (
+          <Input
+            label="Região (Region)"
+            value={formData.region}
+            onChange={(e) => setFormData({ ...formData, region: e.target.value })}
+            placeholder={defaultRegion}
+            hint="ex: us-central1 ou europe-west4. Partner models usam a região global automaticamente."
+          />
+        )}
+
         {/* T07: Extra API Keys for round-robin rotation */}
         {!isOAuth && (
           <div className="flex flex-col gap-2">

src/app/api/providers/[id]/models/route.ts

@@ -4,6 +4,7 @@ import {
   isOpenAICompatibleProvider,
   isAnthropicCompatibleProvider,
 } from "@/shared/constants/providers";
+import { PROVIDER_MODELS } from "@/shared/constants/models";
 
 type JsonRecord = Record<string, unknown>;
 
@@ -336,39 +337,85 @@ export async function GET(request, { params }) {
         );
       }
 
-      let modelsUrl = baseUrl.replace(/\/$/, "");
-      if (modelsUrl.endsWith("/chat/completions")) {
-        modelsUrl = modelsUrl.slice(0, -17) + "/models";
-      } else if (modelsUrl.endsWith("/completions")) {
-        modelsUrl = modelsUrl.slice(0, -12) + "/models";
-      } else {
-        modelsUrl = `${modelsUrl}/models`;
+      let base = baseUrl.replace(/\/$/, "");
+      if (base.endsWith("/chat/completions")) {
+        base = base.slice(0, -17);
+      } else if (base.endsWith("/completions")) {
+        base = base.slice(0, -12);
+      } else if (base.endsWith("/v1")) {
+        base = base.slice(0, -3);
       }
 
-      const response = await fetch(modelsUrl, {
-        method: "GET",
-        headers: {
-          "Content-Type": "application/json",
-          Authorization: `Bearer ${apiKey}`,
-        },
-      });
+      // T39: Try multiple endpoint formats
+      const endpoints = [
+        `${base}/v1/models`,
+        `${base}/models`,
+        `${baseUrl.replace(/\/$/, "")}/models`, // Original fallback
+      ];
 
-      if (!response.ok) {
-        const errorText = await response.text();
-        console.log(`Error fetching models from ${provider}:`, errorText);
-        return NextResponse.json(
-          { error: `Failed to fetch models: ${response.status}` },
-          { status: response.status }
-        );
+      // Remove duplicates
+      const uniqueEndpoints = [...new Set(endpoints)];
+      let models = null;
+      let lastErrorStatus = null;
+
+      for (const modelsUrl of uniqueEndpoints) {
+        try {
+          const response = await fetch(modelsUrl, {
+            method: "GET",
+            headers: {
+              "Content-Type": "application/json",
+              Authorization: `Bearer ${apiKey}`,
+            },
+            signal: AbortSignal.timeout(5000), // Quick timeout for fallbacks
+          });
+
+          if (response.ok) {
+            const data = await response.json();
+            models = data.data || data.models || [];
+            break; // Success!
+          }
+
+          if (response.status === 401 || response.status === 403) {
+            lastErrorStatus = response.status;
+            throw new Error("auth_failed");
+          }
+        } catch (err: any) {
+          if (err.message === "auth_failed") break; // Don't try other endpoints if auth failed
+        }
       }
 
-      const data = await response.json();
-      const models = data.data || data.models || [];
+      // If all endpoints failed (but not because of auth), fallback to local catalog
+      if (!models) {
+        if (lastErrorStatus === 401 || lastErrorStatus === 403) {
+          return NextResponse.json(
+            { error: `Auth failed: ${lastErrorStatus}` },
+            { status: lastErrorStatus }
+          );
+        }
+
+        console.warn(`[models] All endpoints failed for ${provider}, using local catalog`);
+        const localModels = PROVIDER_MODELS[provider] || [];
+        models = localModels.map((m: any) => ({
+          id: m.id,
+          name: m.name || m.id,
+          owned_by: provider,
+        }));
+      }
+
+      // Track source for MCP tool T39 requirement
+      const source =
+        models === null || (models && models.length > 0 && models[0].owned_by === provider)
+          ? "local_catalog"
+          : "api";
 
       return NextResponse.json({
         provider,
         connectionId,
         models,
+        source,
+        ...(source === "local_catalog"
+          ? { warning: "API unavailable — using cached catalog" }
+          : {}),
       });
     }
 

src/app/api/providers/[id]/test/route.ts

@@ -523,9 +523,10 @@ async function testApiKeyConnection(connection: any) {
 /**
  * Core test logic — reusable by test-batch without HTTP self-calls.
  * @param {string} connectionId
+ * @param {string} validationModelId Optional custom model ID to test connection with
  * @returns {Promise<object>} Test result (same shape as the JSON response)
  */
-export async function testSingleConnection(connectionId: string) {
+export async function testSingleConnection(connectionId: string, validationModelId?: string) {
   const connection = await getProviderConnectionById(connectionId);
 
   if (!connection) {
@@ -567,8 +568,17 @@ export async function testSingleConnection(connectionId: string) {
       diagnosis: (runtime as any).diagnosis,
     };
   } else if (connection.authType === "apikey") {
+    const enrichedConnection = validationModelId
+      ? {
+          ...connection,
+          providerSpecificData: {
+            ...((connection.providerSpecificData as any) || {}),
+            validationModelId,
+          },
+        }
+      : connection;
     result = await runWithProxyContext(proxyInfo?.proxy || null, () =>
-      testApiKeyConnection(connection)
+      testApiKeyConnection(enrichedConnection)
     );
   } else {
     result = await runWithProxyContext(proxyInfo?.proxy || null, () =>
@@ -670,7 +680,17 @@ export async function testSingleConnection(connectionId: string) {
 export async function POST(request: Request, { params }: { params: Promise<{ id: string }> }) {
   try {
     const { id } = await params;
-    const data = await testSingleConnection(id);
+
+    // Parse optional body for validationModelId
+    let validationModelId;
+    try {
+      const body = await request.json();
+      validationModelId = body?.validationModelId;
+    } catch {
+      // Body is optional
+    }
+
+    const data = await testSingleConnection(id, validationModelId);
 
     if (data.error === "Connection not found") {
       return NextResponse.json({ error: "Connection not found" }, { status: 404 });

src/app/api/providers/validate/route.ts

@@ -30,9 +30,9 @@ export async function POST(request) {
     if (isValidationFailure(validation)) {
       return NextResponse.json({ error: validation.error }, { status: 400 });
     }
-    const { provider, apiKey } = validation.data;
+    const { provider, apiKey, validationModelId } = validation.data;
 
-    let providerSpecificData = {};
+    let providerSpecificData: any = { validationModelId };
 
     if (isOpenAICompatibleProvider(provider) || isAnthropicCompatibleProvider(provider)) {
       const node: any = await getProviderNodeById(provider);
@@ -44,6 +44,7 @@ export async function POST(request) {
         );
       }
       providerSpecificData = {
+        ...providerSpecificData,
         baseUrl: node.baseUrl,
         apiType: node.apiType,
       };

src/app/api/usage/analytics/route.ts

@@ -1,6 +1,36 @@
 import { NextResponse } from "next/server";
 import { getUsageDb } from "@/lib/usageDb";
 import { computeAnalytics } from "@/lib/usageAnalytics";
+import { getDbInstance } from "@/lib/db/core";
+
+function getRangeStartIso(range: string): string | null {
+  const end = new Date();
+  const start = new Date(end);
+
+  switch (range) {
+    case "1d":
+      start.setDate(start.getDate() - 1);
+      break;
+    case "7d":
+      start.setDate(start.getDate() - 7);
+      break;
+    case "30d":
+      start.setDate(start.getDate() - 30);
+      break;
+    case "90d":
+      start.setDate(start.getDate() - 90);
+      break;
+    case "ytd":
+      start.setMonth(0, 1);
+      start.setHours(0, 0, 0, 0);
+      break;
+    case "all":
+    default:
+      return null;
+  }
+
+  return start.toISOString();
+}
 
 export async function GET(request) {
   try {
@@ -36,6 +66,47 @@ export async function GET(request) {
 
     const analytics = await computeAnalytics(history, range, connectionMap);
 
+    // T01: fallback transparency metrics from call_logs (requested_model vs routed model).
+    try {
+      const db = getDbInstance();
+      const sinceIso = getRangeStartIso(range);
+      const whereClause = sinceIso ? "WHERE timestamp >= @since" : "";
+      const row = db
+        .prepare(
+          `
+          SELECT
+            COUNT(*) as total,
+            SUM(CASE WHEN requested_model IS NOT NULL AND requested_model != '' THEN 1 ELSE 0 END) as with_requested,
+            SUM(CASE
+              WHEN requested_model IS NOT NULL
+               AND requested_model != ''
+               AND model IS NOT NULL
+               AND requested_model != model
+              THEN 1 ELSE 0 END
+            ) as fallbacks
+          FROM call_logs
+          ${whereClause}
+        `
+        )
+        .get(sinceIso ? { since: sinceIso } : {}) as
+        | { total?: number; with_requested?: number; fallbacks?: number }
+        | undefined;
+
+      const total = Number(row?.total || 0);
+      const withRequested = Number(row?.with_requested || 0);
+      const fallbackCount = Number(row?.fallbacks || 0);
+
+      analytics.summary.fallbackCount = fallbackCount;
+      analytics.summary.fallbackRatePct =
+        withRequested > 0 ? Number(((fallbackCount / withRequested) * 100).toFixed(2)) : 0;
+      analytics.summary.requestedModelCoveragePct =
+        total > 0 ? Number(((withRequested / total) * 100).toFixed(2)) : 0;
+    } catch {
+      analytics.summary.fallbackCount = 0;
+      analytics.summary.fallbackRatePct = 0;
+      analytics.summary.requestedModelCoveragePct = 0;
+    }
+
     return NextResponse.json(analytics);
   } catch (error) {
     console.error("Error computing analytics:", error);

src/lib/db/apiKeys.ts

@@ -40,6 +40,8 @@ interface ApiKeyMetadata {
   accessSchedule: AccessSchedule | null;
   maxRequestsPerDay: number | null;
   maxRequestsPerMinute: number | null;
+  // T08: Per-key max concurrent sticky sessions (0 = unlimited)
+  maxSessions: number;
 }
 
 interface ApiKeyRow extends JsonRecord {
@@ -197,6 +199,11 @@ function ensureApiKeysColumns(db: ApiKeysDbLike) {
       db.exec("ALTER TABLE api_keys ADD COLUMN max_requests_per_minute INTEGER");
       console.log("[DB] Added api_keys.max_requests_per_minute column");
     }
+    // T08: max concurrent sticky sessions per key (0 = unlimited)
+    if (!columnNames.has("max_sessions")) {
+      db.exec("ALTER TABLE api_keys ADD COLUMN max_sessions INTEGER NOT NULL DEFAULT 0");
+      console.log("[DB] Added api_keys.max_sessions column");
+    }
     _schemaChecked = true;
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
@@ -222,7 +229,7 @@ function getPreparedStatements(db: ApiKeysDbLike): ApiKeysStatements {
     _stmtGetKeyById = db.prepare<ApiKeyRow>("SELECT * FROM api_keys WHERE id = ?");
     _stmtValidateKey = db.prepare<JsonRecord>("SELECT 1 FROM api_keys WHERE key = ?");
     _stmtGetKeyMetadata = db.prepare<ApiKeyRow>(
-      "SELECT id, name, machine_id, allowed_models, allowed_connections, no_log, auto_resolve, is_active, access_schedule, max_requests_per_day, max_requests_per_minute FROM api_keys WHERE key = ?"
+      "SELECT id, name, machine_id, allowed_models, allowed_connections, no_log, auto_resolve, is_active, access_schedule, max_requests_per_day, max_requests_per_minute, max_sessions FROM api_keys WHERE key = ?"
     );
     _stmtInsertKey = db.prepare(
       "INSERT INTO api_keys (id, name, key, machine_id, allowed_models, no_log, created_at) VALUES (?, ?, ?, ?, ?, ?, ?)"
@@ -418,6 +425,8 @@ export async function updateApiKeyPermissions(
         accessSchedule?: AccessSchedule | null;
         maxRequestsPerDay?: number | null;
         maxRequestsPerMinute?: number | null;
+        // T08: max concurrent sessions for this key (0 = unlimited)
+        maxSessions?: number | null;
       }
 ) {
   const db = getDbInstance() as ApiKeysDbLike;
@@ -436,6 +445,7 @@ export async function updateApiKeyPermissions(
           accessSchedule: update.accessSchedule,
           maxRequestsPerDay: update.maxRequestsPerDay,
           maxRequestsPerMinute: update.maxRequestsPerMinute,
+          maxSessions: (update as { maxSessions?: number | null }).maxSessions,
         };
 
   if (
@@ -447,7 +457,8 @@ export async function updateApiKeyPermissions(
     normalized.isActive === undefined &&
     normalized.accessSchedule === undefined &&
     normalized.maxRequestsPerDay === undefined &&
-    normalized.maxRequestsPerMinute === undefined
+    normalized.maxRequestsPerMinute === undefined &&
+    (normalized as Record<string, unknown>).maxSessions === undefined
   ) {
     return false;
   }
@@ -464,6 +475,7 @@ export async function updateApiKeyPermissions(
     accessSchedule?: string | null;
     maxRequestsPerDay?: number | null;
     maxRequestsPerMinute?: number | null;
+    maxSessions?: number;
   } = { id };
 
   if (normalized.name !== undefined) {
@@ -514,6 +526,12 @@ export async function updateApiKeyPermissions(
     params.maxRequestsPerMinute = normalized.maxRequestsPerMinute;
   }
 
+  const maxSessionsUpdate = (normalized as Record<string, unknown>).maxSessions;
+  if (maxSessionsUpdate !== undefined) {
+    updates.push("max_sessions = @maxSessions");
+    params.maxSessions = typeof maxSessionsUpdate === "number" ? Math.max(0, maxSessionsUpdate) : 0;
+  }
+
   const result = db.prepare(`UPDATE api_keys SET ${updates.join(", ")} WHERE id = @id`).run(params);
 
   if (result.changes === 0) return false;
@@ -605,6 +623,8 @@ export async function getApiKeyMetadata(
   const rawMaxRPD = record.max_requests_per_day ?? record.maxRequestsPerDay;
   const rawMaxRPM = record.max_requests_per_minute ?? record.maxRequestsPerMinute;
 
+  const rawMaxSessions = record.max_sessions ?? record.maxSessions;
+
   const metadata: ApiKeyMetadata = {
     id: metadataId,
     name: metadataName,
@@ -619,6 +639,8 @@ export async function getApiKeyMetadata(
     accessSchedule: parseAccessSchedule(record.access_schedule ?? record.accessSchedule),
     maxRequestsPerDay: typeof rawMaxRPD === "number" && rawMaxRPD > 0 ? rawMaxRPD : null,
     maxRequestsPerMinute: typeof rawMaxRPM === "number" && rawMaxRPM > 0 ? rawMaxRPM : null,
+    // T08: max concurrent sessions; 0 = unlimited (default & backward-compatible)
+    maxSessions: typeof rawMaxSessions === "number" && rawMaxSessions > 0 ? rawMaxSessions : 0,
   };
 
   if (!metadata.id) {

src/lib/db/migrations/009_requested_model.sql

@@ -0,0 +1,9 @@
+-- Migration 009: Add requested_model to call_logs for billing transparency
+-- Tracks the model the client *asked* for vs the model that was *actually routed*.
+-- Needed when a combo falls back: requested_model ≠ model in call_logs.
+-- Ref: sub2api commits 0b845c25 + 4edcfe1f (T01 sub2api gap analysis)
+ALTER TABLE call_logs ADD COLUMN requested_model TEXT DEFAULT NULL;
+
+-- Index for filtering/aggregating by requested_model in Analytics
+CREATE INDEX IF NOT EXISTS idx_call_logs_requested_model
+  ON call_logs(requested_model);

src/lib/db/providers.ts

@@ -513,3 +513,98 @@ export async function deleteProviderNode(id: string) {
   backupDbFile("pre-write");
   return rowToCamel(existing);
 }
+
+// ──────────────── T05: Rate-Limit DB Persistence ──────────────────────────
+// Allows rate-limit state to survive token refresh without being accidentally
+// cleared. DB column rate_limited_until already exists in schema.
+// Ref: sub2api PR #1218 (fix(openai): prevent rescheduling rate-limited accounts)
+
+/**
+ * T05: Persist when a connection is rate-limited, directly in DB.
+ * This survives token refresh — OAuth flows must NOT override this field.
+ *
+ * @param connectionId - The provider_connections.id
+ * @param until - Epoch ms when the rate limit expires (null to clear)
+ */
+export function setConnectionRateLimitUntil(connectionId: string, until: number | null): void {
+  const db = getDbInstance() as unknown as DbLike;
+  db.prepare(
+    "UPDATE provider_connections SET rate_limited_until = ?, updated_at = ? WHERE id = ?"
+  ).run(until, new Date().toISOString(), connectionId);
+  invalidateDbCache("connections");
+}
+
+/**
+ * T05: Check if a connection is currently rate-limited (DB-backed).
+ * Use this before account selection to skip transiently rate-limited accounts.
+ *
+ * @returns true if rate_limited_until is set and in the future
+ */
+export function isConnectionRateLimited(connectionId: string): boolean {
+  const db = getDbInstance() as unknown as DbLike;
+  const row = db
+    .prepare("SELECT rate_limited_until FROM provider_connections WHERE id = ?")
+    .get(connectionId) as { rate_limited_until?: number | null } | undefined;
+  if (!row?.rate_limited_until) return false;
+  return Date.now() < row.rate_limited_until;
+}
+
+/**
+ * T05: Get all connections for a provider that are currently rate-limited.
+ * Returns an array of { id, rateLimitedUntil } for dashboard display.
+ */
+export function getRateLimitedConnections(
+  provider: string
+): Array<{ id: string; rateLimitedUntil: number }> {
+  const db = getDbInstance() as unknown as DbLike;
+  const now = Date.now();
+  const rows = db
+    .prepare(
+      "SELECT id, rate_limited_until FROM provider_connections WHERE provider = ? AND rate_limited_until > ?"
+    )
+    .all(provider, now) as Array<{ id: string; rate_limited_until: number }>;
+  return rows.map((r) => ({ id: r.id, rateLimitedUntil: r.rate_limited_until }));
+}
+
+// ──────────────── T13: Stale Quota Display Fix ─────────────────────────────
+// Codex/Claude quotas display stale cumulative usage after the window resets.
+// By comparing resetAt timestamp to now(), we can show 0 when window has passed.
+// Ref: sub2api PR #1171 (fix: quota display shows stale cumulative usage after reset)
+
+/**
+ * T13: Get effective quota usage, zeroing it out if the window has already reset.
+ *
+ * @param used - Stored usage value (tokens used in the window)
+ * @param resetAt - ISO-8601 string or epoch ms when the window resets, or null
+ * @returns Effective usage: 0 if window expired, original value otherwise
+ */
+export function getEffectiveQuotaUsage(
+  used: number,
+  resetAt: string | number | null | undefined
+): number {
+  if (!resetAt) return used;
+  const resetTime = typeof resetAt === "number" ? resetAt : new Date(resetAt).getTime();
+  if (isNaN(resetTime)) return used;
+  // Window has passed — display should show 0 (pending next snapshot)
+  if (Date.now() >= resetTime) return 0;
+  return used;
+}
+
+/**
+ * T13: Format a reset countdown as a human-readable string: "2h 35m" or "4m 30s".
+ * Returns null if resetAt is in the past or not set.
+ */
+export function formatResetCountdown(resetAt: string | number | null | undefined): string | null {
+  if (!resetAt) return null;
+  const resetTime = typeof resetAt === "number" ? resetAt : new Date(resetAt).getTime();
+  if (isNaN(resetTime)) return null;
+  const diffMs = resetTime - Date.now();
+  if (diffMs <= 0) return null;
+  const totalSeconds = Math.floor(diffMs / 1000);
+  const hours = Math.floor(totalSeconds / 3600);
+  const minutes = Math.floor((totalSeconds % 3600) / 60);
+  const seconds = totalSeconds % 60;
+  if (hours > 0) return `${hours}h ${minutes}m`;
+  if (minutes > 0) return `${minutes}m ${seconds}s`;
+  return `${seconds}s`;
+}

src/lib/ipUtils.ts

@@ -0,0 +1,56 @@
+import { isIP } from "node:net";
+
+/**
+ * T07: Extract the real client IP from X-Forwarded-For header.
+ * Skips invalid entries like "unknown" or empty strings.
+ * Falls back to remoteAddress if no valid IP found.
+ * Ref: sub2api PR #1135
+ *
+ * @param xForwardedFor - Value of the X-Forwarded-For header (may be CSV)
+ * @param remoteAddress - Fallback from the raw socket (req.socket.remoteAddress)
+ * @returns The first valid IP address found, or "unknown"
+ */
+export function extractClientIp(
+  xForwardedFor: string | null | undefined,
+  remoteAddress: string | undefined
+): string {
+  if (xForwardedFor) {
+    const entries = xForwardedFor.split(",");
+    for (const entry of entries) {
+      const trimmed = entry.trim();
+      if (trimmed && isIP(trimmed) !== 0) {
+        return trimmed; // First valid IP wins
+      }
+    }
+  }
+  return remoteAddress?.trim() ?? "unknown";
+}
+
+/**
+ * Extract client IP from a Request or NextRequest object.
+ * Checks X-Forwarded-For, X-Real-IP, CF-Connecting-IP, then socket.
+ */
+export function getClientIpFromRequest(req: {
+  headers?: Headers | { get?: (n: string) => string | null };
+  socket?: { remoteAddress?: string };
+  ip?: string;
+}): string {
+  // Helper to get header value from either Headers object or plain object
+  const getHeader = (name: string): string | null => {
+    if (!req.headers) return null;
+    if (typeof (req.headers as Headers).get === "function") {
+      return (req.headers as Headers).get(name);
+    }
+    return null;
+  };
+
+  // Priority: CF-Connecting-IP (Cloudflare) > X-Forwarded-For > X-Real-IP > socket
+  const cfIp = getHeader("cf-connecting-ip");
+  if (cfIp && isIP(cfIp.trim()) !== 0) return cfIp.trim();
+
+  const xff = getHeader("x-forwarded-for");
+  const realIp = getHeader("x-real-ip");
+  const remoteAddress = req.ip ?? req.socket?.remoteAddress;
+
+  return extractClientIp(xff ?? realIp, remoteAddress);
+}

src/lib/localDb.ts

@@ -23,6 +23,15 @@ export {
   createProviderNode,
   updateProviderNode,
   deleteProviderNode,
+
+  // T05: Rate-limit DB persistence (survives token refresh)
+  setConnectionRateLimitUntil,
+  isConnectionRateLimited,
+  getRateLimitedConnections,
+
+  // T13: Stale quota display fix (zero out usage after window resets)
+  getEffectiveQuotaUsage,
+  formatResetCountdown,
 } from "./db/providers";
 
 export {

src/lib/proxyHealth.ts

@@ -0,0 +1,140 @@
+/**
+ * T14: Proxy Fast-Fail — TCP health check with in-memory cache.
+ *
+ * When a configured HTTP/SOCKS5 proxy is unreachable, every request
+ * through OmniRoute used to wait for the full PROXY_TIMEOUT_MS (30s)
+ * before failing. This module detects dead proxies in <2s via a quick
+ * TCP connection check, caching the result to avoid overhead per request.
+ *
+ * Ref: sub2api PR #1167 (fix: proxy-fast-fail)
+ */
+
+import { createConnection } from "node:net";
+
+// Configurable via env vars
+const FAST_FAIL_TIMEOUT_MS = parseInt(process.env.PROXY_FAST_FAIL_TIMEOUT_MS ?? "2000", 10);
+const HEALTH_CACHE_TTL_MS = parseInt(process.env.PROXY_HEALTH_CACHE_TTL_MS ?? "30000", 10);
+
+interface ProxyHealthEntry {
+  healthy: boolean;
+  checkedAt: number;
+  ttlMs: number;
+}
+
+// In-memory cache: proxyUrl → health entry
+const proxyHealthCache = new Map<string, ProxyHealthEntry>();
+
+/**
+ * T14: Perform a fast TCP check to see if a proxy host:port is reachable.
+ * Results are cached for `cacheTtlMs` (default 30s) to avoid checking every request.
+ *
+ * @param proxyUrl - Full proxy URL, e.g. http://user:pass@1.2.3.4:8080
+ * @param timeoutMs - TCP connection timeout (default 2000ms)
+ * @param cacheTtlMs - How long to cache the health result (default 30000ms)
+ * @returns true if proxy TCP port is open, false otherwise
+ */
+export async function isProxyReachable(
+  proxyUrl: string,
+  timeoutMs = FAST_FAIL_TIMEOUT_MS,
+  cacheTtlMs = HEALTH_CACHE_TTL_MS
+): Promise<boolean> {
+  const cached = proxyHealthCache.get(proxyUrl);
+  if (cached && Date.now() - cached.checkedAt < cached.ttlMs) {
+    return cached.healthy;
+  }
+
+  let url: URL;
+  try {
+    url = new URL(proxyUrl);
+  } catch {
+    // Malformed URL — treat as unreachable
+    proxyHealthCache.set(proxyUrl, {
+      healthy: false,
+      checkedAt: Date.now(),
+      ttlMs: cacheTtlMs,
+    });
+    return false;
+  }
+
+  const host = url.hostname;
+  const port = parseInt(url.port || defaultPortForScheme(url.protocol), 10);
+
+  if (!host || isNaN(port)) {
+    proxyHealthCache.set(proxyUrl, {
+      healthy: false,
+      checkedAt: Date.now(),
+      ttlMs: cacheTtlMs,
+    });
+    return false;
+  }
+
+  const healthy = await tcpCheck(host, port, timeoutMs);
+  proxyHealthCache.set(proxyUrl, { healthy, checkedAt: Date.now(), ttlMs: cacheTtlMs });
+  return healthy;
+}
+
+/**
+ * Get the cached health status of a proxy without re-checking.
+ * Returns null if there is no cached entry.
+ */
+export function getCachedProxyHealth(proxyUrl: string): boolean | null {
+  const cached = proxyHealthCache.get(proxyUrl);
+  if (!cached) return null;
+  if (Date.now() - cached.checkedAt >= cached.ttlMs) return null; // stale
+  return cached.healthy;
+}
+
+/**
+ * Invalidate the cached health for a proxy URL (force re-check on next call).
+ */
+export function invalidateProxyHealth(proxyUrl: string): void {
+  proxyHealthCache.delete(proxyUrl);
+}
+
+/**
+ * Get all currently cached proxy health entries (for dashboard display).
+ */
+export function getAllProxyHealthStatuses(): Array<{
+  proxyUrl: string;
+  healthy: boolean;
+  checkedAt: number;
+  stale: boolean;
+}> {
+  const now = Date.now();
+  return [...proxyHealthCache.entries()].map(([proxyUrl, entry]) => ({
+    proxyUrl,
+    healthy: entry.healthy,
+    checkedAt: entry.checkedAt,
+    stale: now - entry.checkedAt >= entry.ttlMs,
+  }));
+}
+
+// ─── Internals ────────────────────────────────────────────────────────────────
+
+function defaultPortForScheme(protocol: string): string {
+  switch (protocol.replace(":", "").toLowerCase()) {
+    case "https":
+      return "443";
+    case "socks5":
+    case "socks5h":
+      return "1080";
+    case "http":
+    default:
+      return "8080";
+  }
+}
+
+function tcpCheck(host: string, port: number, timeoutMs: number): Promise<boolean> {
+  return new Promise<boolean>((resolve) => {
+    const socket = createConnection({ host, port }, () => {
+      socket.destroy();
+      resolve(true);
+    });
+    socket.setTimeout(timeoutMs);
+    socket.on("error", () => resolve(false));
+    socket.on("timeout", () => {
+      socket.destroy();
+      resolve(false);
+    });
+  });
+}

src/lib/usage/callLogs.ts

@@ -180,6 +180,7 @@ export async function saveCallLog(entry: any) {
       path: entry.path || "/v1/chat/completions",
       status: entry.status || 0,
       model: entry.model || "-",
+      requestedModel: entry.requestedModel || null, // T01: model the client asked for
       provider: entry.provider || "-",
       account,
       connectionId: entry.connectionId || null,
@@ -205,10 +206,10 @@ export async function saveCallLog(entry: any) {
     const db = getDbInstance();
     db.prepare(
       `
-      INSERT INTO call_logs (id, timestamp, method, path, status, model, provider,
+      INSERT INTO call_logs (id, timestamp, method, path, status, model, requested_model, provider,
         account, connection_id, duration, tokens_in, tokens_out, request_type, source_format, target_format,
         api_key_id, api_key_name, combo_name, request_body, response_body, error)
-      VALUES (@id, @timestamp, @method, @path, @status, @model, @provider,
+      VALUES (@id, @timestamp, @method, @path, @status, @model, @requestedModel, @provider,
         @account, @connectionId, @duration, @tokensIn, @tokensOut, @requestType, @sourceFormat, @targetFormat,
         @apiKeyId, @apiKeyName, @comboName, @requestBody, @responseBody, @error)
     `
@@ -374,6 +375,7 @@ export async function getCallLogs(filter: any = {}) {
       path: toStringOrNull(l.path),
       status: toNumber(l.status),
       model: toStringOrNull(l.model),
+      requestedModel: toStringOrNull(l.requested_model), // T01: original model from client
       provider: toStringOrNull(l.provider),
       account: toStringOrNull(l.account),
       duration: toNumber(l.duration),

src/shared/components/RequestLoggerDetail.tsx

@@ -147,6 +147,21 @@ export default function RequestLoggerDetail({ log, detail, loading, onClose, onC
               <div className="text-[10px] text-text-muted uppercase tracking-wider mb-1">Model</div>
               <div className="text-sm font-medium text-primary font-mono">{log.model}</div>
             </div>
+            <div>
+              <div className="text-[10px] text-text-muted uppercase tracking-wider mb-1">
+                Requested Model
+              </div>
+              <div
+                className={`text-sm font-medium font-mono ${
+                  (detail?.requestedModel || log.requestedModel) &&
+                  (detail?.requestedModel || log.requestedModel) !== log.model
+                    ? "text-amber-600 dark:text-amber-400"
+                    : "text-text-muted"
+                }`}
+              >
+                {detail?.requestedModel || log.requestedModel || "—"}
+              </div>
+            </div>
             <div>
               <div className="text-[10px] text-text-muted uppercase tracking-wider mb-1">
                 Provider

src/shared/constants/pricing.ts

@@ -802,6 +802,30 @@ export const DEFAULT_PRICING = {
       reasoning: 1.8,
       cache_creation: 0.3,
     },
+    // T12: MiniMax M2.7 — new default model (sub2api PR #1120)
+    // Upgraded from M2.5, same API endpoint api.minimax.io
+    // Pricing estimated, check https://platform.minimaxi.com/document/Price
+    "minimax-m2.7": {
+      input: 0.4,
+      output: 1.6,
+      cached: 0.2,
+      reasoning: 2.4,
+      cache_creation: 0.4,
+    },
+    "MiniMax-M2.7": {
+      input: 0.4,
+      output: 1.6,
+      cached: 0.2,
+      reasoning: 2.4,
+      cache_creation: 0.4,
+    },
+    "minimax-m2.7-highspeed": {
+      input: 0.4,
+      output: 1.6,
+      cached: 0.2,
+      reasoning: 2.4,
+      cache_creation: 0.4,
+    },
   },
 
   // ─── Free-tier API Key Providers (nominal $0 pricing) ───

src/shared/constants/providers.ts

@@ -546,6 +546,20 @@ export const APIKEY_PROVIDERS = {
     freeNote:
       "No API key needed — access GPT-5, Claude, Gemini, DeepSeek V3, Llama 4 free (1 req/15s)",
   },
+  puter: {
+    id: "puter",
+    alias: "pu",
+    name: "Puter AI",
+    icon: "cloud_circle",
+    color: "#6366F1",
+    textIcon: "PU",
+    website: "https://puter.com",
+    hasFree: true,
+    freeNote:
+      "500+ models (GPT-5, Claude Opus 4, Gemini 3 Pro, Grok 4, DeepSeek V3...) — Users pay via free Puter account",
+    passthroughModels: true,
+    authHint: "Get token at puter.com/dashboard → Copy Auth Token",
+  },
   "cloudflare-ai": {
     id: "cloudflare-ai",
     alias: "cf",

tests/unit/auth-terminal-status.test.mjs

@@ -0,0 +1,90 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import fs from "node:fs";
+import os from "node:os";
+import path from "node:path";
+
+const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-auth-terminal-"));
+process.env.DATA_DIR = TEST_DATA_DIR;
+
+const core = await import("../../src/lib/db/core.ts");
+const providersDb = await import("../../src/lib/db/providers.ts");
+const auth = await import("../../src/sse/services/auth.ts");
+
+async function resetStorage() {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+  fs.mkdirSync(TEST_DATA_DIR, { recursive: true });
+}
+
+test.after(() => {
+  core.resetDbInstance();
+  fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
+});
+
+test("getProviderCredentials skips credits_exhausted connections", async () => {
+  await resetStorage();
+
+  const exhausted = await providersDb.createProviderConnection({
+    provider: "openai",
+    authType: "apikey",
+    apiKey: "sk-exhausted",
+    isActive: true,
+    testStatus: "credits_exhausted",
+  });
+
+  const healthy = await providersDb.createProviderConnection({
+    provider: "openai",
+    authType: "apikey",
+    apiKey: "sk-healthy",
+    isActive: true,
+    testStatus: "active",
+  });
+
+  const selected = await auth.getProviderCredentials("openai");
+  assert.ok(selected);
+  assert.equal(selected.connectionId, healthy.id);
+  assert.notEqual(selected.connectionId, exhausted.id);
+});
+
+test("getProviderCredentials returns null when all active connections are terminal", async () => {
+  await resetStorage();
+
+  await providersDb.createProviderConnection({
+    provider: "openai",
+    authType: "apikey",
+    apiKey: "sk-only-exhausted",
+    isActive: true,
+    testStatus: "credits_exhausted",
+  });
+
+  const selected = await auth.getProviderCredentials("openai");
+  assert.equal(selected, null);
+});
+
+test("markAccountUnavailable does not overwrite terminal status", async () => {
+  await resetStorage();
+
+  const conn = await providersDb.createProviderConnection({
+    provider: "openai",
+    authType: "apikey",
+    apiKey: "sk-terminal",
+    isActive: true,
+    testStatus: "credits_exhausted",
+    lastError: "insufficient_quota",
+  });
+
+  const result = await auth.markAccountUnavailable(
+    conn.id,
+    503,
+    "temporary upstream error",
+    "openai",
+    "gpt-4.1"
+  );
+
+  assert.equal(result.shouldFallback, true);
+  assert.equal(result.cooldownMs, 0);
+
+  const after = await providersDb.getProviderConnectionById(conn.id);
+  assert.equal(after.testStatus, "credits_exhausted");
+});

tests/unit/error-classifier.test.mjs

@@ -0,0 +1,35 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { classifyProviderError, PROVIDER_ERROR_TYPES } =
+  await import("../../open-sse/services/errorClassifier.ts");
+
+test("classifyProviderError: 401 + account_deactivated => ACCOUNT_DEACTIVATED", () => {
+  const body = JSON.stringify({
+    error: { message: "account_deactivated: this account has been disabled" },
+  });
+  const result = classifyProviderError(401, body);
+  assert.equal(result, PROVIDER_ERROR_TYPES.ACCOUNT_DEACTIVATED);
+});
+
+test("classifyProviderError: plain 401 => UNAUTHORIZED", () => {
+  const result = classifyProviderError(401, { error: { message: "token expired" } });
+  assert.equal(result, PROVIDER_ERROR_TYPES.UNAUTHORIZED);
+});
+
+test("classifyProviderError: 402 => QUOTA_EXHAUSTED", () => {
+  const result = classifyProviderError(402, { error: { message: "payment required" } });
+  assert.equal(result, PROVIDER_ERROR_TYPES.QUOTA_EXHAUSTED);
+});
+
+test("classifyProviderError: 400 + billing signal => QUOTA_EXHAUSTED", () => {
+  const result = classifyProviderError(400, {
+    error: { message: "insufficient_quota: exceeded your current quota" },
+  });
+  assert.equal(result, PROVIDER_ERROR_TYPES.QUOTA_EXHAUSTED);
+});
+
+test("classifyProviderError: 429 without billing signal => RATE_LIMITED", () => {
+  const result = classifyProviderError(429, { error: { message: "too many requests" } });
+  assert.equal(result, PROVIDER_ERROR_TYPES.RATE_LIMITED);
+});

tests/unit/ip-filter.test.mjs

@@ -11,6 +11,7 @@ const {
   addToWhitelist,
   removeFromWhitelist,
   getIPFilterConfig,
+  checkRequestIP,
   resetIPFilter,
 } = await import("../../open-sse/services/ipFilter.ts");
 
@@ -111,6 +112,48 @@ test("normalizes ::ffff: prefix", () => {
   assert.equal(checkIP("::ffff:1.2.3.4").allowed, false);
 });
 
+// ─── T07: X-Forwarded-For validation ───────────────────────────────────────
+
+test("checkRequestIP: skips invalid XFF entries and uses next valid IP", () => {
+  configureIPFilter({ enabled: true, mode: "whitelist", whitelist: ["1.2.3.4"] });
+  const req = {
+    headers: {
+      get(name) {
+        if (name === "x-forwarded-for") return "unknown, 1.2.3.4";
+        return null;
+      },
+    },
+  };
+  assert.equal(checkRequestIP(req).allowed, true);
+});
+
+test("checkRequestIP: all-invalid XFF falls back to x-real-ip", () => {
+  configureIPFilter({ enabled: true, mode: "whitelist", whitelist: ["9.9.9.9"] });
+  const req = {
+    headers: {
+      get(name) {
+        if (name === "x-forwarded-for") return "unknown, -, not_an_ip";
+        if (name === "x-real-ip") return "9.9.9.9";
+        return null;
+      },
+    },
+  };
+  assert.equal(checkRequestIP(req).allowed, true);
+});
+
+test("checkRequestIP: empty headers fall back to request.ip", () => {
+  configureIPFilter({ enabled: true, mode: "whitelist", whitelist: ["7.7.7.7"] });
+  const req = {
+    headers: {
+      get() {
+        return null;
+      },
+    },
+    ip: "7.7.7.7",
+  };
+  assert.equal(checkRequestIP(req).allowed, true);
+});
+
 // ─── Config API ─────────────────────────────────────────────────────────────
 
 test("getIPFilterConfig: returns serializable config", () => {

tests/unit/nanobanana-image-handler.test.mjs

@@ -13,7 +13,7 @@ test("handleImageGeneration(nanobanana): async submit+poll returns URL payload",
     if (u.includes("/generate-pro")) {
       const body = JSON.parse(options.body);
       assert.equal(body.prompt, "galaxy test");
-      assert.equal(body.aspectRatio, "4:5");
+      assert.equal(body.aspectRatio, "1:1");
       assert.equal(body.resolution, "2K");
 
       return new Response(

tests/unit/openai-to-claude-strip-empty.test.mjs

@@ -0,0 +1,76 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { stripEmptyTextBlocks, openaiToClaudeRequest } =
+  await import("../../open-sse/translator/request/openai-to-claude.ts");
+
+test("stripEmptyTextBlocks removes empty text recursively inside tool_result content", () => {
+  const input = [
+    { type: "text", text: "" },
+    { type: "text", text: "keep-top-level" },
+    {
+      type: "tool_result",
+      content: [
+        { type: "text", text: "" },
+        { type: "text", text: "keep-nested" },
+        {
+          type: "tool_result",
+          content: [
+            { type: "text", text: "" },
+            { type: "text", text: "keep-deep" },
+          ],
+        },
+      ],
+    },
+  ];
+
+  const out = stripEmptyTextBlocks(input);
+  assert.deepEqual(out, [
+    { type: "text", text: "keep-top-level" },
+    {
+      type: "tool_result",
+      content: [
+        { type: "text", text: "keep-nested" },
+        {
+          type: "tool_result",
+          content: [{ type: "text", text: "keep-deep" }],
+        },
+      ],
+    },
+  ]);
+});
+
+test("openaiToClaudeRequest applies strip to tool message array content", () => {
+  const request = {
+    messages: [
+      { role: "user", content: "run tool" },
+      {
+        role: "assistant",
+        content: "",
+        tool_calls: [
+          {
+            id: "call_1",
+            type: "function",
+            function: { name: "demo_tool", arguments: "{}" },
+          },
+        ],
+      },
+      {
+        role: "tool",
+        tool_call_id: "call_1",
+        content: [
+          { type: "text", text: "" },
+          { type: "text", text: "tool ok" },
+        ],
+      },
+    ],
+  };
+
+  const translated = openaiToClaudeRequest("claude-sonnet-4", request, false);
+  const toolMessage = translated.messages.find(
+    (m) => Array.isArray(m.content) && m.content.some((b) => b.type === "tool_result")
+  );
+  assert.ok(toolMessage, "expected a translated tool_result user message");
+  const toolResult = toolMessage.content.find((b) => b.type === "tool_result");
+  assert.deepEqual(toolResult.content, [{ type: "text", text: "tool ok" }]);
+});

tests/unit/plan3-p0.test.mjs

@@ -11,7 +11,10 @@ import { DefaultExecutor } from "../../open-sse/executors/default.ts";
 import { CodexExecutor, setDefaultFastServiceTierEnabled } from "../../open-sse/executors/codex.ts";
 import { translateNonStreamingResponse } from "../../open-sse/handlers/responseTranslator.ts";
 import { extractUsageFromResponse } from "../../open-sse/handlers/usageExtractor.ts";
-import { parseSSEToResponsesOutput } from "../../open-sse/handlers/sseParser.ts";
+import {
+  parseSSEToOpenAIResponse,
+  parseSSEToResponsesOutput,
+} from "../../open-sse/handlers/sseParser.ts";
 
 test("getModelInfoCore resolves unique non-openai unprefixed model", async () => {
   const info = await getModelInfoCore("claude-haiku-4-5-20251001", {});
@@ -382,3 +385,56 @@ test("parseSSEToResponsesOutput returns null for invalid payload", () => {
   const parsed = parseSSEToResponsesOutput("data: not-json\n\ndata: [DONE]\n", "fallback-model");
   assert.equal(parsed, null);
 });
+
+test("parseSSEToOpenAIResponse merges split tool call chunks by id without duplication", () => {
+  const rawSSE = [
+    `data: ${JSON.stringify({
+      id: "chatcmpl_1",
+      object: "chat.completion.chunk",
+      choices: [
+        {
+          index: 0,
+          delta: {
+            tool_calls: [
+              {
+                id: "call_abc",
+                index: 0,
+                type: "function",
+                function: { name: "sum", arguments: '{"a":' },
+              },
+            ],
+          },
+        },
+      ],
+    })}`,
+    `data: ${JSON.stringify({
+      id: "chatcmpl_1",
+      object: "chat.completion.chunk",
+      choices: [
+        {
+          index: 0,
+          delta: {
+            tool_calls: [
+              {
+                id: "call_abc",
+                index: 0,
+                type: "function",
+                function: { arguments: "1}" },
+              },
+            ],
+          },
+          finish_reason: "tool_calls",
+        },
+      ],
+    })}`,
+    "data: [DONE]",
+  ].join("\n");
+
+  const parsed = parseSSEToOpenAIResponse(rawSSE, "gpt-5.1-codex");
+  assert.ok(parsed);
+  assert.equal(parsed.choices[0].finish_reason, "tool_calls");
+  assert.equal(parsed.choices[0].message.tool_calls.length, 1);
+  assert.equal(parsed.choices[0].message.tool_calls[0].id, "call_abc");
+  assert.equal(parsed.choices[0].message.tool_calls[0].function.name, "sum");
+  assert.equal(parsed.choices[0].message.tool_calls[0].function.arguments, '{"a":1}');
+});

tests/unit/session-manager.test.mjs

@@ -8,6 +8,11 @@ const {
   getSessionConnection,
   getActiveSessionCount,
   getActiveSessions,
+  extractExternalSessionId,
+  checkSessionLimit,
+  registerKeySession,
+  unregisterKeySession,
+  isSessionRegisteredForKey,
   clearSessions,
 } = await import("../../open-sse/services/sessionManager.ts");
 
@@ -39,11 +44,17 @@ test("generateSessionId: different model = different ID", () => {
 test("generateSessionId: different system prompt = different ID", () => {
   const body1 = {
     model: "claude-sonnet-4-20250514",
-    messages: [{ role: "system", content: "A" }, { role: "user", content: "hi" }],
+    messages: [
+      { role: "system", content: "A" },
+      { role: "user", content: "hi" },
+    ],
   };
   const body2 = {
     model: "claude-sonnet-4-20250514",
-    messages: [{ role: "system", content: "B" }, { role: "user", content: "hi" }],
+    messages: [
+      { role: "system", content: "B" },
+      { role: "user", content: "hi" },
+    ],
   };
   assert.notEqual(generateSessionId(body1), generateSessionId(body2));
 });
@@ -147,3 +158,25 @@ test("touchSession with null sessionId: no-op", () => {
   touchSession(null);
   assert.equal(getActiveSessionCount(), 0);
 });
+
+test("extractExternalSessionId accepts hyphen and underscore variants", () => {
+  const h1 = new Headers({ "x-session-id": "abc-123" });
+  const h2 = new Headers({ x_session_id: "def-456" });
+
+  assert.equal(extractExternalSessionId(h1), "ext:abc-123");
+  assert.equal(extractExternalSessionId(h2), "ext:def-456");
+});
+
+test("checkSessionLimit enforces max_sessions for new sessions only", () => {
+  const keyId = "key-1";
+  registerKeySession(keyId, "ext:sess-a");
+  assert.equal(isSessionRegisteredForKey(keyId, "ext:sess-a"), true);
+
+  const violation = checkSessionLimit(keyId, 1);
+  assert.ok(violation);
+  assert.equal(violation.code, "SESSION_LIMIT_EXCEEDED");
+
+  unregisterKeySession(keyId, "ext:sess-a");
+  assert.equal(isSessionRegisteredForKey(keyId, "ext:sess-a"), false);
+  assert.equal(checkSessionLimit(keyId, 1), null);
+});

tests/unit/thinking-budget.test.mjs

@@ -101,7 +101,8 @@ test("CUSTOM: sets OpenAI reasoning_effort from budget", () => {
     reasoning_effort: "low",
   };
   const result = applyThinkingBudget(body);
-  assert.equal(result.reasoning_effort, "high");
+  // T11 (sub2api gap): full budget (131072) now maps to "max" instead of "high"
+  assert.equal(result.reasoning_effort, "max");
   setThinkingBudgetConfig(DEFAULT_THINKING_CONFIG);
 });
 
@@ -212,8 +213,8 @@ test("normalizeThinkingLevel: converts Gemini thinkingConfig.thinkingLevel", ()
     },
   };
   const result = normalizeThinkingLevel(body);
-  assert.equal(result.generationConfig.thinking_config.thinking_budget, 131072);
-  assert.equal(result.generationConfig.thinkingConfig, undefined);
+  assert.equal(result.generationConfig.thinkingConfig.thinkingBudget, 131072);
+  assert.equal(result.generationConfig.thinking_config, undefined);
 });
 
 test("normalizeThinkingLevel: ignores unknown string values", () => {