converse-mcp-server 3.2.3 → 3.3.0

package/README.md

@@ -234,6 +234,7 @@ SUMMARIZATION_MODEL=gpt-5-nano        # Default: gpt-5-nano
 
 - **gemini-3.1-pro-preview** (aliases: `pro`, `gemini-pro`): Most advanced reasoning with expanded thinking levels (1M context, 64K output)
 - **gemini-3.5-flash** (aliases: `gemini-3.5`, `flash-3.5`): Frontier-level agentic and coding performance at Flash speed (1M context, 65K output)
+- **gemini-3.8-flash** (aliases: `gemini-3.8`, `flash-3.8`): Current-generation Flash with stronger long-horizon agentic performance (1M context, 65K output; thinking levels low/medium/high — no minimal)
 - **gemini-2.5-pro** (alias: `pro 2.5`): Deep reasoning with thinking budget (1M context, 65K output)
 - **gemini-2.5-flash** (alias: `flash`): Ultra-fast (1M context, 65K output)
 - **gemini-2.5-flash-lite** (alias: `flash-lite`): Lightweight fast model (1M context, 65K output)
@@ -301,7 +302,7 @@ Reach these with the `copilot:` namespace (e.g. `copilot:gpt-5.6-terra`); uses y
 
 - **OpenAI**: `gpt-5.6-sol` (aliases: `gpt-5.6`, `gpt-5`), `gpt-5.6-terra`, `gpt-5.6-luna` (all support `reasoning_effort`)
 - **Anthropic**: `claude-fable-5` (alias: `fable`), `claude-sonnet-5` (alias: `sonnet`), `claude-opus-4.8` (aliases: `opus`, `claude`)
-- **Google**: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.5-flash` (alias: `gemini-flash`)
+- **Google**: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.8-flash` (aliases: `gemini-3.8`, `flash-3.8`), `gemini-3.5-flash` (alias: `gemini-flash`)
 - Any other `copilot:<id>` is forwarded to the Copilot backend verbatim
 
 ## 📚 Help & Documentation
@@ -371,18 +372,42 @@ CODEX_APPROVAL_POLICY=never                  # never (default), untrusted, on-fa
 
 These must be set in your system environment or when launching Claude Code, NOT in the project .env file:
 
-| Variable                | Description                 | Default  | Example                              |
-| ----------------------- | --------------------------- | -------- | ------------------------------------ |
-| `MAX_MCP_OUTPUT_TOKENS` | Token response limit        | `25000`  | `200000`                             |
-| `MCP_TOOL_TIMEOUT`      | Tool execution timeout (ms) | `120000` | `5400000` (90 min for deep research) |
+| Variable                             | Description                                                             | Default                         | Example                     |
+| ------------------------------------ | ----------------------------------------------------------------------- | ------------------------------- | --------------------------- |
+| `MAX_MCP_OUTPUT_TOKENS`              | Token response limit                                                    | `25000`                         | `200000`                    |
+| `MCP_TOOL_TIMEOUT`                   | Wall-clock limit per tool call (ms)                                     | ~28 hours when unset            | `7200000` (2 h)             |
+| `CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT`  | Idle window (ms) — aborts a call that produces no output for this long  | 30 min (stdio) / 5 min (HTTP)   | `3600000` (60 min)          |
 
 ```bash
 # Example: Set globally before starting Claude Code
 export MAX_MCP_OUTPUT_TOKENS=200000
-export MCP_TOOL_TIMEOUT=5400000  # 90 minutes for deep research models
+export CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT=3600000  # 60 min for long silent agentic calls
 claude  # Then start Claude Code
 ```
 
+Or persist them in `~/.claude/settings.json`:
+
+```json
+{
+	"env": {
+		"MAX_MCP_OUTPUT_TOKENS": "200000",
+		"CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT": "3600000"
+	}
+}
+```
+
+**The idle timeout is usually what kills long calls.** Agentic models (Codex, Claude Agent SDK) can work silently for 30+ minutes; Converse holds one MCP request open the whole time, and Claude Code aborts it after the idle window with an error like _"failed after 30 minutes of silence. The idle timeout aborted it."_ Progress-notification heartbeats can't prevent this — Claude Code doesn't send a `progressToken` on `tools/call`, so an MCP server has no spec-compliant way to emit them ([claude-code#58687](https://github.com/anthropics/claude-code/issues/58687)). Raising the idle window is the only fix.
+
+#### Codex CLI Tool Timeout
+
+If you register Converse in OpenAI's Codex CLI, note that Codex enforces its own **hard 300-second default** per MCP tool call (`tool_timeout_sec`, undocumented — it exists only in Codex's config schema). Progress notifications don't extend it. Raise it in `~/.codex/config.toml`:
+
+```toml
+[mcp_servers.converse]
+# ... command/env ...
+tool_timeout_sec = 3600  # default 300 kills long calls at 5 minutes
+```
+
 ### Model Selection
 
 Use `"auto"` for automatic model selection, or specify exact models:
@@ -421,7 +446,7 @@ Use `"auto"` for automatic model selection, or specify exact models:
 Provider priority order (subscription-based SDK providers first, then API-key providers):
 
 1. Codex (`codex`)
-2. Gemini via Antigravity CLI (`gemini`, `gemini:flash`)
+2. Gemini via Antigravity CLI (`gemini` → Gemini 3.8 Flash, `gemini:pro`)
 3. Claude Agent SDK (`claude` → Claude Fable 5)
 4. Copilot (`copilot`)
 5. OpenAI (`gpt-5.6`)
@@ -532,6 +557,13 @@ For development setup, see the [Development](#-development) section below.
 
 - Clear cache and reinstall: `npm run clean`
 
+**Long tool calls aborted mid-run (idle/timeout errors):**
+
+- The abort almost always comes from the MCP _client_, not Converse — Converse's own limits are 30 min per provider call and 90 min per async job.
+- Claude Code: raise `CLAUDE_CODE_MCP_TOOL_IDLE_TIMEOUT` (idle window, default 30 min stdio / 5 min HTTP) and check `MCP_TOOL_TIMEOUT` (wall-clock). See [Claude Code Environment Variables](#claude-code-environment-variables-systemglobal).
+- Codex CLI: set `tool_timeout_sec` under `[mcp_servers.converse]` in `~/.codex/config.toml` — the undocumented default is 300 seconds.
+- Immune alternative: run the call with `async: true` and poll `check_status` — each poll is a fresh short request, so no client timeout applies.
+
 ### Debug Mode
 
 ```bash

package/docs/API.md

@@ -391,6 +391,7 @@ Provide models as plain name strings in the `models` array. Bare names and alias
 |-------|---------|---------|--------|-------|
 | `gemini-3.1-pro-preview` | `pro`, `gemini-pro` | 1M | 64K | Most advanced reasoning, expanded thinking levels |
 | `gemini-3.5-flash` | `gemini-3.5`, `flash-3.5` | 1M | 65K | Frontier agentic/coding at Flash speed |
+| `gemini-3.8-flash` | `gemini-3.8`, `flash-3.8` | 1M | 65K | Current-generation Flash; thinking low/medium/high only |
 | `gemini-2.5-pro` | `pro 2.5` | 1M | 65K | Deep reasoning with thinking budget |
 | `gemini-2.5-flash` | `flash` | 1M | 65K | Ultra-fast |
 | `gemini-2.5-flash-lite` | `flash-lite` | 1M | 65K | Lightweight fast model |
@@ -476,7 +477,7 @@ Any other model works via its full `provider/model` slug (e.g. `anthropic/claude
 
 The **Antigravity CLI** (`agy`) provides subscription-based access to Gemini models through Google OAuth:
 
-- **Models** (text-only): `gemini` (= `gemini:pro`, Gemini 3.1 Pro), `gemini:flash` (Gemini 3.5 Flash)
+- **Models** (text-only): `gemini` (= `gemini:flash`, Gemini 3.8 Flash), `gemini:pro` (Gemini 3.1 Pro)
 - **Authentication**: Google OAuth via `agy` (one-time interactive login)
 - **Setup**: install the Antigravity CLI and run `agy` once to log in
 - **Billing**: uses your Antigravity subscription/compute allowance instead of API credits
@@ -502,7 +503,7 @@ Reach these with the `copilot:` namespace (e.g. `copilot:gpt-5.6-terra`); uses y
 
 - **OpenAI**: `gpt-5.6-sol` (aliases: `gpt-5.6`, `gpt-5`), `gpt-5.6-terra`, `gpt-5.6-luna` (all accept `reasoning_effort`)
 - **Anthropic**: `claude-fable-5` (alias: `fable`), `claude-sonnet-5` (alias: `sonnet`), `claude-opus-5` (aliases: `opus`, `claude`), `claude-opus-4.8`
-- **Google**: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.5-flash` (alias: `gemini-flash`)
+- **Google**: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.8-flash` (aliases: `gemini-3.8`, `flash-3.8`), `gemini-3.5-flash` (alias: `gemini-flash`)
 - Any other `copilot:<id>` is forwarded to the Copilot backend verbatim
 
 ### Model Selection
@@ -522,7 +523,7 @@ Use `"auto"` for automatic selection, or specify exact models:
 "opus"                     // Anthropic API (-> claude-opus-5)
 "claude"                   // Claude Agent SDK (-> Claude Fable 5)
 "claude:opus"              // Claude Agent SDK (Claude Opus 5)
-"gemini"                   // Antigravity CLI (Gemini 3.1 Pro)
+"gemini"                   // Antigravity CLI (Gemini 3.8 Flash)
 "copilot:gpt-5.6-terra"    // GitHub Copilot SDK
 ```
 

package/docs/PROVIDERS.md

@@ -24,6 +24,7 @@ This guide documents all supported AI providers in the Converse MCP Server and t
 - **Supported Models**:
   - `gemini-3.1-pro-preview` (aliases: `pro`, `gemini-pro`) - Most advanced reasoning with expanded thinking levels (1M context, 64K output)
   - `gemini-3.5-flash` (aliases: `gemini-3.5`, `flash-3.5`) - Frontier-level agentic and coding performance at Flash speed (1M context, 65K output)
+  - `gemini-3.8-flash` (aliases: `gemini-3.8`, `flash-3.8`) - Current-generation Flash with stronger long-horizon agentic performance (1M context, 65K output; thinking levels low/medium/high — no minimal)
   - `gemini-2.5-pro` (alias: `pro 2.5`) - Deep reasoning with thinking budget (1M context, 65K output)
   - `gemini-2.5-flash` (alias: `flash`) - Ultra-fast model with thinking budget (1M context, 65K output)
   - `gemini-2.5-flash-lite` (alias: `flash-lite`) - Lightweight fast model (1M context, 65K output)
@@ -129,8 +130,8 @@ This guide documents all supported AI providers in the Converse MCP Server and t
   2. Authenticate: run `agy` once interactively and complete the Google OAuth login. This also establishes workspace trust for your home directory (the provider spawns each call in a per-call subdirectory under `~/.converse/agy-runs`).
 - **Environment Variables**: None (the provider detects the `agy` binary on PATH or at the platform install location)
 - **Supported Models** (text-only — print mode has no image input channel):
-  - `gemini` (= `gemini:pro`) - Gemini 3.1 Pro
-  - `gemini:flash` - Gemini 3.5 Flash
+  - `gemini` (= `gemini:flash`) - Gemini 3.8 Flash (default)
+  - `gemini:pro` - Gemini 3.1 Pro
   - `reasoning_effort` selects the variant: `low` → (Low), `medium` → (Medium) for Flash / (High) for Pro, `high`/`max` → (High); unset defaults to (High)
 
 **Key Features:**
@@ -216,7 +217,7 @@ agy
   - `copilot` - Uses Copilot's default or env-configured model
   - OpenAI: `gpt-5.6-sol` (aliases: bare `gpt-5.6`, `gpt-5`), `gpt-5.6-terra` (recommended balanced tier), `gpt-5.6-luna`
   - Anthropic: `claude-fable-5` (alias: `fable`), `claude-sonnet-5` (alias: `sonnet`), `claude-opus-5` (aliases: `opus`, `claude`), `claude-opus-4.8`
-  - Google: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.5-flash` (alias: `gemini-flash`)
+  - Google: `gemini-3.1-pro-preview` (aliases: `gemini`, `gemini-3.1-pro`), `gemini-3.8-flash` (aliases: `gemini-3.8`, `flash-3.8`), `gemini-3.5-flash` (alias: `gemini-flash`)
 - **Reasoning**: The `gpt-5.6-sol`/`terra`/`luna` tiers accept `reasoning_effort`.
 - **Explicit pass-through**: Any other `copilot:<id>` model string is forwarded to the Copilot backend verbatim, so IDs outside the curated list still work while the backend accepts them.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "converse-mcp-server",
-  "version": "3.2.3",
+  "version": "3.3.0",
   "description": "Converse MCP Server - Converse with other LLMs with chat and consensus tools",
   "type": "module",
   "main": "src/index.js",
@@ -93,28 +93,28 @@
     ".env.example"
   ],
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.3.220",
-    "@anthropic-ai/sdk": "^0.115.0",
-    "@github/copilot-sdk": "^1.0.8",
-    "@google/genai": "^2.13.0",
-    "@lydell/node-pty": "1.2.0-beta.12",
-    "@mistralai/mistralai": "^2.5.0",
-    "@modelcontextprotocol/sdk": "^1.29.0",
-    "@openai/codex-sdk": "^0.145.0",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.258",
+    "@anthropic-ai/sdk": "^0.123.0",
+    "@github/copilot-sdk": "^1.0.11",
+    "@google/genai": "^2.20.0",
+    "@lydell/node-pty": "1.2.0-beta.15",
+    "@mistralai/mistralai": "^2.6.4",
+    "@modelcontextprotocol/sdk": "^1.30.0",
+    "@openai/codex-sdk": "^0.152.1",
     "cors": "^2.8.6",
     "dotenv": "^17.4.2",
     "express": "^5.2.1",
     "lru-cache": "^11.5.2",
-    "nanoid": "^6.0.0",
-    "openai": "^6.49.0",
-    "p-limit": "^7.3.1",
-    "vite": "^8.1.5"
+    "nanoid": "^6.0.1",
+    "openai": "^7.9.0",
+    "p-limit": "^7.3.2",
+    "vite": "^8.2.2"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^4.1.10",
+    "@vitest/coverage-v8": "^4.1.11",
     "cross-env": "^10.1.0",
-    "eslint": "^10.8.0",
+    "eslint": "^10.9.1",
     "rimraf": "^6.1.3",
-    "vitest": "^4.1.10"
+    "vitest": "^4.1.11"
   }
 }

package/src/async/jobRunner.js

@@ -35,7 +35,7 @@ export class JobRunner extends EventEmitter {
    * @param {object} dependencies.eventBus - EventBus instance (optional, uses global if not provided)
    * @param {object} options - Configuration options
    * @param {number} options.concurrency - Maximum concurrent jobs (default: 10)
-   * @param {number} options.defaultTimeout - Default job timeout in ms (default: 30 minutes)
+   * @param {number} options.defaultTimeout - Default job timeout in ms (default: 90 minutes)
    */
   constructor(dependencies, options = {}) {
     super();
@@ -54,7 +54,7 @@ export class JobRunner extends EventEmitter {
 
     // Configuration
     this.concurrency = options.concurrency || 10;
-    this.defaultTimeout = options.defaultTimeout || 30 * 60 * 1000; // 30 minutes
+    this.defaultTimeout = options.defaultTimeout || 90 * 60 * 1000; // 90 minutes
 
     // Create bounded concurrency limiter
     this.limiter = pLimit(this.concurrency);

package/src/config.js

@@ -78,8 +78,8 @@ const CONFIG_SCHEMA = {
     },
     HTTP_REQUEST_TIMEOUT: {
       type: 'number',
-      default: 300000,
-      description: 'HTTP request timeout in milliseconds (5 minutes)',
+      default: 900000,
+      description: 'HTTP request timeout in milliseconds (15 minutes)',
     },
     HTTP_MAX_REQUEST_SIZE: {
       type: 'string',
@@ -768,7 +768,7 @@ export function getHttpTransportConfig(config) {
     // Server settings
     port: transport.port || 3157,
     host: transport.host || 'localhost',
-    requestTimeout: transport.requesttimeout || 300000,
+    requestTimeout: transport.requesttimeout || 900000,
     maxRequestSize: transport.maxrequestsize || '10mb',
 
     // Session management

package/src/providers/anthropic.js

@@ -23,7 +23,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: false,
     supportsThinking: true,
     supportsAdaptiveThinking: true, // Adaptive thinking is the only thinking mode
-    timeout: 600000,
+    timeout: 1800000,
     supportsEffort: true,
     effortGA: true,
     supportsCompaction: true,
@@ -48,7 +48,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: false,
     supportsThinking: true,
     supportsAdaptiveThinking: true, // Thinking is on by default; adaptive is the only on-mode
-    timeout: 600000,
+    timeout: 1800000,
     supportsEffort: true,
     effortGA: true,
     supportsCompaction: true,
@@ -76,7 +76,7 @@ const SUPPORTED_MODELS = {
     supportsAdaptiveThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 128000,
-    timeout: 600000,
+    timeout: 1800000,
     supportsEffort: true,
     effortGA: true,
     supports1MContext: true,
@@ -106,7 +106,7 @@ const SUPPORTED_MODELS = {
     supportsAdaptiveThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 128000,
-    timeout: 600000,
+    timeout: 1800000,
     supportsEffort: true,
     effortGA: true,
     supports1MContext: true,
@@ -136,7 +136,7 @@ const SUPPORTED_MODELS = {
     supportsAdaptiveThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 128000,
-    timeout: 600000,
+    timeout: 1800000,
     supportsEffort: true,
     effortGA: true,
     supports1MContext: true,
@@ -165,7 +165,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 64000,
-    timeout: 300000,
+    timeout: 900000,
     supportsEffort: true, // Opus 4.5 effort parameter (requires beta header)
     description:
       'Claude Opus 4.5 - Previous most intelligent model combining maximum capability with practical performance',
@@ -191,7 +191,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 32000,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Claude Opus 4.1 - Highest level of intelligence and capability with extended thinking',
     aliases: [
@@ -220,7 +220,7 @@ const SUPPORTED_MODELS = {
     supportsAdaptiveThinking: true, // Sonnet 4.6: thinking: {type: "adaptive"} recommended
     minThinkingTokens: 1024,
     maxThinkingTokens: 64000,
-    timeout: 300000,
+    timeout: 900000,
     supportsEffort: true,
     effortGA: true, // Effort is generally available, no beta header required
     supports1MContext: true, // Beta 1M context support
@@ -251,7 +251,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 64000,
-    timeout: 300000,
+    timeout: 900000,
     supports1MContext: true, // Beta 1M context support
     deprecated: true,
     description:
@@ -275,7 +275,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     minThinkingTokens: 1024,
     maxThinkingTokens: 64000,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Claude Haiku 4.5 - Fast and intelligent model with extended thinking',
     aliases: [
@@ -598,8 +598,8 @@ export const anthropicProvider = {
     // Initialize Anthropic client
     const anthropic = new Anthropic({
       apiKey: config.apiKeys.anthropic,
-      // Increase timeout to 20 minutes for thinking models that may take longer
-      timeout: 20 * 60 * 1000,
+      // Increase timeout to 60 minutes for thinking models that may take longer
+      timeout: 60 * 60 * 1000,
     });
 
     // Build beta features array for the request

package/src/providers/claude.js

@@ -29,7 +29,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: true, // Supported via streaming input mode
     supportsWebSearch: false, // SDK accesses files directly, not web
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Claude Fable 5 via Agent SDK (default) - requires claude login authentication',
     aliases: [
@@ -50,7 +50,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: true, // Supported via streaming input mode
     supportsWebSearch: false, // SDK accesses files directly, not web
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Claude Opus 5 via Agent SDK - requires claude login authentication',
     aliases: ['claude:opus', 'claude-opus-5'],

package/src/providers/codex.js

@@ -32,7 +32,7 @@ const SUPPORTED_MODELS = {
     // still advertises the pre-5.6 set — the backend is the authority, so the
     // accepted tiers are declared per model and requests are clamped onto them.
     supportedEfforts: ['none', 'low', 'medium', 'high', 'xhigh'],
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'OpenAI Codex agentic coding assistant with local file access and tool execution (GPT-5.6)',
     aliases: [

package/src/providers/copilot.js

@@ -27,7 +27,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description:
 			'GitHub Copilot via CLI SDK - uses default or env-configured model',
     aliases: ['copilot-sdk', 'github-copilot'],
@@ -48,7 +48,7 @@ const SUPPORTED_MODELS = {
     supportsImages: false,
     supportsWebSearch: false,
     supportsReasoningEffort: true,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'OpenAI GPT-5.6 Sol via Copilot subscription',
     aliases: ['gpt-5.6', 'gpt-5', 'gpt', 'codex'],
   },
@@ -61,7 +61,7 @@ const SUPPORTED_MODELS = {
     supportsImages: false,
     supportsWebSearch: false,
     supportsReasoningEffort: true,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'OpenAI GPT-5.6 Terra via Copilot subscription',
     aliases: [],
   },
@@ -74,7 +74,7 @@ const SUPPORTED_MODELS = {
     supportsImages: false,
     supportsWebSearch: false,
     supportsReasoningEffort: true,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'OpenAI GPT-5.6 Luna via Copilot subscription',
     aliases: [],
   },
@@ -88,7 +88,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Anthropic Claude Fable 5 via Copilot subscription',
     aliases: ['fable'],
   },
@@ -100,7 +100,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Anthropic Claude Sonnet 5 via Copilot subscription',
     aliases: ['sonnet'],
   },
@@ -112,7 +112,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Anthropic Claude Opus 5 via Copilot subscription',
     aliases: ['opus', 'claude'],
   },
@@ -124,7 +124,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Anthropic Claude Opus 4.8 via Copilot subscription',
     aliases: [],
   },
@@ -139,7 +139,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Google Gemini 3.1 Pro Preview via Copilot subscription',
     aliases: ['gemini', 'gemini-3.1-pro'],
   },
@@ -151,7 +151,7 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: false,
     supportsWebSearch: false,
-    timeout: 600000,
+    timeout: 1800000,
     description: 'Google Gemini 3.5 Flash via Copilot subscription',
     aliases: ['gemini-flash'],
   },
@@ -553,7 +553,7 @@ async function checkReasoningSupport(client, modelId) {
 }
 
 async function* createStreamingGenerator(client, prompt, options, signal, config) {
-  const { model, timeout = 600000, reasoning_effort } = options;
+  const { model, timeout = 1800000, reasoning_effort } = options;
 
   const sessionModel = resolveSessionModel(model, config);
   const accessLevel = getToolAccessLevel(config);

package/src/providers/deepseek.js

@@ -29,7 +29,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     supportsJsonOutput: true,
     supportsFunctionCalling: true,
-    timeout: 600000, // Longer timeout for reasoning
+    timeout: 1800000, // Longer timeout for reasoning
     description:
       'DeepSeek V4 Pro - flagship MoE model with 1M context and thinking mode',
     aliases: ['deepseek', 'deepseek-pro'],
@@ -46,7 +46,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     supportsJsonOutput: true,
     supportsFunctionCalling: true,
-    timeout: 600000, // Longer timeout for reasoning
+    timeout: 1800000, // Longer timeout for reasoning
     description:
       'DeepSeek V4 Flash - faster, lower-cost V4 tier with 1M context and thinking mode',
     aliases: ['deepseek-flash'],

package/src/providers/gemini-cli.js

@@ -20,7 +20,9 @@
  *
  * The provider registry key remains 'gemini-cli' and the user-facing alias
  * remains 'gemini' for routing/normalization stability. Only three user-facing
- * model names are exposed: gemini (= gemini:pro), gemini:pro, gemini:flash.
+ * model names are exposed: gemini (= gemini:flash), gemini:flash, gemini:pro.
+ * Flash is the default: Gemini 3.8 Flash is the current-generation model agy
+ * lists first, while 3.1 Pro remains the only Pro tier Antigravity offers.
  */
 
 import { existsSync, mkdirSync, writeFileSync, rmSync } from 'node:fs';
@@ -37,7 +39,7 @@ import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 const ARGV_PROMPT_LIMIT = 24000;
 
 // Default print timeout (ms) when the tool layer passes none.
-const DEFAULT_TIMEOUT_MS = 600000;
+const DEFAULT_TIMEOUT_MS = 1800000;
 
 // Extra wall-clock grace before the JS-side hard kill fires (ms).
 const HARD_KILL_GRACE_MS = 15000;
@@ -57,7 +59,7 @@ const PTY_COLS = 1000;
 const SUPPORTED_MODELS = {
   gemini: {
     modelName: 'gemini',
-    friendlyName: 'Gemini 3.1 Pro (via Antigravity CLI)',
+    friendlyName: 'Gemini 3.8 Flash (via Antigravity CLI)',
     contextWindow: 1048576,
     maxOutputTokens: 65536,
     supportsStreaming: true,
@@ -66,14 +68,14 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     timeout: DEFAULT_TIMEOUT_MS,
     description:
-      'Gemini 3.1 Pro via Antigravity CLI (agy) - requires Antigravity Google OAuth login',
+      'Gemini 3.8 Flash via Antigravity CLI (agy) - requires Antigravity Google OAuth login',
     aliases: ['gemini-cli'],
     // agy display-name base; reasoning_effort selects the parenthesized variant
-    agyModelBase: 'Gemini 3.1 Pro',
+    agyModelBase: 'Gemini 3.8 Flash',
   },
-  'gemini:pro': {
-    modelName: 'gemini:pro',
-    friendlyName: 'Gemini 3.1 Pro (via Antigravity CLI)',
+  'gemini:flash': {
+    modelName: 'gemini:flash',
+    friendlyName: 'Gemini 3.8 Flash (via Antigravity CLI)',
     contextWindow: 1048576,
     maxOutputTokens: 65536,
     supportsStreaming: true,
@@ -82,13 +84,13 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     timeout: DEFAULT_TIMEOUT_MS,
     description:
-      'Gemini 3.1 Pro via Antigravity CLI (agy) - explicit alias of `gemini`',
-    aliases: [],
-    agyModelBase: 'Gemini 3.1 Pro',
+      'Gemini 3.8 Flash via Antigravity CLI (agy) - explicit alias of `gemini`',
+    aliases: ['flash'],
+    agyModelBase: 'Gemini 3.8 Flash',
   },
-  'gemini:flash': {
-    modelName: 'gemini:flash',
-    friendlyName: 'Gemini 3.5 Flash (via Antigravity CLI)',
+  'gemini:pro': {
+    modelName: 'gemini:pro',
+    friendlyName: 'Gemini 3.1 Pro (via Antigravity CLI)',
     contextWindow: 1048576,
     maxOutputTokens: 65536,
     supportsStreaming: true,
@@ -97,9 +99,9 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     timeout: DEFAULT_TIMEOUT_MS,
     description:
-      'Gemini 3.5 Flash via Antigravity CLI (agy) - requires Antigravity Google OAuth login',
-    aliases: ['flash'],
-    agyModelBase: 'Gemini 3.5 Flash',
+      'Gemini 3.1 Pro via Antigravity CLI (agy) - requires Antigravity Google OAuth login',
+    aliases: ['pro'],
+    agyModelBase: 'Gemini 3.1 Pro',
   },
 };
 
@@ -175,7 +177,7 @@ export function findAgyBinary() {
 /**
  * Map a reasoning_effort value to the agy parenthesized variant suffix.
  * Flash supports Low/Medium/High; Pro supports Low/High (no Medium).
- * @param {string} base - agy model base ('Gemini 3.5 Flash' / 'Gemini 3.1 Pro')
+ * @param {string} base - agy model base ('Gemini 3.8 Flash' / 'Gemini 3.1 Pro')
  * @param {string} [reasoningEffort]
  * @returns {string} e.g. '(Low)', '(Medium)', '(High)'
  */
@@ -207,7 +209,7 @@ function effortSuffix(base, reasoningEffort) {
  * verbatim so power users aren't blocked.
  * @param {string} model - e.g. 'gemini', 'gemini:flash', or a full agy name
  * @param {string} [reasoningEffort]
- * @returns {string} agy --model value, e.g. 'Gemini 3.1 Pro (High)'
+ * @returns {string} agy --model value, e.g. 'Gemini 3.8 Flash (High)'
  */
 export function resolveAgyModel(model, reasoningEffort) {
   const raw = typeof model === 'string' ? model.trim() : '';
@@ -230,11 +232,11 @@ export function resolveAgyModel(model, reasoningEffort) {
     !nameLower ||
     nameLower === 'gemini' ||
     nameLower === 'gemini-cli' ||
-    nameLower === 'pro'
+    nameLower === 'flash'
   ) {
     base = SUPPORTED_MODELS.gemini.agyModelBase;
-  } else if (nameLower === 'flash') {
-    base = SUPPORTED_MODELS['gemini:flash'].agyModelBase;
+  } else if (nameLower === 'pro') {
+    base = SUPPORTED_MODELS['gemini:pro'].agyModelBase;
   } else {
     // Unknown suffix: pass through verbatim (power-user agy display name)
     return raw;
@@ -762,19 +764,16 @@ export const geminiCliProvider = {
 
     const name = modelName.toLowerCase().trim();
 
-    // Full agy display-name passthrough → matching base config.
-    if (/gemini 3\.5 flash/i.test(modelName)) {
+    // Full agy display-name passthrough → matching tier config. Any 3.x Flash
+    // (agy also lists 3.6/3.7) shares the Flash config; any 3.x Pro the Pro one.
+    if (/gemini 3\.\d+ flash/i.test(modelName)) {
       return SUPPORTED_MODELS['gemini:flash'];
     }
-    if (/gemini 3\.1 pro/i.test(modelName)) {
-      return SUPPORTED_MODELS.gemini;
-    }
-
-    if (name === 'pro') {
+    if (/gemini 3\.\d+ pro/i.test(modelName)) {
       return SUPPORTED_MODELS['gemini:pro'];
     }
 
-    // Exact key match (gemini, gemini:pro, gemini:flash)
+    // Exact key match (gemini, gemini:flash, gemini:pro)
     if (SUPPORTED_MODELS[name]) {
       return SUPPORTED_MODELS[name];
     }

package/src/providers/google.js

@@ -20,7 +20,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     supportsWebSearch: true,
     maxThinkingTokens: 24576,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Ultra-fast (1M context) - Quick analysis, simple queries, rapid iterations with grounding',
     aliases: [
@@ -45,7 +45,7 @@ const SUPPORTED_MODELS = {
     supportsThinking: true,
     supportsWebSearch: true,
     maxThinkingTokens: 24576,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Lightweight fast model (1M context) - Efficient quick responses with grounding',
     aliases: [
@@ -68,7 +68,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsThinking: true,
     maxThinkingTokens: 32768,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Deep reasoning + thinking mode (1M context) - Complex problems, architecture, deep analysis',
     aliases: ['pro 2.5', 'gemini pro 2.5', 'gemini-2.5-pro-latest'],
@@ -84,7 +84,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     thinkingMode: 'level',
     thinkingLevels: ['minimal', 'low', 'medium', 'high'],
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Gemini 3.1 Pro - Most advanced reasoning with expanded thinking levels (1M context)',
     aliases: [
@@ -113,7 +113,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     thinkingMode: 'level',
     thinkingLevels: ['minimal', 'low', 'medium', 'high'],
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Gemini 3.5 Flash - Frontier-level agentic and coding performance at Flash speed (1M context)',
     aliases: [
@@ -127,6 +127,32 @@ const SUPPORTED_MODELS = {
       '3.5-flash',
     ],
   },
+  'gemini-3.8-flash': {
+    modelName: 'gemini-3.8-flash',
+    friendlyName: 'Gemini (Flash 3.8)',
+    contextWindow: 1048576, // 1M tokens
+    maxOutputTokens: 65536,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsThinking: true,
+    supportsWebSearch: true,
+    thinkingMode: 'level',
+    // 3.8 Flash rejects thinkingLevel "minimal" outright (API error), unlike 3.5.
+    thinkingLevels: ['low', 'medium', 'high'],
+    timeout: 900000,
+    description:
+      'Gemini 3.8 Flash - Current-generation Flash with stronger long-horizon agentic performance (1M context)',
+    aliases: [
+      'gemini-3.8',
+      'gemini3.8',
+      'gemini-3.8-flash-latest',
+      'flash-3.8',
+      'flash3.8',
+      'gemini-flash-3.8',
+      'gemini flash 3.8',
+      '3.8-flash',
+    ],
+  },
 };
 
 // Thinking mode budget percentages
@@ -514,10 +540,17 @@ export const googleProvider = {
           };
           thinkingLevel = levelMap[reasoning_effort] || 'high';
           if (!modelConfig.thinkingLevels.includes(thinkingLevel)) {
+            // Clamp to the nearest supported level: a request below the
+            // model's floor (e.g. minimal on 3.8 Flash) takes the lowest
+            // level, anything else the highest.
+            const rank = ['minimal', 'low', 'medium', 'high'];
+            const [lowest] = modelConfig.thinkingLevels;
             thinkingLevel =
-              modelConfig.thinkingLevels[
-                modelConfig.thinkingLevels.length - 1
-              ];
+              rank.indexOf(thinkingLevel) < rank.indexOf(lowest)
+                ? lowest
+                : modelConfig.thinkingLevels[
+                  modelConfig.thinkingLevels.length - 1
+                ];
           }
         } else {
           // Binary levels only (Gemini 3.0 Pro: low/high)

package/src/providers/mistral.js

@@ -21,7 +21,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: false,
     supportsReasoning: true,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Mistral Medium 3.5 - Frontier-class multimodal model with adjustable reasoning',
     aliases: [
@@ -40,7 +40,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: false,
     supportsReasoning: true,
-    timeout: 180000,
+    timeout: 540000,
     description:
       'Mistral Small 4 - Hybrid multimodal model unifying instruct, reasoning, and coding',
     aliases: ['mistral-small', 'mistral-small-latest'],
@@ -54,7 +54,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: false,
     supportsReasoning: false,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Mistral Large 3 - Open-weight MoE flagship (no adjustable reasoning)',
     aliases: ['mistral-large', 'mistral-large-latest'],

package/src/providers/openai.js

@@ -20,7 +20,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsNoneReasoningEffort: true,
-    timeout: 3600000, // 1 hour
+    timeout: 10800000, // 3 hours
     description:
       'Flagship GPT-5.6 model (1M context, 128K output) - Frontier reasoning, coding, agentic workflows. Most token-efficient flagship',
     aliases: [
@@ -45,7 +45,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsNoneReasoningEffort: true,
-    timeout: 1800000, // 30 minutes
+    timeout: 5400000, // 90 minutes
     description:
       'Lower-cost GPT-5.6 (400K context, 128K output) - Performance competitive with GPT-5.5 at half the flagship price',
     aliases: ['gpt5.6-terra', 'gpt-5.6terra', 'gpt 5.6 terra', 'terra'],
@@ -60,7 +60,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsNoneReasoningEffort: true,
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Fastest, most affordable GPT-5.6 (400K context, 128K output) - High-volume, latency-sensitive workloads',
     aliases: ['gpt5.6-luna', 'gpt-5.6luna', 'gpt 5.6 luna', 'luna'],
@@ -75,7 +75,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsNoneReasoningEffort: true,
-    timeout: 3600000, // 1 hour
+    timeout: 10800000, // 3 hours
     description:
       'Latest flagship model (1M context, 128K output) - Superior reasoning, coding, agentic workflows, computer use. Most token-efficient reasoning model',
     aliases: [
@@ -92,7 +92,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 1800000, // 30 minutes
+    timeout: 5400000, // 90 minutes
     description:
       'Faster, cost-efficient GPT-5 (400K context, 128K output) - Well-defined tasks, precise prompts',
     aliases: ['gpt5-mini', 'gpt-5mini', 'gpt 5 mini', 'gpt-5-mini-2025-08-07'],
@@ -106,7 +106,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: false, // GPT-5-nano doesn't support web search
     supportsResponsesAPI: true,
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Fastest, most cost-efficient GPT-5 (400K context, 128K output) - Summarization, classification',
     aliases: ['gpt5-nano', 'gpt-5nano', 'gpt 5 nano', 'gpt-5-nano-2025-08-07'],
@@ -120,7 +120,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 1800000, // 30 minutes
+    timeout: 5400000, // 90 minutes
     description:
       'Fast, efficient GPT-5.4 (400K context, 128K output) - Coding, subagents, computer use, tool use. 2x faster than GPT-5 mini',
     aliases: [
@@ -139,7 +139,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: false,
     supportsResponsesAPI: true,
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Smallest, cheapest GPT-5.4 (400K context, 128K output) - Classification, data extraction, ranking, coding subagents',
     aliases: [
@@ -159,7 +159,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsDeepResearch: false,
-    timeout: 3600000, // 60 minutes
+    timeout: 10800000, // 180 minutes
     description:
       'Maximum performance reasoning model (1M context, 272K output) - Most complex tasks, extended compute time (EXPENSIVE)',
     aliases: [
@@ -179,7 +179,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 600000, // 10 minutes
+    timeout: 1800000, // 30 minutes
     description:
       'Strong reasoning (200K context) - Logical problems, code generation, systematic analysis',
     aliases: ['o3-2025-01-31'],
@@ -193,7 +193,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 3600000, // 60 minutes
+    timeout: 10800000, // 180 minutes
     description:
       'Professional-grade reasoning (200K context) - EXTREMELY EXPENSIVE: Only for the most complex problems',
     aliases: ['o3-pro', 'o3pro', 'o3 pro'],
@@ -207,7 +207,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 180000, // 3 minutes
+    timeout: 540000, // 9 minutes
     description:
       'Latest reasoning model (200K context) - Optimized for shorter contexts, rapid reasoning',
     aliases: ['o4mini', 'o4', 'o4 mini', 'o4-mini-2025-01-30'],
@@ -221,7 +221,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsResponsesAPI: true,
-    timeout: 300000,
+    timeout: 900000,
     description:
       'GPT-4.1 (1M context) - Advanced reasoning model with large context window',
     aliases: ['gpt4.1', 'gpt-4.1', 'gpt 4.1', 'gpt-4.1-latest'],
@@ -236,7 +236,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsDeepResearch: true,
-    timeout: 7200000, // 120 minutes for deep research
+    timeout: 21600000, // 360 minutes for deep research
     description:
       'Deep research model (200K context) - In-depth synthesis, comprehensive reports, multi-source analysis (30-90 min runtime)',
     aliases: [
@@ -256,7 +256,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: true,
     supportsResponsesAPI: true,
     supportsDeepResearch: true,
-    timeout: 3600000, // 60 minutes for faster deep research
+    timeout: 10800000, // 180 minutes for faster deep research
     description:
       'Fast deep research model (200K context) - Lightweight research, faster results, latency-sensitive analysis (15-60 min runtime)',
     aliases: [

package/src/providers/openrouter-discovery.js

@@ -90,7 +90,7 @@ export function metadataToModelConfig(raw) {
     supportsReasoning,
     // Structured capability object consumed by the OpenRouter reasoning mapper.
     reasoning,
-    timeout: 300000,
+    timeout: 900000,
     isDynamic: true,
   };
 }

package/src/providers/openrouter.js

@@ -54,7 +54,7 @@ const SUPPORTED_MODELS = {
       supported_efforts: ['xhigh', 'high'],
       default_effort: 'high',
     },
-    timeout: 300000,
+    timeout: 900000,
     description:
       'Z.ai GLM 5.2 — large-scale reasoning model with a 1M-token context',
     aliases: ['glm-5.2', 'glm5.2', 'glm'],
@@ -74,7 +74,7 @@ const SUPPORTED_MODELS = {
       supported_efforts: ['xhigh', 'high'],
       default_effort: 'high',
     },
-    timeout: 300000,
+    timeout: 900000,
     description: 'DeepSeek V4 Pro reasoning model (via OpenRouter)',
     aliases: [],
   },
@@ -93,7 +93,7 @@ const SUPPORTED_MODELS = {
       supported_efforts: ['xhigh', 'high'],
       default_effort: 'high',
     },
-    timeout: 300000,
+    timeout: 900000,
     description: 'DeepSeek V4 Flash — faster, lower-cost DeepSeek V4 tier',
     aliases: [],
   },
@@ -108,7 +108,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     // Enable/disable-only — no effort tiers exposed.
     reasoning: { mandatory: false, default_enabled: true },
-    timeout: 300000,
+    timeout: 900000,
     description: 'Qwen3.7 Max — flagship Qwen with a 1M-token context',
     aliases: ['qwen3.7-max'],
   },
@@ -123,7 +123,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     // Enable/disable-only — no effort tiers exposed.
     reasoning: { mandatory: false, default_enabled: true },
-    timeout: 300000,
+    timeout: 900000,
     description: 'Qwen3.7 Plus — image-capable Qwen with a 1M-token context',
     aliases: ['qwen3.7-plus'],
   },
@@ -138,7 +138,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     // Mandatory reasoning — cannot be disabled.
     reasoning: { mandatory: true, default_enabled: true },
-    timeout: 300000,
+    timeout: 900000,
     description: 'Moonshot Kimi K2.7 Code — coding model with mandatory reasoning',
     aliases: ['kimi-k2.7-code'],
   },
@@ -153,7 +153,7 @@ const SUPPORTED_MODELS = {
     supportsReasoning: true,
     // Enable/disable-only — no effort tiers exposed.
     reasoning: { mandatory: false, default_enabled: true },
-    timeout: 300000,
+    timeout: 900000,
     description: 'Moonshot Kimi K2.6 — image-capable general model',
     aliases: ['kimi-k2.6'],
   },
@@ -169,7 +169,7 @@ const SUPPORTED_MODELS = {
     // Router selects the underlying model (and its effort) — do not fabricate a
     // reasoning field.
     reasoning: { passthrough: true },
-    timeout: 300000,
+    timeout: 900000,
     description: 'Auto-selects the best model for your prompt via OpenRouter',
     aliases: ['auto-router', 'openrouter-auto'],
   },
@@ -489,7 +489,7 @@ function createConservativeModelConfig(modelName) {
     maxOutputTokens: 8192,
     supportsStreaming: true,
     supportsWebSearch: false,
-    timeout: 300000,
+    timeout: 900000,
     isDynamic: true,
   };
 }

package/src/providers/xai.js

@@ -36,7 +36,7 @@ const SUPPORTED_MODELS = {
     supportsImages: true,
     supportsWebSearch: true,
     supportsReasoning: true,
-    timeout: 300000, // 5 minutes
+    timeout: 900000, // 15 minutes
     description:
       'Grok 4.5 (500K context) - Flagship X.AI model with image input, reasoning content, and native web/X search via Agent Tools',
     aliases: ['grok', 'grok-4.5', 'grok-4.5-latest', 'grok-build-latest'],

package/src/transport/httpTransport.js

@@ -28,7 +28,7 @@ export class HTTPTransportServer {
       // Server settings
       port: config.port || 3157,
       host: config.host || 'localhost',
-      requestTimeout: config.requestTimeout || 300000,
+      requestTimeout: config.requestTimeout || 900000,
       maxRequestSize: config.maxRequestSize || '10mb',
 
       // Session management