@starlink-awaken/agentmesh 1.2.8 → 1.3.2

package/config/gateway.yaml

@@ -1,268 +1,230 @@
-port: 3000
-wsPort: 3001
-host: "0.0.0.0"
-dataDir: "./data"
-logDir: "./logs"
-
-routing:
-  defaultAgent: "claude-code"
-
-  rules:
-    # Claude Code - 通用代码任务
-    - name: code-generation
-      keywords:
-        - write code
-        - generate code
-        - create function
-        - create class
-        - 写代码
-        - 生成代码
-      agent: claude-code
-      priority: 10
-
-    - name: code-review
-      keywords:
-        - review
-        - code review
-        - pr review
-        - review code
-        - 代码审查
-        - review code
-      agent: claude-code
-      priority: 15
-
-    - name: debugging
-      keywords:
-        - debug
-        - fix bug
-        - error
-        - bug
-        - 调试
-        - 修复错误
-      agent: claude-code
-      priority: 15
-
-    - name: refactoring
-      keywords:
-        - refactor
-        - 重构
-        - improve code
-      agent: claude-code
-      priority: 10
-
-    - name: documentation
-      keywords:
-        - docs
-        - document
-        - 文档
-        - write docs
-      agent: claude-code
-      priority: 10
-
-    # OpenClaw - 浏览器自动化
-    - name: browser-automation
-      keywords:
-        - browser
-        - scrape
-        - click
-        - screenshot
-        - web automation
-        - 浏览器
-        - 爬虫
-        - 截图
-      agent: openclaw
-      priority: 15
-
-    - name: web-scraping
-      keywords:
-        - scrap
-        - crawl
-        - extract data
-        - 抓取
-        - 采集
-      agent: openclaw
-      priority: 15
-
-    # Cursor - AI 编程
-    - name: cursor-task
-      keywords:
-        - cursor
-        - cursor task
-      agent: cursor
-      priority: 12
-
-    # Windsurf - Flow 状态编程
-    - name: windsurf-task
-      keywords:
-        - windsurf
-        - flow state
-      agent: windsurf
-      priority: 12
-
-    # Qwen Code - 中文编程
-    - name: qwen-task
-      keywords:
-        - qwen
-        - 通义千问
-      agent: qwen-code
-      priority: 12
-
-    # Gemini - 多模态
-    - name: gemini-task
-      keywords:
-        - gemini
-        - multimodal
-        - 多模态
-      agent: gemini
-      priority: 12
-
-    # Droid - Android 开发
-    - name: android-development
-      keywords:
-        - android
-        - apk
-        - mobile
-        - 安卓
-        - 手机应用
-      agent: droid
-      priority: 15
-
-    # Aider - Git 集成编辑
-    - name: aider-task
-      keywords:
-        - aider
-        - git edit
-        - refactor git
-      agent: aider
-      priority: 12
-
-    # Cline - 自主编程
-    - name: cline-task
-      keywords:
-        - cline
-        - autonomous
-      agent: cline
-      priority: 12
-
-    # Roo Code
-    - name: roo-code-task
-      keywords:
-        - roo-code
-        - roo
-      agent: roo-code
-      priority: 12
-
-    # 多 Agent 协作
-    - name: multi-agent
-      keywords:
-        - collaborate
-        - team
-        - together
-        - multiple agents
-        - 协作
-        - 多个
-      strategy: broadcast
-      agents:
-        - claude-code
-        - openclaw
-      priority: 20
-
-# =============================================================================
-# 模型网关配置 — 多 Provider 路由 + 配额感知 Fallback
-# =============================================================================
+agents:
+- capabilities:
+  - code-generation
+  - code-review
+  - debugging
+  - refactoring
+  - documentation
+  - file-operations
+  id: claude-code
+  name: Claude Code
+  type: claude-code
+- capabilities:
+  - browser-automation
+  - web-scraping
+  - form-filling
+  - ui-testing
+  id: openclaw
+  name: OpenClaw
+  type: openclaw
+dataDir: ./data
+host: 0.0.0.0
+logDir: ./logs
 models:
   default_model: deepseek-chat
-
-  # 默认配置（各 Provider 可覆盖）
   defaults:
     circuit_breaker:
       failure_threshold: 3
-      reset_timeout_ms: 30000
       half_open_max_requests: 1
+      reset_timeout_ms: 30000
     retry:
-      max_retries: 3
       base_delay_ms: 500
       max_delay_ms: 10000
-      retryable_statuses: [429, 500, 502, 503, 504]
-
+      max_retries: 3
+      retryable_statuses:
+      - 429
+      - 500
+      - 502
+      - 503
+      - 504
+  fallback_chain:
+  - deepseek
+  - openrouter
+  - ollama
+  model_routing:
+    claude:
+    - openrouter
+    codestral:
+    - ollama
+    deepseek:
+    - deepseek
+    gemini:
+    - openrouter
+    gpt-:
+    - openai
+    - deepseek
+    gpt-5.3-codex:
+    - deepseek
+    llama:
+    - ollama
+    o1:
+    - openai
+    - deepseek
+    o4:
+    - openai
+    - deepseek
+    qwen:
+    - ollama
   providers:
     deepseek:
-      base_url: https://api.deepseek.com/v1
       api_key_env: DEEPSEEK_API_KEY
+      base_url: https://api.deepseek.com/v1
+      models:
+      - deepseek-chat
+      - deepseek-reasoner
+      - deepseek-v4-pro
+      - deepseek-v4-flash
+    ollama:
+      api_key: ollama
+      base_url: http://127.0.0.1:11434/v1
       models:
-        - deepseek-chat
-        - deepseek-reasoner
-        - deepseek-v4-pro
-        - deepseek-v4-flash
+      - qwen3:14b
+      - codestral:22b
+      - llama3.1:8b
     openai:
-      base_url: https://api.openai.com/v1
       api_key_env: OPENAI_API_KEY
+      base_url: https://api.openai.com/v1
       models:
-        - gpt-5.1
-        - gpt-5.1-codex
-        - o4-mini
+      - gpt-5.1
+      - gpt-5.1-codex
+      - o4-mini
     openrouter:
-      base_url: https://openrouter.ai/api/v1
       api_key_env: OPENROUTER_API_KEY
-    ollama:
-      base_url: http://127.0.0.1:11434/v1
-      api_key: ollama
-      models:
-        - qwen3:14b
-        - codestral:22b
-        - llama3.1:8b
-
-  # Fallback 链：主 Provider 不可用时依次尝试
-  fallback_chain:
-    - deepseek
-    - openrouter
-    - ollama
-
-  # 模型名 → 优先 Provider 列表
-  model_routing:
-    "deepseek":
-      - deepseek
-    "gpt-":
-      - openai
-      - deepseek
-    "o1":
-      - openai
-      - deepseek
-    "o4":
-      - openai
-      - deepseek
-    "claude":
-      - openrouter
-    "gemini":
-      - openrouter
-    "qwen":
-      - ollama
-    "codestral":
-      - ollama
-    "llama":
-      - ollama
-
-# =============================================================================
-# Agent 配置
-# =============================================================================
-agents:
-  # 内置 Agent（在代码中注册）
-  - id: claude-code
-    name: Claude Code
-    type: claude-code
-    capabilities:
-      - code-generation
-      - code-review
-      - debugging
-      - refactoring
-      - documentation
-      - file-operations
-
-  - id: openclaw
-    name: OpenClaw
-    type: openclaw
-    capabilities:
-      - browser-automation
-      - web-scraping
-      - form-filling
-      - ui-testing
+      base_url: https://openrouter.ai/api/v1
+port: 3000
+routing:
+  defaultAgent: claude-code
+  rules:
+  - agent: claude-code
+    keywords:
+    - write code
+    - generate code
+    - create function
+    - create class
+    - 写代码
+    - 生成代码
+    name: code-generation
+    priority: 10
+  - agent: claude-code
+    keywords:
+    - review
+    - code review
+    - pr review
+    - review code
+    - 代码审查
+    - review code
+    name: code-review
+    priority: 15
+  - agent: claude-code
+    keywords:
+    - debug
+    - fix bug
+    - error
+    - bug
+    - 调试
+    - 修复错误
+    name: debugging
+    priority: 15
+  - agent: claude-code
+    keywords:
+    - refactor
+    - 重构
+    - improve code
+    name: refactoring
+    priority: 10
+  - agent: claude-code
+    keywords:
+    - docs
+    - document
+    - 文档
+    - write docs
+    name: documentation
+    priority: 10
+  - agent: openclaw
+    keywords:
+    - browser
+    - scrape
+    - click
+    - screenshot
+    - web automation
+    - 浏览器
+    - 爬虫
+    - 截图
+    name: browser-automation
+    priority: 15
+  - agent: openclaw
+    keywords:
+    - scrap
+    - crawl
+    - extract data
+    - 抓取
+    - 采集
+    name: web-scraping
+    priority: 15
+  - agent: cursor
+    keywords:
+    - cursor
+    - cursor task
+    name: cursor-task
+    priority: 12
+  - agent: windsurf
+    keywords:
+    - windsurf
+    - flow state
+    name: windsurf-task
+    priority: 12
+  - agent: qwen-code
+    keywords:
+    - qwen
+    - 通义千问
+    name: qwen-task
+    priority: 12
+  - agent: gemini
+    keywords:
+    - gemini
+    - multimodal
+    - 多模态
+    name: gemini-task
+    priority: 12
+  - agent: droid
+    keywords:
+    - android
+    - apk
+    - mobile
+    - 安卓
+    - 手机应用
+    name: android-development
+    priority: 15
+  - agent: aider
+    keywords:
+    - aider
+    - git edit
+    - refactor git
+    name: aider-task
+    priority: 12
+  - agent: cline
+    keywords:
+    - cline
+    - autonomous
+    name: cline-task
+    priority: 12
+  - agent: roo-code
+    keywords:
+    - roo-code
+    - roo
+    name: roo-code-task
+    priority: 12
+  - agents:
+    - claude-code
+    - openclaw
+    keywords:
+    - collaborate
+    - team
+    - together
+    - multiple agents
+    - 协作
+    - 多个
+    name: multi-agent
+    priority: 20
+    strategy: broadcast
+wsPort: 3001

package/dist/src/cli.js

@@ -7,7 +7,7 @@ import { existsSync, readFileSync } from 'node:fs';
 import { resolve, dirname, join } from 'node:path';
 import { initLogger } from './core/logger.js';
 const PROJECT_ROOT = resolve(dirname(import.meta.dir), '..');
-const VERSION = '1.2.8';
+const VERSION = '1.3.2';
 const BANNER = `
    █████╗  ██████╗ ███████╗███╗   ██╗████████╗
   ██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝
@@ -250,6 +250,42 @@ async function cmdConfig(args) {
         }
     }
 }
+async function cmdStatus() {
+    try {
+        const [health, models, stats] = await Promise.all([
+            apiRequest('/health'),
+            apiRequest('/v1/models').catch(() => ({ data: [] })),
+            apiRequest('/model-gateway/stats').catch(() => null),
+        ]);
+        console.log(`
+╔═══════════════════════════════════════════════════╗
+║     Agent Mesh Gateway Status                    ║
+╠═══════════════════════════════════════════════════╣
+║  Status:    ${health.status}                                      ║
+║  Agents:    ${String(health.agents?.length || 0).padStart(2)} online                               ║
+║  Models:    ${String(models.data?.length || 0).padStart(2)} available                            ║
+║  Uptime:    ${stats?.uptime_seconds ? Math.floor(stats.uptime_seconds) + 's' : 'N/A'}                                   ║
+╠═══════════════════════════════════════════════════╣`);
+        if (stats?.providers) {
+            console.log('║  Provider Metrics:                              ║');
+            for (const [name, m] of Object.entries(stats.providers)) {
+                console.log(`║  ${name.padEnd(12)} reqs:${String(m.requests).padStart(5)}  ok:${(m.success_rate || 'N/A').padStart(6)}  avg:${String(m.avg_latency_ms || 0).padStart(4)}ms     ║`);
+            }
+        }
+        if (stats?.recent?.length) {
+            console.log('╠═══════════════════════════════════════════════════╣');
+            console.log('║  Recent:                                        ║');
+            for (const r of stats.recent.slice(0, 5)) {
+                const time = new Date(r.time).toLocaleTimeString();
+                console.log(`║  ${r.status >= 400 ? '❌' : '✅'} ${time} ${r.model} → ${r.actual}  ${r.latency_ms}ms              ║`);
+            }
+        }
+        console.log('╚═══════════════════════════════════════════════════╝\n');
+    }
+    catch {
+        console.error('\n  ❌ Gateway not reachable. Start: agentmesh start\n');
+    }
+}
 async function cmdDoctor() {
     console.log('\n  🔍 Agent Mesh Gateway Diagnostics\n');
     const checks = [];
@@ -326,9 +362,12 @@ async function main() {
                 await runSetup();
                 break;
             case 'health':
-            case 'status':
                 await cmdHealth();
                 break;
+            case 'status':
+            case 'info':
+                await cmdStatus();
+                break;
             case 'models':
             case 'model':
                 await cmdModels();

package/dist/src/model-gateway/metrics.d.ts

@@ -0,0 +1,28 @@
+interface RequestLog {
+    timestamp: number;
+    model: string;
+    provider: string;
+    actualModel: string;
+    latencyMs: number;
+    status: number;
+    error?: string;
+    streaming: boolean;
+}
+export declare function recordRequest(log: RequestLog): void;
+export declare function getMetrics(): {
+    uptime_seconds: number;
+    total_requests: number;
+    total_failures: number;
+    providers: Record<string, any>;
+    recent: {
+        time: string;
+        model: string;
+        provider: string;
+        actual: string;
+        latency_ms: number;
+        status: number;
+        streaming: boolean;
+        error: string | undefined;
+    }[];
+};
+export {};

package/dist/src/model-gateway/metrics.js

@@ -0,0 +1,60 @@
+// 网关运行时指标收集
+const providerMetrics = new Map();
+const recentRequests = [];
+const MAX_RECENT = 200;
+function getOrInitProvider(name) {
+    if (!providerMetrics.has(name)) {
+        providerMetrics.set(name, { requests: 0, success: 0, failures: 0, totalLatencyMs: 0 });
+    }
+    return providerMetrics.get(name);
+}
+export function recordRequest(log) {
+    const m = getOrInitProvider(log.provider);
+    m.requests++;
+    m.totalLatencyMs += log.latencyMs;
+    if (log.status >= 200 && log.status < 400) {
+        m.success++;
+        m.lastSuccessTime = log.timestamp;
+    }
+    else {
+        m.failures++;
+        m.lastError = log.error;
+        m.lastErrorTime = log.timestamp;
+    }
+    recentRequests.unshift(log);
+    if (recentRequests.length > MAX_RECENT)
+        recentRequests.pop();
+}
+export function getMetrics() {
+    const providers = {};
+    let totalRequests = 0;
+    let totalFailures = 0;
+    for (const [name, m] of providerMetrics) {
+        totalRequests += m.requests;
+        totalFailures += m.failures;
+        providers[name] = {
+            requests: m.requests,
+            success_rate: m.requests > 0 ? ((m.success / m.requests) * 100).toFixed(1) + '%' : 'N/A',
+            avg_latency_ms: m.requests > 0 ? Math.round(m.totalLatencyMs / m.requests) : 0,
+            last_success: m.lastSuccessTime ? new Date(m.lastSuccessTime).toISOString() : null,
+            last_error: m.lastError || null,
+            last_error_time: m.lastErrorTime ? new Date(m.lastErrorTime).toISOString() : null,
+        };
+    }
+    return {
+        uptime_seconds: Math.round(process.uptime()),
+        total_requests: totalRequests,
+        total_failures: totalFailures,
+        providers,
+        recent: recentRequests.slice(0, 20).map(r => ({
+            time: new Date(r.timestamp).toISOString(),
+            model: r.model,
+            provider: r.provider,
+            actual: r.actualModel,
+            latency_ms: r.latencyMs,
+            status: r.status,
+            streaming: r.streaming,
+            error: r.error,
+        })),
+    };
+}

package/dist/src/model-gateway/providers.d.ts

@@ -1,4 +1,3 @@
 import type { ChatCompletionRequest, ResolvedProvider } from './types.js';
 export declare function callChatCompletions(provider: ResolvedProvider, request: ChatCompletionRequest): Promise<Response>;
 export declare function callResponsesApi(provider: ResolvedProvider, body: Record<string, any>): Promise<Response>;
-export declare function buildStreamingResponse(upstreamResp: Response): Response;

package/dist/src/model-gateway/providers.js

@@ -1,6 +1,5 @@
 import { circuitBreakerRegistry } from './circuit-breaker.js';
 import { withRetry, isRetryable } from './retry.js';
-// 所有目标 Provider 都兼容 OpenAI API 格式，统一客户端即可
 export async function callChatCompletions(provider, request) {
     const { base_url, api_key, name: providerName } = provider;
     const { model, messages, stream, temperature, max_tokens, tools, tool_choice } = request;
@@ -19,22 +18,18 @@ export async function callChatCompletions(provider, request) {
         'Content-Type': 'application/json',
         Authorization: `Bearer ${api_key}`,
     };
-    // OpenRouter 需要额外的头部
     if (providerName === 'openrouter') {
         headers['HTTP-Referer'] = 'http://127.0.0.1:3000';
         headers['X-Title'] = 'Agent Mesh Gateway';
     }
     const url = `${base_url.replace(/\/$/, '')}/chat/completions`;
-    // 熔断器检查
     if (!circuitBreakerRegistry.canRequest(providerName)) {
         throw new Error(`Circuit breaker open for ${providerName}`);
     }
     try {
         const resp = await withRetry(providerName, async () => {
             const r = await fetch(url, {
-                method: 'POST',
-                headers,
-                body: JSON.stringify(body),
+                method: 'POST', headers, body: JSON.stringify(body),
                 signal: AbortSignal.timeout(120_000),
             });
             return r;
@@ -42,7 +37,6 @@ export async function callChatCompletions(provider, request) {
             console.warn(`[Retry] ${providerName} attempt ${attempt} after ${status} — retrying in ${delayMs}ms`);
         });
         if (!resp.ok && isRetryable(resp.status)) {
-            // Retry logic already handled in withRetry, but if we get here after max retries:
             circuitBreakerRegistry.recordFailure(providerName);
         }
         else if (resp.ok) {
@@ -58,38 +52,260 @@ export async function callChatCompletions(provider, request) {
         throw err;
     }
 }
+// ============================================================================
+// Responses API → Chat Completions 双向转换（含 tool_calls 往返）
+// ============================================================================
 export async function callResponsesApi(provider, body) {
-    // Responses API → Chat Completions 转换
-    const messages = convertResponsesInputToMessages(body.input || []);
+    const messages = convertInputToMessages(body.input || []);
     if (body.instructions) {
         messages.unshift({ role: 'system', content: body.instructions });
     }
-    return callChatCompletions(provider, {
+    // 转换 tools 定义（Codex 的 tool schema → OpenAI format）
+    const tools = convertToolSchemas(body.tools);
+    const chatResp = await callChatCompletions(provider, {
         model: body.model,
         messages,
         stream: body.stream,
-        tools: body.tools,
+        tools,
+        tool_choice: body.tool_choice,
+    });
+    // 非流式：直接转换响应
+    if (!body.stream) {
+        const ccData = (await chatResp.json());
+        return new Response(JSON.stringify(convertChatToResponses(ccData)), {
+            status: 200,
+            headers: { 'Content-Type': 'application/json' },
+        });
+    }
+    // 流式：解析 SSE → 转换 → 重新打包 SSE
+    const transformed = transformSSEStream(chatResp.body);
+    return new Response(transformed, {
+        status: 200,
+        headers: { 'Content-Type': 'text/event-stream', 'Cache-Control': 'no-cache' },
     });
 }
-function convertResponsesInputToMessages(input) {
+// ============================================================================
+// 输入转换: Responses input[] → Chat messages[]
+// ============================================================================
+function convertInputToMessages(input) {
     const messages = [];
     for (const item of input) {
-        if (item.role === 'system') {
-            messages.push({ role: 'system', content: extractTextContent(item.content) });
-        }
-        else if (item.role === 'user') {
-            messages.push({ role: 'user', content: extractTextContent(item.content) });
-        }
-        else if (item.role === 'assistant') {
-            messages.push({ role: 'assistant', content: extractTextContent(item.content) });
-        }
-        else if (item.type === 'message') {
-            const role = item.role || 'user';
-            messages.push({ role, content: extractTextContent(item.content) });
+        switch (item.type || item.role) {
+            // 标准消息类型
+            case 'message':
+                messages.push({ role: item.role || 'user', content: extractTextContent(item.content) });
+                break;
+            // Function call（Assistant 侧发起工具调用）
+            case 'function_call': {
+                messages.push({
+                    role: 'assistant',
+                    content: null,
+                    tool_calls: [{
+                            id: item.call_id,
+                            type: 'function',
+                            function: {
+                                name: item.name,
+                                arguments: typeof item.arguments === 'string' ? item.arguments : JSON.stringify(item.arguments),
+                            },
+                        }],
+                });
+                break;
+            }
+            // Function call output（Tool 返回结果）
+            case 'function_call_output':
+                messages.push({
+                    role: 'tool',
+                    tool_call_id: item.call_id,
+                    content: typeof item.output === 'string' ? item.output : JSON.stringify(item.output),
+                });
+                break;
+            // 简单角色
+            case 'system':
+                messages.push({ role: 'system', content: extractTextContent(item.content) });
+                break;
+            case 'user':
+                messages.push({ role: 'user', content: extractTextContent(item.content) });
+                break;
+            case 'assistant':
+                messages.push({ role: 'assistant', content: extractTextContent(item.content) });
+                break;
+            default:
+                // 回退: role 字段
+                if (item.role) {
+                    messages.push({ role: item.role, content: extractTextContent(item.content) });
+                }
         }
     }
     return messages;
 }
+// ============================================================================
+// 输出转换: Chat completions response → Responses API response
+// ============================================================================
+function convertChatToResponses(ccData) {
+    const choice = ccData.choices?.[0];
+    if (!choice) {
+        return { id: ccData.id, object: 'response', model: ccData.model, output: [], usage: ccData.usage };
+    }
+    const output = [];
+    // 文本回复
+    if (choice.message?.content) {
+        output.push({
+            type: 'message',
+            role: 'assistant',
+            content: [{ type: 'output_text', text: choice.message.content }],
+        });
+    }
+    // 工具调用
+    if (choice.message?.tool_calls) {
+        for (const tc of choice.message.tool_calls) {
+            output.push({
+                type: 'function_call',
+                call_id: tc.id,
+                name: tc.function?.name,
+                arguments: tc.function?.arguments,
+            });
+        }
+    }
+    // finish_reason = 'tool_calls' 表示等待 tool 结果（Codex 需要此信息）
+    const status = choice.finish_reason === 'tool_calls' ? 'requires_action' : 'completed';
+    return {
+        id: ccData.id,
+        object: 'response',
+        model: ccData.model,
+        status,
+        output,
+        usage: ccData.usage,
+    };
+}
+// ============================================================================
+// SSE 流式转换: Chat SSE → Responses SSE
+// ============================================================================
+function transformSSEStream(upstreamBody) {
+    const encoder = new TextEncoder();
+    let buffer = '';
+    let responseId = '';
+    let modelName = '';
+    let contentBuffer = '';
+    let toolCallAccum = {};
+    return new ReadableStream({
+        async start(controller) {
+            const reader = upstreamBody.getReader();
+            const decoder = new TextDecoder();
+            try {
+                while (true) {
+                    const { done, value } = await reader.read();
+                    if (done)
+                        break;
+                    buffer += decoder.decode(value, { stream: true });
+                    const lines = buffer.split('\n');
+                    buffer = lines.pop() || '';
+                    for (const line of lines) {
+                        if (!line.startsWith('data: '))
+                            continue;
+                        const data = line.slice(6).trim();
+                        if (data === '[DONE]') {
+                            // 发送最终事件
+                            const finalEvt = buildResponseEvent(responseId, modelName, contentBuffer, toolCallAccum, true);
+                            controller.enqueue(encoder.encode(finalEvt));
+                            controller.enqueue(encoder.encode('data: [DONE]\n\n'));
+                            continue;
+                        }
+                        try {
+                            const chunk = JSON.parse(data);
+                            responseId = chunk.id || responseId;
+                            modelName = chunk.model || modelName;
+                            const delta = chunk.choices?.[0]?.delta;
+                            if (!delta)
+                                continue;
+                            // 文本增量
+                            if (delta.content) {
+                                contentBuffer += delta.content;
+                                const evt = `data: ${JSON.stringify({
+                                    type: 'response.output_text.delta',
+                                    delta: delta.content,
+                                })}\n\n`;
+                                controller.enqueue(encoder.encode(evt));
+                            }
+                            // 工具调用增量
+                            if (delta.tool_calls) {
+                                for (const tc of delta.tool_calls) {
+                                    if (!toolCallAccum[tc.index]) {
+                                        toolCallAccum[tc.index] = {
+                                            id: tc.id || '',
+                                            type: 'function',
+                                            function: { name: tc.function?.name || '', arguments: '' },
+                                        };
+                                    }
+                                    if (tc.function?.arguments) {
+                                        toolCallAccum[tc.index].function.arguments += tc.function.arguments;
+                                    }
+                                    if (tc.id)
+                                        toolCallAccum[tc.index].id = tc.id;
+                                    if (tc.function?.name)
+                                        toolCallAccum[tc.index].function.name = tc.function.name;
+                                }
+                                // 发送 tool_call delta 事件
+                                const tcEvt = `data: ${JSON.stringify({
+                                    type: 'response.function_call_arguments.delta',
+                                    tool_calls: Object.values(toolCallAccum),
+                                })}\n\n`;
+                                controller.enqueue(encoder.encode(tcEvt));
+                            }
+                        }
+                        catch {
+                            // 非 JSON 行直接透传
+                            controller.enqueue(encoder.encode(line + '\n'));
+                        }
+                    }
+                }
+            }
+            catch (err) {
+                controller.error(err);
+            }
+            finally {
+                reader.releaseLock();
+                controller.close();
+            }
+        },
+    });
+}
+function buildResponseEvent(id, model, content, toolCalls, isFinal) {
+    const output = [];
+    if (content) {
+        output.push({ type: 'message', role: 'assistant', content: [{ type: 'output_text', text: content }] });
+    }
+    for (const tc of Object.values(toolCalls)) {
+        output.push({
+            type: 'function_call',
+            call_id: tc.id,
+            name: tc.function.name,
+            arguments: tc.function.arguments,
+        });
+    }
+    return `data: ${JSON.stringify({
+        type: isFinal ? 'response.completed' : 'response.output_text.delta',
+        response: isFinal ? { id, model, object: 'response', status: 'completed', output } : undefined,
+        delta: isFinal ? undefined : content,
+    })}\n\n`;
+}
+// ============================================================================
+// 工具定义转换
+// ============================================================================
+function convertToolSchemas(tools) {
+    if (!tools || !Array.isArray(tools))
+        return undefined;
+    return tools.map(t => ({
+        type: 'function',
+        function: {
+            name: t.name,
+            description: t.description,
+            parameters: t.parameters || t.input_schema,
+        },
+    }));
+}
+// ============================================================================
+// 辅助函数
+// ============================================================================
 function extractTextContent(content) {
     if (typeof content === 'string')
         return content;
@@ -101,13 +317,3 @@ function extractTextContent(content) {
     }
     return String(content || '');
 }
-export function buildStreamingResponse(upstreamResp) {
-    return new Response(upstreamResp.body, {
-        status: upstreamResp.status,
-        headers: {
-            'Content-Type': 'text/event-stream',
-            'Cache-Control': 'no-cache',
-            Connection: 'keep-alive',
-        },
-    });
-}

package/dist/src/model-gateway/router.d.ts

@@ -2,3 +2,4 @@ import type { ModelGatewayConfig, ResolvedProvider } from './types.js';
 export declare function initModelRouter(cfg: ModelGatewayConfig): void;
 export declare function getConfig(): ModelGatewayConfig;
 export declare function resolveProvider(model: string): ResolvedProvider | null;
+export declare function remapModel(model: string, providerName: string): string;

package/dist/src/model-gateway/router.js

@@ -1,8 +1,25 @@
 import { isProviderAvailable } from './quota.js';
 import { circuitBreakerRegistry } from './circuit-breaker.js';
+// 模型名重映射：对外模型名 → 实际 Provider 的模型名（可从 config 覆盖）
+let modelAliases = {
+    deepseek: {
+        'gpt-5.3-codex': 'deepseek-v4-pro',
+        'gpt-5.4': 'deepseek-v4-pro',
+        'gpt-5.5': 'deepseek-v4-pro',
+        'o4-mini': 'deepseek-v4-flash',
+        'claude-sonnet-4-6': 'deepseek-v4-pro',
+    },
+};
 let config;
 export function initModelRouter(cfg) {
     config = cfg;
+    // 从 config 加载模型别名（覆盖默认）
+    if (cfg.model_aliases) {
+        for (const [key, val] of Object.entries(cfg.model_aliases)) {
+            modelAliases.deepseek = modelAliases.deepseek || {};
+            modelAliases.deepseek[key] = val;
+        }
+    }
 }
 export function getConfig() {
     return config;
@@ -14,7 +31,7 @@ export function resolveProvider(model) {
     // 1. 按 model_routing 配置查找
     const routingEntries = Object.entries(config.model_routing);
     for (const [pattern, providers] of routingEntries) {
-        if (model.includes(pattern)) {
+        if (model.startsWith(pattern)) {
             for (const providerName of providers) {
                 const providerCfg = config.providers[providerName];
                 if (!providerCfg)
@@ -68,6 +85,9 @@ export function resolveProvider(model) {
     }
     return null;
 }
+export function remapModel(model, providerName) {
+    return modelAliases[providerName]?.[model] || model;
+}
 function resolveApiKey(_name, providerCfg) {
     if (providerCfg.api_key && providerCfg.api_key !== '') {
         return providerCfg.api_key;

package/dist/src/model-gateway/routes.js

@@ -1,8 +1,9 @@
-import { resolveProvider, getConfig } from './router.js';
+import { resolveProvider, getConfig, remapModel } from './router.js';
 import { callChatCompletions, callResponsesApi } from './providers.js';
 import { getQuotaSummary, probeQuota } from './quota.js';
 import { circuitBreakerRegistry } from './circuit-breaker.js';
 import { checkAllProviders } from './health.js';
+import { getMetrics, recordRequest } from './metrics.js';
 export async function modelGatewayRoutes(fastify) {
     // 健康检查 + 配额总览
     fastify.get('/model-gateway/health', async (_req, _reply) => {
@@ -44,14 +45,16 @@ export async function modelGatewayRoutes(fastify) {
         if (!body || !body.messages) {
             return reply.code(400).send({ error: { message: 'messages is required' } });
         }
-        const model = body.model || 'deepseek-chat';
-        const provider = resolveProvider(model);
+        const originalModel = body.model || 'deepseek-chat';
+        const provider = resolveProvider(originalModel);
         if (!provider) {
             return reply.code(503).send({
                 error: { message: 'No available provider. Check API keys and quota.' },
             });
         }
-        console.log(`[ModelGW] ${model} → ${provider.name} (${body.stream ? 'stream' : 'sync'})`);
+        const model = remapModel(originalModel, provider.name);
+        const reqStart = Date.now();
+        console.log(`[ModelGW] ${originalModel} → ${provider.name}/${model} (${body.stream ? 'stream' : 'sync'})`);
         try {
             const upstreamResp = await callChatCompletions(provider, {
                 model,
@@ -64,12 +67,14 @@ export async function modelGatewayRoutes(fastify) {
             });
             if (!upstreamResp.ok && upstreamResp.status !== 200) {
                 const errText = await upstreamResp.text();
+                recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart, status: upstreamResp.status, error: errText.slice(0, 200), streaming: !!body.stream });
                 console.error(`[ModelGW] ${provider.name} error ${upstreamResp.status}: ${errText.slice(0, 200)}`);
                 return reply.code(upstreamResp.status).send({
                     error: { message: `${provider.name}: ${errText.slice(0, 500)}` },
                 });
             }
             if (body.stream) {
+                recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart, status: 200, streaming: true });
                 return reply.headers({
                     'Content-Type': 'text/event-stream',
                     'Cache-Control': 'no-cache',
@@ -77,9 +82,11 @@ export async function modelGatewayRoutes(fastify) {
                 }).send(upstreamResp.body);
             }
             const data = await upstreamResp.json();
+            recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart, status: 200, streaming: false });
             reply.send(data);
         }
         catch (err) {
+            recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart, status: 502, error: err.message, streaming: !!body.stream });
             console.error(`[ModelGW] Error calling ${provider.name}:`, err.message);
             reply.code(502).send({
                 error: { message: `Provider error: ${err.message}` },
@@ -92,58 +99,48 @@ export async function modelGatewayRoutes(fastify) {
         if (!body || !body.input) {
             return reply.code(400).send({ error: { message: 'input is required' } });
         }
-        const model = body.model || 'deepseek-chat';
-        const provider = resolveProvider(model);
+        const originalModel = body.model || 'deepseek-chat';
+        const provider = resolveProvider(originalModel);
         if (!provider) {
             return reply.code(503).send({
                 error: { message: 'No available provider. Check API keys and quota.' },
             });
         }
-        console.log(`[ModelGW:Responses] ${model} → ${provider.name}`);
+        const model = remapModel(originalModel, provider.name);
+        body.model = model;
+        console.log(`[ModelGW:Responses] ${originalModel} → ${provider.name}/${model}`);
+        const reqStart2 = Date.now();
         try {
             const upstreamResp = await callResponsesApi(provider, body);
             if (!upstreamResp.ok) {
                 const errText = await upstreamResp.text();
+                recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart2, status: upstreamResp.status, error: errText.slice(0, 200), streaming: !!body.stream });
                 return reply.code(upstreamResp.status).send({
                     error: { message: `${provider.name}: ${errText.slice(0, 500)}` },
                 });
             }
+            recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart2, status: 200, streaming: !!body.stream });
             if (body.stream) {
                 return reply.headers({
                     'Content-Type': 'text/event-stream',
                     'Cache-Control': 'no-cache',
                 }).send(upstreamResp.body);
             }
-            // 将 Chat Completions 响应转回 Responses API 格式
-            const ccData = (await upstreamResp.json());
-            const choice = ccData.choices?.[0];
-            const responsesData = {
-                id: ccData.id,
-                object: 'response',
-                model: ccData.model,
-                output: [
-                    {
-                        type: 'message',
-                        role: 'assistant',
-                        content: [
-                            {
-                                type: 'output_text',
-                                text: choice?.message?.content || '',
-                            },
-                        ],
-                    },
-                ],
-                usage: ccData.usage,
-            };
-            reply.send(responsesData);
+            const data = await upstreamResp.json();
+            reply.send(data);
         }
         catch (err) {
+            recordRequest({ timestamp: Date.now(), model: originalModel, provider: provider.name, actualModel: model, latencyMs: Date.now() - reqStart2, status: 502, error: err.message, streaming: !!body.stream });
             console.error(`[ModelGW:Responses] Error:`, err.message);
             reply.code(502).send({
                 error: { message: `Provider error: ${err.message}` },
             });
         }
     });
+    // 运行时统计
+    fastify.get('/model-gateway/stats', async (_request, reply) => {
+        reply.send(getMetrics());
+    });
     // Provider 健康检查 + 熔断器状态
     fastify.get('/model-gateway/health/:provider', async (request, reply) => {
         const { provider } = request.params;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@starlink-awaken/agentmesh",
-  "version": "1.2.8",
+  "version": "1.3.2",
   "description": "Unified Agent Gateway - Multi-Agent Scheduler and Router",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",