@starlink-awaken/agentmesh 1.3.2 → 1.4.0

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 = '1.3.2';
+const VERSION = '1.4.0';
 const BANNER = `
    █████╗  ██████╗ ███████╗███╗   ██╗████████╗
   ██╔══██╗██╔════╝ ██╔════╝████╗  ██║╚══██╔══╝

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

@@ -21,9 +21,9 @@ export function recordRequest(log) {
         m.lastError = log.error;
         m.lastErrorTime = log.timestamp;
     }
-    recentRequests.unshift(log);
+    recentRequests.push(log);
     if (recentRequests.length > MAX_RECENT)
-        recentRequests.pop();
+        recentRequests.shift();
 }
 export function getMetrics() {
     const providers = {};
@@ -46,7 +46,7 @@ export function getMetrics() {
         total_requests: totalRequests,
         total_failures: totalFailures,
         providers,
-        recent: recentRequests.slice(0, 20).map(r => ({
+        recent: recentRequests.slice(-20).reverse().map(r => ({
             time: new Date(r.timestamp).toISOString(),
             model: r.model,
             provider: r.provider,

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

@@ -52,9 +52,6 @@ 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) {
     const messages = convertInputToMessages(body.input || []);
     if (body.instructions) {
@@ -84,9 +81,6 @@ export async function callResponsesApi(provider, body) {
         headers: { 'Content-Type': 'text/event-stream', 'Cache-Control': 'no-cache' },
     });
 }
-// ============================================================================
-// 输入转换: Responses input[] → Chat messages[]
-// ============================================================================
 function convertInputToMessages(input) {
     const messages = [];
     for (const item of input) {
@@ -95,7 +89,6 @@ function convertInputToMessages(input) {
             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',
@@ -111,7 +104,6 @@ function convertInputToMessages(input) {
                 });
                 break;
             }
-            // Function call output（Tool 返回结果）
             case 'function_call_output':
                 messages.push({
                     role: 'tool',
@@ -119,7 +111,6 @@ function convertInputToMessages(input) {
                     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;
@@ -130,7 +121,6 @@ function convertInputToMessages(input) {
                 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) });
                 }
@@ -138,16 +128,12 @@ function convertInputToMessages(input) {
     }
     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',
@@ -155,7 +141,6 @@ function convertChatToResponses(ccData) {
             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({
@@ -177,16 +162,16 @@ function convertChatToResponses(ccData) {
         usage: ccData.usage,
     };
 }
-// ============================================================================
-// SSE 流式转换: Chat SSE → Responses SSE
-// ============================================================================
+const EVENT_DELTA = 'response.output_text.delta';
+const EVENT_TOOL_DELTA = 'response.function_call_arguments.delta';
+const EVENT_COMPLETED = 'response.completed';
+const sseEncoder = new TextEncoder();
 function transformSSEStream(upstreamBody) {
-    const encoder = new TextEncoder();
     let buffer = '';
     let responseId = '';
     let modelName = '';
     let contentBuffer = '';
-    let toolCallAccum = {};
+    const toolCallAccum = {};
     return new ReadableStream({
         async start(controller) {
             const reader = upstreamBody.getReader();
@@ -197,17 +182,17 @@ function transformSSEStream(upstreamBody) {
                     if (done)
                         break;
                     buffer += decoder.decode(value, { stream: true });
-                    const lines = buffer.split('\n');
-                    buffer = lines.pop() || '';
-                    for (const line of lines) {
+                    let newlineIdx;
+                    while ((newlineIdx = buffer.indexOf('\n')) !== -1) {
+                        const line = buffer.slice(0, newlineIdx);
+                        buffer = buffer.slice(newlineIdx + 1);
                         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'));
+                            controller.enqueue(sseEncoder.encode(finalEvt));
+                            controller.enqueue(sseEncoder.encode('data: [DONE]\n\n'));
                             continue;
                         }
                         try {
@@ -217,16 +202,14 @@ function transformSSEStream(upstreamBody) {
                             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',
+                                    type: EVENT_DELTA,
                                     delta: delta.content,
                                 })}\n\n`;
-                                controller.enqueue(encoder.encode(evt));
+                                controller.enqueue(sseEncoder.encode(evt));
                             }
-                            // 工具调用增量
                             if (delta.tool_calls) {
                                 for (const tc of delta.tool_calls) {
                                     if (!toolCallAccum[tc.index]) {
@@ -246,15 +229,15 @@ function transformSSEStream(upstreamBody) {
                                 }
                                 // 发送 tool_call delta 事件
                                 const tcEvt = `data: ${JSON.stringify({
-                                    type: 'response.function_call_arguments.delta',
+                                    type: EVENT_TOOL_DELTA,
                                     tool_calls: Object.values(toolCallAccum),
                                 })}\n\n`;
-                                controller.enqueue(encoder.encode(tcEvt));
+                                controller.enqueue(sseEncoder.encode(tcEvt));
                             }
                         }
                         catch {
                             // 非 JSON 行直接透传
-                            controller.enqueue(encoder.encode(line + '\n'));
+                            controller.enqueue(sseEncoder.encode(line + '\n'));
                         }
                     }
                 }
@@ -283,14 +266,11 @@ function buildResponseEvent(id, model, content, toolCalls, isFinal) {
         });
     }
     return `data: ${JSON.stringify({
-        type: isFinal ? 'response.completed' : 'response.output_text.delta',
+        type: isFinal ? EVENT_COMPLETED : EVENT_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;
@@ -303,9 +283,6 @@ function convertToolSchemas(tools) {
         },
     }));
 }
-// ============================================================================
-// 辅助函数
-// ============================================================================
 function extractTextContent(content) {
     if (typeof content === 'string')
         return content;

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

@@ -15,9 +15,22 @@ 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)) {
+        // cfg.model_aliases: { "gpt-x": "deepseek-v4-pro", "claude-y": "claude-opus" }
+        // 按目标模型名反查对应的 provider → 写入对应 provider 的别名表
+        for (const [aliasKey, realModel] of Object.entries(cfg.model_aliases)) {
+            for (const [providerName, providerCfg] of Object.entries(config.providers)) {
+                const providerModels = providerCfg.models || [];
+                if (providerModels.includes(realModel)) {
+                    modelAliases[providerName] = modelAliases[providerName] || {};
+                    modelAliases[providerName][aliasKey] = realModel;
+                    break;
+                }
+            }
+            // 兜底：如果没匹配到任何 provider，放入 deepseek
             modelAliases.deepseek = modelAliases.deepseek || {};
-            modelAliases.deepseek[key] = val;
+            if (!modelAliases.deepseek[aliasKey]) {
+                modelAliases.deepseek[aliasKey] = realModel;
+            }
         }
     }
 }

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

@@ -4,6 +4,14 @@ import { getQuotaSummary, probeQuota } from './quota.js';
 import { circuitBreakerRegistry } from './circuit-breaker.js';
 import { checkAllProviders } from './health.js';
 import { getMetrics, recordRequest } from './metrics.js';
+import { checkRateLimit } from './rate-limit.js';
+function logReq(originalModel, providerName, actualModel, reqStart, status, streaming, error) {
+    recordRequest({
+        timestamp: Date.now(), model: originalModel,
+        provider: providerName, actualModel, latencyMs: Date.now() - reqStart,
+        status, streaming, error,
+    });
+}
 export async function modelGatewayRoutes(fastify) {
     // 健康检查 + 配额总览
     fastify.get('/model-gateway/health', async (_req, _reply) => {
@@ -41,6 +49,11 @@ export async function modelGatewayRoutes(fastify) {
     });
     // POST /v1/chat/completions — 标准 OpenAI 兼容端点
     fastify.post('/v1/chat/completions', async (request, reply) => {
+        const ip = request.ip || '127.0.0.1';
+        const rl = checkRateLimit('/v1/chat/completions', ip);
+        if (!rl.allowed) {
+            return reply.code(429).header('Retry-After', String(rl.resetSeconds)).header('X-RateLimit-Limit', rl.limit).header('X-RateLimit-Remaining', '0').send({ error: { message: 'Rate limit exceeded' } });
+        }
         const body = request.body;
         if (!body || !body.messages) {
             return reply.code(400).send({ error: { message: 'messages is required' } });
@@ -65,16 +78,16 @@ export async function modelGatewayRoutes(fastify) {
                 tools: body.tools,
                 tool_choice: body.tool_choice,
             });
-            if (!upstreamResp.ok && upstreamResp.status !== 200) {
+            if (!upstreamResp.ok) {
                 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 });
+                logReq(originalModel, provider.name, model, reqStart, upstreamResp.status, !!body.stream, errText.slice(0, 200));
                 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 });
+                logReq(originalModel, provider.name, model, reqStart, 200, true);
                 return reply.headers({
                     'Content-Type': 'text/event-stream',
                     'Cache-Control': 'no-cache',
@@ -82,11 +95,11 @@ 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 });
+            logReq(originalModel, provider.name, model, reqStart, 200, 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 });
+            logReq(originalModel, provider.name, model, reqStart, 502, !!body.stream, err.message);
             console.error(`[ModelGW] Error calling ${provider.name}:`, err.message);
             reply.code(502).send({
                 error: { message: `Provider error: ${err.message}` },
@@ -95,6 +108,11 @@ export async function modelGatewayRoutes(fastify) {
     });
     // POST /v1/responses — Codex Desktop Responses API 适配
     fastify.post('/v1/responses', async (request, reply) => {
+        const ip = request.ip || '127.0.0.1';
+        const rl = checkRateLimit('/v1/responses', ip);
+        if (!rl.allowed) {
+            return reply.code(429).header('Retry-After', String(rl.resetSeconds)).send({ error: { message: 'Rate limit exceeded' } });
+        }
         const body = request.body;
         if (!body || !body.input) {
             return reply.code(400).send({ error: { message: 'input is required' } });
@@ -109,17 +127,17 @@ export async function modelGatewayRoutes(fastify) {
         const model = remapModel(originalModel, provider.name);
         body.model = model;
         console.log(`[ModelGW:Responses] ${originalModel} → ${provider.name}/${model}`);
-        const reqStart2 = Date.now();
+        const reqStart = 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 });
+                logReq(originalModel, provider.name, model, reqStart, upstreamResp.status, !!body.stream, errText.slice(0, 200));
                 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 });
+            logReq(originalModel, provider.name, model, reqStart, 200, !!body.stream);
             if (body.stream) {
                 return reply.headers({
                     'Content-Type': 'text/event-stream',
@@ -130,7 +148,7 @@ export async function modelGatewayRoutes(fastify) {
             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 });
+            logReq(originalModel, provider.name, model, reqStart, 502, !!body.stream, err.message);
             console.error(`[ModelGW:Responses] Error:`, err.message);
             reply.code(502).send({
                 error: { message: `Provider error: ${err.message}` },

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

@@ -9,6 +9,7 @@ export interface ModelGatewayConfig {
     providers: Record<string, ModelProviderConfig>;
     fallback_chain: string[];
     model_routing: Record<string, string[]>;
+    model_aliases?: Record<string, string>;
 }
 export interface ResolvedProvider {
     name: string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@starlink-awaken/agentmesh",
-  "version": "1.3.2",
+  "version": "1.4.0",
   "description": "Unified Agent Gateway - Multi-Agent Scheduler and Router",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",