llm-simple-router 0.10.13 → 0.11.0

package/dist/config/recommended-providers.json

@@ -0,0 +1,355 @@
+[
+  {
+    "group": "DeepSeek",
+    "presets": [
+      {
+        "plan": "Anthropic",
+        "presetName": "deepseek",
+        "apiType": "anthropic",
+        "baseUrl": "https://api.deepseek.com/anthropic",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "deepseek-v4-flash",
+          "deepseek-v4-pro"
+        ]
+      },
+      {
+        "plan": "OpenAI",
+        "presetName": "deepseek-openai",
+        "apiType": "openai",
+        "baseUrl": "https://api.deepseek.com",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "deepseek-v4-flash",
+          "deepseek-v4-pro"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "百度千帆",
+    "presets": [
+      {
+        "plan": "API",
+        "presetName": "qianfan",
+        "apiType": "openai",
+        "baseUrl": "https://qianfan.baidubce.com/v2",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "ernie-4.0-8k",
+          "ernie-4.0-turbo-8k",
+          "ernie-3.5-8k",
+          "ernie-speed-8k",
+          "ernie-lite-8k",
+          "ernie-x1-32k-preview",
+          "deepseek-v3",
+          "deepseek-r1"
+        ],
+        "upstreamPath": "/chat/completions"
+      }
+    ]
+  },
+  {
+    "group": "科大讯飞",
+    "presets": [
+      {
+        "plan": "API",
+        "presetName": "iflytek-spark",
+        "apiType": "openai",
+        "baseUrl": "https://spark-api-open.xf-yun.com",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "4.0Ultra",
+          "generalv3.5",
+          "max-32k",
+          "generalv3",
+          "pro-128k",
+          "lite"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "硅基流动",
+    "presets": [
+      {
+        "plan": "API",
+        "presetName": "siliconflow",
+        "apiType": "openai",
+        "baseUrl": "https://api.siliconflow.cn",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "deepseek-ai/DeepSeek-V3.2-Exp",
+          "deepseek-ai/DeepSeek-R1",
+          "Qwen/Qwen3-8B",
+          "Qwen/Qwen2.5-72B-Instruct",
+          "Qwen/Qwen2.5-Coder-32B-Instruct",
+          "moonshotai/Kimi-K2-Instruct",
+          "moonshotai/Kimi-K2.5"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "智谱",
+    "presets": [
+      {
+      "plan": "Coding Plan",
+    "presetName": "zhipu-coding-plan",
+    "apiType": "openai",
+    "baseUrl": "https://open.bigmodel.cn",
+    "upstreamPath": "/api/coding/paas/v4/chat/completions",
+    "modelsEndpoint": "/v1/models",
+        "models": [
+          "glm-5.1",
+          "glm-5",
+          "glm-5-turbo",
+          "glm-4.7",
+          "glm-4.5-air"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "zhipu",
+        "apiType": "openai",
+        "baseUrl": "https://open.bigmodel.cn/api/paas/v4",
+        "modelsEndpoint": "/models",
+        "models": [
+          "glm-5.1",
+          "glm-5",
+          "glm-5-turbo",
+          "glm-4.7",
+          "glm-4.7-flash",
+          "glm-4.6"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "月之暗面",
+    "presets": [
+      {
+        "plan": "Coding Plan",
+        "presetName": "kimi-coding-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://api.kimi.com/coding",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "kimi-for-coding",
+          "kimi-k2.5"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "kimi",
+        "apiType": "openai",
+        "baseUrl": "https://api.moonshot.cn",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "kimi-k2.6",
+          "kimi-k2.5",
+          "kimi-k2-turbo-preview",
+          "kimi-k2-thinking",
+          "moonshot-v1-128k"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "Minimax",
+    "presets": [
+      {
+        "plan": "Token Plan",
+        "presetName": "minimax-token-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://api.minimaxi.com/anthropic",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "MiniMax-M2.7"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "minimax",
+        "apiType": "openai",
+        "baseUrl": "https://api.minimax.chat",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "MiniMax-M2.7",
+          "MiniMax-M2.7-highspeed",
+          "MiniMax-M2.5",
+          "MiniMax-M2.5-highspeed",
+          "MiniMax-M2.1",
+          "MiniMax-M2"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "火山引擎",
+    "presets": [
+      {
+        "plan": "Coding Plan",
+        "presetName": "volcengine-coding-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://ark.cn-beijing.volces.com/api/coding",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "ark-code-latest",
+          "doubao-seed-2.0-code",
+          "kimi-k2.5",
+          "glm-4.7",
+          "deepseek-v3.2"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "volcengine",
+        "apiType": "openai",
+        "baseUrl": "https://ark.cn-beijing.volces.com/api/v3",
+        "modelsEndpoint": "/models",
+        "models": [
+          "doubao-seed-2-0-pro-260215",
+          "doubao-seed-1-8-251228",
+          "doubao-seed-code-preview-251028"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "阿里云",
+    "presets": [
+      {
+        "plan": "Coding Plan",
+        "presetName": "aliyun-coding-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://coding.dashscope.aliyuncs.com/apps/anthropic",
+        "models": [
+          "qwen3.6-plus",
+          "qwen3-coder-next",
+          "qwen3-coder-plus",
+          "kimi-k2.5",
+          "glm-5",
+          "MiniMax-M2.5"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "aliyun",
+        "apiType": "openai",
+        "baseUrl": "https://dashscope.aliyuncs.com/compatible-mode",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "qwen3.6-plus",
+          "qwen3.5-plus",
+          "qwen3-max",
+          "qwen3.5-flash",
+          "qwen3-coder-plus",
+          "qwen3-coder-next"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "腾讯云",
+    "presets": [
+      {
+        "plan": "Coding Plan",
+        "presetName": "tencent-coding-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://api.lkeap.cloud.tencent.com/coding/anthropic",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "tc-code-latest",
+          "hunyuan-2.0-instruct",
+          "hunyuan-2.0-thinking",
+          "hunyuan-turbos",
+          "hunyuan-t1",
+          "glm-5",
+          "kimi-k2.5"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "tencent",
+        "apiType": "openai",
+        "baseUrl": "https://api.hunyuan.cloud.tencent.com",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "hunyuan-2.0-thinking",
+          "hunyuan-2.0-instruct",
+          "hunyuan-t1-latest",
+          "hunyuan-a13b",
+          "hunyuan-turbos-latest"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "OpenCode",
+    "presets": [
+      {
+        "plan": "Go OpenAI",
+        "presetName": "opencode-go-openai",
+        "apiType": "openai",
+        "baseUrl": "https://opencode.ai/zen/go/v1/chat/completions",
+        "modelsEndpoint": "/models",
+        "models": [
+          "glm-5.1",
+          "glm-5",
+          "kimi-k2.5",
+          "kimi-k2.6",
+          "deepseek-v4-pro",
+          "deepseek-v4-flash",
+          "mimo-v2-pro",
+          "mimo-v2-omni",
+          "mimo-v2.5-pro",
+          "mimo-v2.5",
+          "qwen3.6-plus",
+          "qwen3.5-plus"
+        ]
+      },
+      {
+        "plan": "Go Anthropic",
+        "presetName": "opencode-go-anthropic",
+        "apiType": "anthropic",
+        "baseUrl": "https://opencode.ai/zen/go/v1/messages",
+        "models": [
+          "minimax-m2.7",
+          "minimax-m2.5"
+        ]
+      }
+    ]
+  },
+  {
+    "group": "阶跃星辰",
+    "presets": [
+      {
+        "plan": "Step Plan",
+        "presetName": "stepfun-step-plan",
+        "apiType": "anthropic",
+        "baseUrl": "https://api.stepfun.com/step_plan",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "step-3.5-flash-2603",
+          "step-3.5-flash"
+        ]
+      },
+      {
+        "plan": "API",
+        "presetName": "stepfun",
+        "apiType": "openai",
+        "baseUrl": "https://api.stepfun.com",
+        "modelsEndpoint": "/v1/models",
+        "models": [
+          "step-3.5-flash",
+          "step-3",
+          "step-2-mini",
+          "step-2-16k",
+          "step-1-8k",
+          "step-1-32k"
+        ]
+      }
+    ]
+  }
+]

package/dist/config/recommended-retry-rules.json

@@ -0,0 +1,12 @@
+[
+  { "name": "429 Too Many Requests", "status_code": 429, "body_pattern": ".*", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": [] },
+  { "name": "503 Service Unavailable", "status_code": 503, "body_pattern": ".*", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": [] },
+  { "name": "ZAI 网络错误 (code 1234)", "status_code": 400, "body_pattern": "\"type\"\\s*:\\s*\"error\".*\"code\"\\s*:\\s*\"1234\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI 临时不可用", "status_code": 400, "body_pattern": "\"type\"\\s*:\\s*\"error\".*请稍后重试", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI 操作失败 (code 500)", "status_code": 400, "body_pattern": "\"type\"\\s*:\\s*\"error\".*\"code\"\\s*:\\s*\"500\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI 速率限制 (HTTP 200, code 1302)", "status_code": 200, "body_pattern": "\"error\".*\"code\"\\s*:\\s*\"1302\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI SSE 错误 (HTTP 200, code 500)", "status_code": 200, "body_pattern": "\"error\".*\"code\"\\s*:\\s*\"500\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI SSE 错误 (HTTP 200, code 1234)", "status_code": 200, "body_pattern": "\"error\".*\"code\"\\s*:\\s*\"1234\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "ZAI 模型过载 (HTTP 200, code 1305)", "status_code": 200, "body_pattern": "\"error\".*\"code\"\\s*:\\s*\"1305\"", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 10, "max_delay_ms": 60000, "providers": ["智谱"] },
+  { "name": "KIMI 401 认证错误", "status_code": 401, "body_pattern": ".*authentication_error.*", "retry_strategy": "exponential", "retry_delay_ms": 5000, "max_retries": 3, "max_delay_ms": 60000, "providers": ["月之暗面"] }
+]

package/dist/config/recommended.d.ts

@@ -7,6 +7,8 @@ export interface ProviderPreset {
     /** 上游模型列表端点路径，如 /v1/models 或 /models；拼接在 baseUrl 后 */
     modelsEndpoint?: string;
     models: string[];
+    /** 由 API handler 补充：模型名 → capabilities 映射 */
+    modelCapabilities?: Record<string, string[]>;
 }
 export interface ProviderGroup {
     group: string;

package/dist/config/version.json

@@ -0,0 +1 @@
+{"providers":1,"retryRules":1}

package/dist/core/monitor/request-tracker.d.ts

@@ -53,7 +53,7 @@ export declare class RequestTracker {
         attempts?: AttemptSnapshot[];
     }): void;
     /** Update stream metrics for a completed request (e.g., after cache estimation) */
-    updateCompletedMetrics(id: string, cacheReadTokens: number): void;
+    updateCompletedMetrics(id: string, cacheReadTokens: number, cacheReadTokensEstimated?: boolean): void;
     getActive(): ActiveRequest[];
     getRecent(limit?: number): ActiveRequest[];
     get(id: string): ActiveRequest | undefined;

package/dist/core/monitor/request-tracker.js

@@ -144,13 +144,14 @@ export class RequestTracker {
         this.broadcast("request_complete", completed);
     }
     /** Update stream metrics for a completed request (e.g., after cache estimation) */
-    updateCompletedMetrics(id, cacheReadTokens) {
+    updateCompletedMetrics(id, cacheReadTokens, cacheReadTokensEstimated) {
         const req = this.recentCompleted.find(r => r.id === id);
         if (!req || !req.streamMetrics)
             return;
         req.streamMetrics = {
             ...req.streamMetrics,
             cacheReadTokens,
+            cacheReadTokensEstimated: cacheReadTokensEstimated ? 1 : 0,
         };
         this.broadcast("request_complete", req);
     }

package/dist/core/monitor/types.d.ts

@@ -51,6 +51,7 @@ export interface StreamMetricsSnapshot {
     inputTokens: number | null;
     outputTokens: number | null;
     cacheReadTokens: number | null;
+    cacheReadTokensEstimated?: number;
     ttftMs: number | null;
     tokensPerSecond: number | null;
     stopReason: string | null;

package/dist/core/types.d.ts

@@ -37,6 +37,7 @@ export interface MetricsResult {
     output_tokens: number | null;
     cache_creation_tokens: number | null;
     cache_read_tokens: number | null;
+    cache_read_tokens_estimated?: number;
     ttft_ms: number | null;
     /** T6 - T0: proxy end-to-end streaming duration */
     total_duration_ms: number | null;

package/dist/index.js

@@ -6,6 +6,7 @@ import { randomUUID } from "crypto";
 import Fastify from "fastify";
 import { insertRequestLog } from "./db/logs.js";
 import { HTTP_NOT_FOUND, HTTP_INTERNAL_ERROR, getProxyApiType } from "./core/constants.js";
+import { loadModelDirectory } from "./config/model-context.js";
 import { API_CODE, apiError, isAdminApiResponse, statusToApiCode } from "./admin/api-response.js";
 const PROVIDER_DEFAULT_QUEUE_TIMEOUT_MS = 5000;
 const PROVIDER_DEFAULT_MAX_QUEUE_SIZE = 100;
@@ -82,6 +83,8 @@ export async function buildApp(options) {
     else {
         db = initDatabase(config.DB_PATH);
     }
+    // 加载外部模型目录（ai-model-directory），fallback 到硬编码白名单
+    loadModelDirectory();
     const isDev = process.env.NODE_ENV !== "production";
     const MAX_BODY_SIZE_MB = 50;
     const KB = 1024;
@@ -120,8 +123,21 @@ export async function buildApp(options) {
         return new Error(message);
     });
     // 记录请求到达时间，供全局错误处理计算延迟
-    app.addHook("onRequest", (request, _reply, done) => {
+    app.addHook("onRequest", (request, reply, done) => {
         request.receivedAt = Date.now();
+        // 全局 EPIPE 防护：ServerResponse 的 write 异步完成失败时，
+        // 内部 socketErrorListener → response.destroy(err) → response.emit('error')。
+        // 若无 listener 则该 error 成为 uncaught exception。
+        // 代理路由在 create-proxy-handler.ts 中已有额外监听，此处覆盖所有路由。
+        reply.raw.on("error", (err) => {
+            const code = err.code;
+            if (code === 'EPIPE') {
+                request.log.debug({ err }, "client disconnected (EPIPE)");
+            }
+            else {
+                request.log.warn({ err }, "response stream error");
+            }
+        });
         done();
     });
     // 统一错误处理：代理路由保持 {error:{message}}，Admin API 使用信封格式

package/dist/metrics/metrics-extractor.js

@@ -99,6 +99,7 @@ export class MetricsExtractor {
             output_tokens: this.outputTokens,
             cache_creation_tokens: this.cacheCreationTokens,
             cache_read_tokens: this.cacheReadTokens,
+            cache_read_tokens_estimated: this.cacheReadTokens !== null ? 0 : undefined,
             ttft_ms: this.ttftMs,
             total_duration_ms: totalDurationMs,
             tokens_per_second: totalTps,
@@ -306,6 +307,7 @@ function extractOpenAINonStream(parsed) {
         output_tokens: usage?.completion_tokens ?? null,
         cache_creation_tokens: null,
         cache_read_tokens: details?.cached_tokens ?? null,
+        cache_read_tokens_estimated: details?.cached_tokens != null ? 0 : undefined,
         ttft_ms: null,
         total_duration_ms: null,
         tokens_per_second: null,
@@ -321,6 +323,7 @@ function extractAnthropicNonStream(parsed) {
         output_tokens: usage?.output_tokens ?? null,
         cache_creation_tokens: usage?.cache_creation_input_tokens ?? null,
         cache_read_tokens: usage?.cache_read_input_tokens ?? null,
+        cache_read_tokens_estimated: usage?.cache_read_input_tokens != null ? 0 : undefined,
         ttft_ms: null,
         total_duration_ms: null,
         tokens_per_second: null,

package/dist/proxy/handler/create-proxy-handler.js

@@ -192,8 +192,23 @@ export function createProxyHandler(config) {
             // Socket error handling
             const socketErrorHandler = (err) => request.log.debug({ err }, "client socket error");
             request.raw.socket.on("error", socketErrorHandler);
+            // reply.raw (ServerResponse) error handling
+            // Node.js 中，TCP socket write 异步完成失败时（如 EPIPE），
+            // 内部 socketErrorListener → response.destroy(err) → response.emit('error')。
+            // 若无 listener，该 error 成为 uncaught exception 导致进程退出。
+            const replyErrorHandler = (err) => {
+                const code = err.code;
+                if (code === 'EPIPE') {
+                    request.log.debug({ err }, "client disconnected (EPIPE)");
+                }
+                else {
+                    request.log.warn({ err }, "response stream error");
+                }
+            };
+            reply.raw.on("error", replyErrorHandler);
             reply.raw.on("close", () => {
                 request.raw.socket.removeListener("error", socketErrorHandler);
+                reply.raw.removeListener("error", replyErrorHandler);
             });
             // 创建 pipeline context
             const ctx = createPipelineContext(request, reply, apiType);