@templmf/temp-solf-lmf 0.0.43 → 0.0.45

package/ai-gateway/package.json

@@ -0,0 +1,18 @@
+{
+  "name": "ai-gateway",
+  "version": "1.0.0",
+  "description": "AI API Gateway - 兼容 OpenAI 格式，支持内网 Qwen 模型",
+  "type": "module",
+  "main": "src/index.js",
+  "scripts": {
+    "start": "node src/index.js",
+    "dev": "node --watch src/index.js"
+  },
+  "dependencies": {
+    "express": "^4.18.2",
+    "openai": "^4.28.0",
+    "dotenv": "^16.4.1",
+    "uuid": "^9.0.0",
+    "winston": "^3.11.0"
+  }
+}

package/ai-gateway/src/index.js

@@ -0,0 +1,132 @@
+/**
+ * AI Gateway - 入口
+ *
+ * 对外接口（全部兼容 OpenAI 格式）：
+ *
+ *   POST /v1/chat/completions    核心接口
+ *   GET  /v1/models              模型列表（VSCode Roo 等插件需要）
+ *   GET  /health                 健康检查 + Skill / Session 状态
+ *   DELETE /v1/sessions/:id      清除指定会话
+ */
+
+import "dotenv/config";
+import express from "express";
+import { authMiddleware } from "./middleware/auth.js";
+import { rateLimitMiddleware } from "./middleware/rateLimit.js";
+import { chatRouter } from "./routes/chat.js";
+import { loadAllSkills, getSkillRegistry } from "./skills/registry.js";
+import { getSessionStats, deleteSession } from "./utils/sessionManager.js";
+import { logger } from "./utils/logger.js";
+
+const app = express();
+const PORT = process.env.PORT || 3000;
+
+// ─────────────────────────────────────────────────────
+// 全局中间件
+// ─────────────────────────────────────────────────────
+app.use(express.json({ limit: "10mb" }));
+
+// 请求日志（跳过健康检查，避免刷屏）
+app.use((req, res, next) => {
+  if (req.path !== "/health") {
+    logger.debug(`${req.method} ${req.path}`, { ip: req.ip });
+  }
+  next();
+});
+
+// ─────────────────────────────────────────────────────
+// 健康检查（无需鉴权）
+// ─────────────────────────────────────────────────────
+app.get("/health", (req, res) => {
+  res.json({
+    status: "ok",
+    timestamp: new Date().toISOString(),
+    upstream: {
+      baseUrl: process.env.UPSTREAM_BASE_URL,
+      defaultModel: process.env.DEFAULT_MODEL
+    },
+    skills: getSkillRegistry(),
+    sessions: getSessionStats()
+  });
+});
+
+// ─────────────────────────────────────────────────────
+// 模型列表（VSCode Roo / Open WebUI 等需要）
+// ─────────────────────────────────────────────────────
+app.get("/v1/models", authMiddleware, (req, res) => {
+  const models = (process.env.AVAILABLE_MODELS || process.env.DEFAULT_MODEL || "qwen2.5-72b-instruct")
+    .split(",")
+    .map(m => m.trim())
+    .filter(Boolean);
+
+  res.json({
+    object: "list",
+    data: models.map(id => ({
+      id,
+      object: "model",
+      created: Math.floor(Date.now() / 1000),
+      owned_by: "local",
+      permission: [],
+      root: id,
+      parent: null
+    }))
+  });
+});
+
+// ─────────────────────────────────────────────────────
+// 会话管理接口
+// ─────────────────────────────────────────────────────
+app.delete("/v1/sessions/:sessionId", authMiddleware, (req, res) => {
+  const { sessionId } = req.params;
+  const deleted = deleteSession(sessionId);
+  if (deleted) {
+    res.json({ deleted: true, session_id: sessionId });
+  } else {
+    res.status(404).json({
+      error: { message: "Session not found", type: "not_found_error" }
+    });
+  }
+});
+
+// ─────────────────────────────────────────────────────
+// 核心接口：鉴权 → 限流 → 处理
+// ─────────────────────────────────────────────────────
+app.use("/v1/chat/completions", authMiddleware, rateLimitMiddleware, chatRouter);
+
+// ─────────────────────────────────────────────────────
+// 全局错误兜底
+// ─────────────────────────────────────────────────────
+app.use((err, req, res, _next) => {
+  logger.error("Unhandled error", { error: err.message, stack: err.stack });
+  res.status(500).json({
+    error: { message: "Internal server error", type: "server_error" }
+  });
+});
+
+// 404
+app.use((req, res) => {
+  res.status(404).json({
+    error: { message: `Route ${req.method} ${req.path} not found`, type: "not_found_error" }
+  });
+});
+
+// ─────────────────────────────────────────────────────
+// 启动
+// ─────────────────────────────────────────────────────
+const skillsDir = process.env.SKILLS_DIR || "./src/skills";
+const { loaded, missing, total } = loadAllSkills(skillsDir);
+
+app.listen(PORT, () => {
+  logger.info("═══════════════════════════════════════");
+  logger.info("  AI Gateway started");
+  logger.info("═══════════════════════════════════════");
+  logger.info(`  Port        : ${PORT}`);
+  logger.info(`  Upstream    : ${process.env.UPSTREAM_BASE_URL}`);
+  logger.info(`  Default model: ${process.env.DEFAULT_MODEL}`);
+  logger.info(`  Skills      : ${loaded}/${total} loaded${missing > 0 ? ` (${missing} missing)` : ""}`);
+  logger.info("───────────────────────────────────────");
+  logger.info(`  Health      : http://localhost:${PORT}/health`);
+  logger.info(`  API         : http://localhost:${PORT}/v1/chat/completions`);
+  logger.info(`  Models      : http://localhost:${PORT}/v1/models`);
+  logger.info("═══════════════════════════════════════");
+});

package/ai-gateway/src/middleware/auth.js

@@ -0,0 +1,45 @@
+/**
+ * 鉴权中间件
+ *
+ * 客户端（VSCode Roo、前端应用等）调用网关时需要携带 Gateway API Key。
+ * 支持两种传入方式（与 OpenAI 兼容）：
+ *   1. Header:  Authorization: Bearer gw-key-xxx
+ *   2. Header:  x-api-key: gw-key-xxx
+ */
+
+export function authMiddleware(req, res, next) {
+  // 从环境变量读取允许的 key 列表
+  const allowedKeys = (process.env.GATEWAY_API_KEYS || "")
+    .split(",")
+    .map(k => k.trim())
+    .filter(Boolean);
+
+  // 开发模式：未配置 key 则跳过鉴权
+  if (allowedKeys.length === 0) {
+    req.clientId = "anonymous";
+    return next();
+  }
+
+  // 从 Header 提取 key
+  const authHeader = req.headers["authorization"] || "";
+  const xApiKey = req.headers["x-api-key"] || "";
+
+  let clientKey = xApiKey;
+  if (!clientKey && authHeader.startsWith("Bearer ")) {
+    clientKey = authHeader.slice(7).trim();
+  }
+
+  if (!clientKey || !allowedKeys.includes(clientKey)) {
+    return res.status(401).json({
+      error: {
+        message: "Invalid API key. Please check your GATEWAY_API_KEYS configuration.",
+        type: "authentication_error",
+        code: "invalid_api_key"
+      }
+    });
+  }
+
+  // 把 key 的索引作为 clientId（用于日志追踪，不暴露原始 key）
+  req.clientId = `client-${allowedKeys.indexOf(clientKey)}`;
+  next();
+}

package/ai-gateway/src/middleware/rateLimit.js

@@ -0,0 +1,87 @@
+/**
+ * 限流中间件
+ *
+ * 基于内存的滑动窗口限流，无需 Redis 依赖。
+ * 生产环境建议换成 Redis 实现以支持多实例部署。
+ *
+ * 限流维度（按优先级）：
+ *   1. clientId（API Key 级别）
+ *   2. IP 地址（兜底，防止未鉴权滥用）
+ */
+
+// 滑动窗口记录：{ key -> [timestamp, ...] }
+const windows = new Map();
+
+// 定期清理过期记录，防止内存泄漏
+setInterval(() => {
+  const now = Date.now();
+  for (const [key, timestamps] of windows.entries()) {
+    const windowMs = parseInt(process.env.RATE_LIMIT_WINDOW_MS || "60000");
+    const valid = timestamps.filter(t => now - t < windowMs);
+    if (valid.length === 0) {
+      windows.delete(key);
+    } else {
+      windows.set(key, valid);
+    }
+  }
+}, 30_000);
+
+function isRateLimited(key, maxRequests, windowMs) {
+  const now = Date.now();
+  const timestamps = windows.get(key) || [];
+
+  // 过滤掉窗口外的记录
+  const valid = timestamps.filter(t => now - t < windowMs);
+
+  if (valid.length >= maxRequests) {
+    windows.set(key, valid);
+    return {
+      limited: true,
+      remaining: 0,
+      resetMs: windowMs - (now - valid[0])
+    };
+  }
+
+  valid.push(now);
+  windows.set(key, valid);
+  return {
+    limited: false,
+    remaining: maxRequests - valid.length,
+    resetMs: 0
+  };
+}
+
+export function rateLimitMiddleware(req, res, next) {
+  const windowMs = parseInt(process.env.RATE_LIMIT_WINDOW_MS || "60000");
+  const maxPerClient = parseInt(process.env.RATE_LIMIT_MAX_PER_CLIENT || "100");
+  const maxPerIp = parseInt(process.env.RATE_LIMIT_MAX_PER_IP || "200");
+
+  // 优先按 clientId 限流（鉴权后才有 clientId）
+  const limitKey = req.clientId && req.clientId !== "anonymous"
+    ? `client:${req.clientId}`
+    : `ip:${req.ip}`;
+
+  const max = req.clientId && req.clientId !== "anonymous"
+    ? maxPerClient
+    : maxPerIp;
+
+  const result = isRateLimited(limitKey, max, windowMs);
+
+  // 标准限流响应头
+  res.setHeader("X-RateLimit-Limit", max);
+  res.setHeader("X-RateLimit-Remaining", result.remaining);
+  res.setHeader("X-RateLimit-Window-Ms", windowMs);
+
+  if (result.limited) {
+    res.setHeader("Retry-After", Math.ceil(result.resetMs / 1000));
+    return res.status(429).json({
+      error: {
+        message: `Rate limit exceeded. Try again in ${Math.ceil(result.resetMs / 1000)}s.`,
+        type: "rate_limit_error",
+        code: "rate_limit_exceeded"
+      }
+    });
+  }
+
+  next();
+}