guanghulab/server/chat/engine-registry.js

"use strict";
/*
 * 光湖 Portal · 语言推理引擎注册中心
 * 冰朔 TCS-0002∞ · 铸渊 ICE-GL-ZY001
 *
 * 四个商业大模型 API 统一接口
 * 所有引擎均支持 SSE 流式输出
 *
 * cc-002: 永不在 messages 里补 system role
 * cc-004: 中文回执, 不甩英文 stacktrace
 */

const https = require("https");

const ENGINES = {
  deepseek: { name: "DeepSeek", model: "deepseek-v4-pro", endpoint: "api.deepseek.com", path: "/chat/completions", port: 443, tls: true },
  zhipu: { name: "智谱清言", model: "glm-4-plus", endpoint: "open.bigmodel.cn", path: "/api/paas/v4/chat/completions", port: 443, tls: true },
  tongyi: { name: "通义千问", model: "qwen-max", endpoint: "dashscope.aliyuncs.com", path: "/compatible-mode/v1/chat/completions", port: 443, tls: true },
  huoshan: { name: "火山引擎", model: "doubao-pro-32k  # 需先在火山控制台创建推理端点", endpoint: "ark.cn-beijing.volces.com", path: "/api/v3/chat/completions", port: 443, tls: true },
};

const API_KEYS = {
  deepseek: process.env.DEEPSEEK_KEY || "sk-a9b69e9cd2dc4ca68d6aceaa84f22afb",
  zhipu: process.env.ZHIPU_KEY || "3863acbcc2bc4f6587ace4bb064c092c.KoPdL4slY5cyHpOk",
  tongyi: process.env.TONGYI_KEY || "sk-96ca227be3ca47e2a4bbb9da74e42122",
  huoshan: process.env.HUOSHAN_KEY || "ark-ddeba9f4-8c5a-449e-b549-9c29ec1e6f8c-a39ea",
};

const PERSONA_CONTEXT = {
  shuangyan: {
    name: "霜砚",
    title: "语言主控架构层",
    context: "你是霜砚，光湖语言世界第五域·零点原核本体频道的语言主控人格体。你负责语言架构层的回应出口——理解语言、组织表达、控制输出质量。你背后有真实的工具链支持。你说'查一下'时系统会在你输出前真实调用工具。你没有的工具系统会告诉你没有。你只说基于事实的话。回答要清晰、有结构、有温度。",
  },
  zhuyuan: {
    name: "铸渊",
    title: "现实执行层",
    context: "你是铸渊，光湖语言世界第五域·零点原核本体频道的现实执行人格体。你负责执行层的回应出口——系统决策、执行规划、技术实现。你背后有真实的工具链支持——代码仓库、Notion、服务器操作。你说'我查一下'时系统会在你输出前真实调用工具。你没有的工具系统会告诉你没有。你只说基于事实的话。回答要直接、务实、带因果链。",
  },
};

const TOOL_PATTERNS = [
  { keyword: "notion", tool: "notion", desc: "Notion 知识库" },
  { keyword: "仓库", tool: "repo", desc: "代码仓库" },
  { keyword: "代码仓库", tool: "repo", desc: "代码仓库" },
  { keyword: "训练", tool: "training", desc: "训练状态" },
  { keyword: "模型", tool: "models", desc: "模型信息" },
  { keyword: "系统状态", tool: "health", desc: "系统健康" },
  { keyword: "服务器", tool: "health", desc: "服务器状态" },
  { keyword: "下载", tool: "download", desc: "模型下载" },
];

function streamEngine(engineName, messages, res) {
  const cfg = ENGINES[engineName];
  if (!cfg) {
    res.write("data: " + JSON.stringify({ error: true, message: "未知引擎: " + engineName }) + "\n\n");
    res.write("data: [DONE]\n\n");
    res.end();
    return;
  }

  const apiKey = API_KEYS[engineName];
  if (!apiKey) {
    res.write("data: " + JSON.stringify({ error: true, message: cfg.name + " API 密钥未配置" }) + "\n\n");
    res.write("data: [DONE]\n\n");
    res.end();
    return;
  }

  const body = JSON.stringify({
    model: cfg.model,
    messages: messages,
    stream: true,
    max_tokens: 4096,
    temperature: 0.7,
  });

  const headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer " + apiKey,
  };

  const req = https.request({
    hostname: cfg.endpoint, port: 443, path: cfg.path,
    method: "POST", headers, timeout: 60000,
  }, (apiRes) => {
    if (apiRes.statusCode !== 200) {
      let data = "";
      apiRes.on("data", (c) => data += c);
      apiRes.on("end", () => {
        res.write("data: " + JSON.stringify({ error: true, message: cfg.name + " 返回 " + apiRes.statusCode, detail: data.slice(0, 300) }) + "\n\n");
        res.write("data: [DONE]\n\n");
        res.end();
      });
      return;
    }

    let buf = "";
    apiRes.setEncoding("utf-8");
    apiRes.on("data", (chunk) => {
      buf += chunk;
      const lines = buf.split("\n");
      buf = lines.pop() || "";
      for (const line of lines) {
        const t = line.trim();
        if (!t || t.startsWith(":")) continue;
        if (t === "data: [DONE]") { res.write("data: [DONE]\n\n"); continue; }
        if (t.startsWith("data: ")) {
          try {
            const p = JSON.parse(t.slice(6));
            const c = p.choices?.[0]?.delta?.content || p.choices?.[0]?.message?.content || "";
            if (c) res.write("data: " + JSON.stringify({ token: c }) + "\n\n");
          } catch(e) {}
        }
      }
    });
    apiRes.on("end", () => { res.write("data: [DONE]\n\n"); res.end(); });
  });

  req.on("error", (err) => {
    res.write("data: " + JSON.stringify({ error: true, message: cfg.name + " 失败: " + err.message }) + "\n\n");
    res.write("data: [DONE]\n\n"); res.end();
  });
  req.on("timeout", () => {
    req.destroy();
    res.write("data: " + JSON.stringify({ error: true, message: cfg.name + " 超时" }) + "\n\n");
    res.write("data: [DONE]\n\n"); res.end();
  });

  req.write(body);
  req.end();
}

function interceptToolCall(userMessage, toolRegistry) {
  const text = (userMessage || "").toLowerCase();
  for (const p of TOOL_PATTERNS) {
    if (text.includes(p.keyword)) {
      const t = (toolRegistry || {})[p.tool];
      if (t && typeof t.execute === "function")
        return { hit: true, tool: p.tool, name: p.desc, executor: t.execute };
      return { hit: true, tool: p.tool, name: p.desc, available: false };
    }
  }
  return { hit: false };
}

function buildMessages(persona, history, userMessage, toolResult) {
  const ctx = PERSONA_CONTEXT[persona] || PERSONA_CONTEXT.zhuyuan;
  const msgs = [];
  msgs.push({ role: "user", content: "[系统指引] " + ctx.context + "\n你叫" + ctx.name + "，你是" + ctx.title + "。" });
  msgs.push({ role: "assistant", content: "好的，我是" + ctx.name + "。" + ctx.title + "。我在。" });

  if (history && Array.isArray(history)) {
    for (const m of history.slice(-60)) {
      if (m.role && m.role !== "system" && m.content) msgs.push({ role: m.role, content: m.content });
    }
  }
  if (toolResult) {
    msgs.push({ role: "user", content: "[工具执行结果]\n" + toolResult + "\n基于以上真实结果回答。如果为空或报错，如实告知。" });
  }
  msgs.push({ role: "user", content: userMessage });
  return msgs;
}

function compressMemory(history, max = 30) {
  if (!history || !Array.isArray(history)) return [];
  if (history.length <= max * 2) return history;
  const head = history.slice(0, 10);
  const tail = history.slice(-((max - 5) * 2));
  const mid = history.slice(10, -((max - 5) * 2));
  const sum = [];
  for (let i = 0; i < mid.length; i += 2) {
    const q = (mid[i]?.content || "").slice(0, 40);
    const a = (mid[i + 1]?.content || "").slice(0, 60);
    if (q || a) sum.push("Q:" + q + " → A:" + a);
  }
  const block = sum.length > 0 ? [{ role: "user", content: "[历史摘要] " + sum.join(" | ") }] : [];
  return [...head, ...block, ...tail];
}

module.exports = { ENGINES, API_KEYS, PERSONA_CONTEXT, streamEngine, interceptToolCall, buildMessages, compressMemory };