项目实战：多工具 Agent 助手

// agent-assistant.ts
import "dotenv/config";
import { tool } from "@langchain/core/tools";
import { z } from "zod";
import { ChatOpenAI } from "@langchain/openai";
import { createToolCallingAgent, AgentExecutor } from "langchain/agents";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import { BufferMemory } from "langchain/memory";
import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { OpenAIEmbeddings } from "@langchain/openai";
 
// 工具 1：计算器
const calculator = tool(
  async ({ expr }: { expr: string }) => String(eval(expr)),
  { name: "calculator", description: "计算数学表达式，如 '23*7+4'", schema: z.object({ expr: z.string() }) }
);
 
// 工具 2：查天气（演示用假数据）
const getWeather = tool(
  async ({ city }: { city: string }) => `${city} 当前 22℃，晴。`,
  { name: "getWeather", description: "查询城市当前天气，如 '北京'", schema: z.object({ city: z.string() }) }
);
 
// 工具 3：RAG 检索（复用上一章的知识库）
function makeRetriever() {
  // 假设已 builtVectorStore；这里直接 asRetriever
  return builtVectorStore.asRetriever({ k: 3 });
}
const retrieverTool = tool(
  async ({ q }: { q: string }) => {
    const docs = await makeRetriever().getRelevantDocuments(q);
    return docs.map((d) => d.pageContent).join("\n---\n");
  },
  { name: "knowledgeBase", description: "从公司手册检索答案，如 '报销流程'", schema: z.object({ q: z.string() }) }
);
 
const tools = [calculator, getWeather, retrieverTool];
 
const prompt = ChatPromptTemplate.fromMessages([
  ["system", "你是全能助手，必要时使用工具。涉及公司内部问题用 knowledgeBase。"],
  ["placeholder", "{chat_history}"],
  ["human", "{input}"],
  ["placeholder", "{agent_scratchpad}"],
]);
 
const agent = createToolCallingAgent({
  llm: new ChatOpenAI({ model: "gpt-4o-mini", temperature: 0 }),
  tools,
  prompt,
});
 
const executor = new AgentExecutor({
  agent,
  tools,
  memory: new BufferMemory({ memoryKey: "chat_history", returnMessages: true }),
  verbose: true,
  maxIterations: 6,
});
 
// 使用
const r1 = await executor.invoke({ input: "北京比上海温度高多少？" });
const r2 = await executor.invoke({ input: "顺便算一下 (22-19)*3" });
console.log(r1.output, r2.output);

// app/api/assistant/route.ts
export const runtime = "nodejs";
export async function POST(req: Request) {
  const { input } = await req.json();
  const stream = await executor.stream({ input }); // AgentExecutor 支持 stream
  return new Response(stream as any, {
    headers: { "Content-Type": "text/plain; charset=utf-8" },
  });
}