Cloudflare D1 支持的聊天记忆

信息

此集成仅在 Cloudflare Workers 中受支持。

对于跨聊天会话的长期持久性，您可以将默认的内存 chatHistory 替换为 Cloudflare D1 实例，该实例支持诸如 BufferMemory 之类的聊天记忆类。

设置

您需要安装 LangChain Cloudflare 集成包。在下面的示例中，我们还使用了 Anthropic，但您可以使用任何您喜欢的模型

提示

有关安装集成包的通用说明，请参阅此部分。

npm

npm install @langchain/cloudflare @langchain/anthropic @langchain/core

Yarn

yarn add @langchain/cloudflare @langchain/anthropic @langchain/core

pnpm

pnpm add @langchain/cloudflare @langchain/anthropic @langchain/core

按照官方文档为您的 worker 设置 D1 实例。您项目的 wrangler.toml 文件应如下所示

name = "YOUR_PROJECT_NAME"
main = "src/index.ts"
compatibility_date = "2024-01-10"

[vars]
ANTHROPIC_API_KEY = "YOUR_ANTHROPIC_KEY"

[[d1_databases]]
binding = "DB"                                       # available in your Worker as env.DB
database_name = "YOUR_D1_DB_NAME"
database_id = "YOUR_D1_DB_ID"

使用方法

然后，您可以按如下方式使用 D1 存储您的历史记录

import type { D1Database } from "@cloudflare/workers-types";
import { BufferMemory } from "langchain/memory";
import { CloudflareD1MessageHistory } from "@langchain/cloudflare";
import {
  ChatPromptTemplate,
  MessagesPlaceholder,
} from "@langchain/core/prompts";
import { RunnableSequence } from "@langchain/core/runnables";
import { StringOutputParser } from "@langchain/core/output_parsers";
import { ChatAnthropic } from "@langchain/anthropic";

export interface Env {
  DB: D1Database;

  ANTHROPIC_API_KEY: string;
}

export default {
  async fetch(request: Request, env: Env): Promise<Response> {
    try {
      const { searchParams } = new URL(request.url);
      const input = searchParams.get("input");
      if (!input) {
        throw new Error(`Missing "input" parameter`);
      }
      const memory = new BufferMemory({
        returnMessages: true,
        chatHistory: new CloudflareD1MessageHistory({
          tableName: "stored_message",
          sessionId: "example",
          database: env.DB,
        }),
      });
      const prompt = ChatPromptTemplate.fromMessages([
        ["system", "You are a helpful chatbot"],
        new MessagesPlaceholder("history"),
        ["human", "{input}"],
      ]);
      const model = new ChatAnthropic({
        apiKey: env.ANTHROPIC_API_KEY,
      });

      const chain = RunnableSequence.from([
        {
          input: (initialInput) => initialInput.input,
          memory: () => memory.loadMemoryVariables({}),
        },
        {
          input: (previousOutput) => previousOutput.input,
          history: (previousOutput) => previousOutput.memory.history,
        },
        prompt,
        model,
        new StringOutputParser(),
      ]);

      const chainInput = { input };

      const res = await chain.invoke(chainInput);
      await memory.saveContext(chainInput, {
        output: res,
      });

      return new Response(JSON.stringify(res), {
        headers: { "content-type": "application/json" },
      });
    } catch (err: any) {
      console.log(err.message);
      return new Response(err.message, { status: 500 });
    }
  },
};

API 参考

• BufferMemory 来自 langchain/memory
• CloudflareD1MessageHistory 来自 @langchain/cloudflare
• ChatPromptTemplate 来自 @langchain/core/prompts
• MessagesPlaceholder 来自 @langchain/core/prompts
• RunnableSequence 来自 @langchain/core/runnables
• StringOutputParser 来自 @langchain/core/output_parsers
• ChatAnthropic 来自 @langchain/anthropic