Vercel AI LangChain Adaptor

generated by gpt-4o

Description

vercel-ai-langchain-adaptor is a lightweight and flexible adaptor for integrating LangChain-based applications with Vercel's AI platform. This package provides extended support for handling key events during AI interactions, offering a streamlined way to manage AI workflows, especially in serverless environments.

Features:

• Enhanced Event Handling: Supports critical AI and LLM-related events, providing hooks for monitoring and extending behavior.
• Streamlined Integration: Simplifies the process of connecting LangChain workflows to Vercel's infrastructure.
• Customizability: Enables developers to adapt event handling (on_tool_* and on_chat_model_*) to fit specific application needs.
• Serverless Ready: Designed to work seamlessly with Vercel's serverless deployment model.

Supported Events

This package focuses on select events from LangChain's ecosystem, with additional integration into Vercel's AI platform through useChat.

| Event | Official Support | This Package | Use in useChat | |------------------------|-----------------------|-------------------|-----------------------| | on_chat_model_start | 🔴 | 🔴 | | | on_chat_model_end | 🔴 | 🔴 | | | on_chat_model_stream | 🟢 | 🟢 | messages | | on_llm_start | 🔴 | 🔴 | | | on_llm_new_token | 🔴 | 🔴 | | | on_llm_end | 🔴 | 🔴 | | | on_llm_error | 🔴 | 🔴 | | | on_chain_start | 🔴 | 🔴 | | | on_chain_end | 🔴 | 🔴 | | | on_tool_start | 🔴 | 🟢 | data | | on_tool_end | 🔴 | 🟢 | data | | on_tool_error | 🔴 | 🟢 | data | | on_text | 🔴 | 🔴 | | | on_agent_action | 🔴 | 🔴 | | | on_agent_finish | 🔴 | 🔴 | |

Use Cases

• Real-time AI model updates through streaming (on_chat_model_stream).
• Tool usage monitoring and error handling with events like on_tool_start, on_tool_end, and on_tool_error.
• Extending LangChain's capabilities in serverless environments without significant overhead.

toDataStreamResponse

import { ChatOpenAI } from '@langchain/openai';
import { LangChainAdapter } from 'vercel-ai-langchain-adaptor';

export async function POST(req: Request) {
  const { prompt } = await req.json();

  const model = new ChatOpenAI({
    model: 'gpt-3.5-turbo-0125',
    temperature: 0,
  });

  const stream = await model.stream(prompt);

  return LangChainAdapter.toDataStreamResponse(stream);
}

https://sdk.vercel.ai/docs/reference/stream-helpers/langchain-adapter

Installation

Install the package using your favorite package manager:

npm install vercel-ai-langchain-adaptor

Contributions

Feel free to contribute to improve the adaptor or extend event support. See our CONTRIBUTING.md for guidelines.