How to connect Python LangChain to Vercel AI SDK

AI SDK
, , ,
1

Hey everyone! I’ve been working on a small but useful adapter that solves a real interoperability problem: connecting Python LangChain backends with Vercel AI SDK frontend
clients.

The Problem

If you’ve tried pairing LangChain (Python) on the backend with Vercel AI SDK’s useChat hook on the frontend, you’ve hit a wall — LangChain’s streaming output format doesn’t
match what Vercel AI SDK expects. The SSE format, message structure, and ID generation all differ.

The Solution

langchain-vercel-ai-sdk-adapter is a lightweight bridge that handles the translation:

from langchain_anthropic import ChatAnthropic
from langchain_ai_sdk_adapter import to_base_messages, to_ui_message_stream, create_ui_message_stream_response

@app.post(“/chat”)
async def chat(request: Request):
model = ChatAnthropic(model=“claude-3-5-sonnet-20241022”)
langchain_messages = await to_base_messages(await request.json())
stream = await model.astream(langchain_messages)
return create_ui_message_stream_response(to_ui_message_stream(stream))

Key Features

  • to_base_messages() — Converts AI SDK UIMessage format to LangChain HumanMessage/AIMessage objects
  • to_ui_message_stream() — Transforms LangChain AIMessageChunk streams into SSE-formatted strings
  • create_ui_message_stream_response() — Wraps the stream in a Starlette/FastAPI StreamingResponse with proper headers
  • Thought block filtering — Skips thinking blocks (e.g., from Claude), only emitting text content
  • LangGraph compatible — Works with graph.astream_events()

Why This Exists

LangChain is maturing fast on the Python side, and Vercel AI SDK is dominating the React/Next.js frontend space. But integrating them has been painful. This adapter is the
minimal, focused solution I wished existed.

Check It Out

Would love to hear if this solves a problem for anyone else, or if there are additional features you’d want see added.

Want me to adjust the tone, add anything, or change the focus?