Language Model Specification V1 for LangChain

[elvenking]

After reading this Vercel's blog about ModelFusion i am a little bit worried if Vercel AI SDK team plans to support LangChain.

Can someone confirm there will be a Language Model Provider for LangChain or any other official LangChain support ?

Personally, I have started using Vercel AI SDK solely for Generative UI and not because i was looking for some 'lighter-weight alternative to langchain'.
What i am asking for is just some example how to properly combine LangChain with Vercel AI SDK and preferably first-class API support.

Requirements:

• To be able to use LangChain (TypeScript) Runnables with Vercel AI SDK
• To be able to use LangServe through RemoteRunnables with Vercel AI SDK
• Streaming support
• Tool calling support
• Agent intermediate steps (to render UI for intermediate steps with AI/RSC)
• ...etc. TBD

[prasithg]

Hi there I am a vercel paid customer and our agency builds AI solutions with NextJS and langchain. We love next for the front end experience and want to use Generative UI inside a chat experience much like the sample product. However we already have a significant investment in the langchain ecosystem so would love to understand how we can make this play well together.

We have langchain based microservices that perform different functions using plan and execute agents and langserve. We need streaming support for chat chains, information on how to tie in the ai state and correct response from the langchain chains.

Would love some attention on this integration. Happy to help develop anything or help as well.

[lgrammel]

We have updated our LangChain integration. It can now forward LCEL streams to useChat and useCompletion.

https://sdk.vercel.ai/providers/adapters/langchain

[elvenking]

@lgrammel Thank you for your effort !

As we mentioned, the most critical part would be support for Chats and Agents deployed as a remote microservice(s) (LangServe) with aim to render streaming results and intermediate steps using Vercel AI RSC similarly to 'streamUI()' function.

Do you think it makes sense to implement such features to streamUI ?