-
-
Notifications
You must be signed in to change notification settings - Fork 800
Open
Description
First of all thank you @developersdigest for this excellent youtube tutorial and repo.
Please take following as an idea and potential feature request.
I feel like it is not ideal to mix all the AI logic with the web app. I would like to see some stable example of separation of concerns of "complex AI backend" and consumer NextJS web app.
Imagine following setup of Answer Engine
Backend:
- Python - as it is lingua franca of LLM and the better part of LangChain libs
- LangServe (with LCEL streaming) as API
- LangGraph for complex state & process management and tool invocation
- Persistence of history (graph) e.g. in EdgeDB or Zep
- Rate limiting
- Semantic Caching
Frontend:
- NextJS app
- Streaming AI backend consumed as RemoteRunnable of LangChain
- Vercel AI SDK with AI/RSC for streaming React components from AI backend
- Able to stream intermediate agentic results / graph state from remote backend
There are multiple requests for such setup and nobody came with a stable solution
I have filled similar request to Vercel AI SDK repo:
vercel/ai#1506
Metadata
Metadata
Assignees
Labels
No labels