My plan of attack:
Replicate main functionality of tools in deepagents with equivalent MCP tool for DSPy
Create DSPy server with history and postgres persistence/memory
Endpoints for streaming and for no streaming that emit hooks for frontend
Discover how frontend is working, how it is showing everything in langgraph
There are most likely many endpoints it is interacting with to view state of graph
Port state/fake filesystem into DSPy or figure out how to use a real one
Build frontend to mirror functionality of current frontend in DSPy
Add memory with mem0 at some point
Instrument with langfuse
Get traces from langfuse
Implement an endpoint for optimizing programs
Connect to prompt managment in langfuse
So yeah I am declaring war on langgraph after my day today. Buddy of mine built an end-to-end system based on deepagents from langgraph and instead of building endpoints and a service to call the langgraph he just said langgraph dev and I’m stuck trying to reverse engineer whatever the shit langgraph platform is doing and I’m just about to use it as a target instead of a framework.