Brilliant; your solution for persistent state management in conversational AI, using LangGraph's checkpoint system, is exactly the step needed for truely intelligent agents.
The checkpoint persitence using SQLite is really clever here. I've been strugling with multi-agent state management and losing context after server restarts has been a nightmare. Your thread_id format (user:topic) is so simple yet solves the whole problem. The way you keep the last 5 exchanges and summarize older ones is a nice balance for avoiding token bloat. Did you run into any issues with concurrent users hitting the same SQLite database?
Brilliant; your solution for persistent state management in conversational AI, using LangGraph's checkpoint system, is exactly the step needed for truely intelligent agents.
Will try today definetly let u know my learning as well by EOD.
The checkpoint persitence using SQLite is really clever here. I've been strugling with multi-agent state management and losing context after server restarts has been a nightmare. Your thread_id format (user:topic) is so simple yet solves the whole problem. The way you keep the last 5 exchanges and summarize older ones is a nice balance for avoiding token bloat. Did you run into any issues with concurrent users hitting the same SQLite database?
I didn't test it with concurrent users, but this is doable for sure. And summarizing is necessary. Otherwise, the context grows too much for the model
Great stuff, subscribed!