FEATURE REQUEST: Persistent Memory Storage for checkpoints

[smokeyScraper]

Is your feature request related to a problem?

• Yes, it is related to a problem

Describe the feature you'd like

Implement Persistent Memory Storage for LG Checkpoints

Refers #73

🌟 Feature Description

Implement persistent database storage for conversation summaries, user interactions, and agent state management. Currently, all memory is lost when the application restarts.

🔍 Problem Statement

• InMemorySaver() loses all conversation history and summaries on application restart
• store_summary_to_database() function in summarization_node.py is a placeholder
• User interaction history is not persisted to the database
• No long-term memory across sessions for personalized user experiences
• Thread management relies on volatile memory storage

🎯 Expected Outcome

• Persistent conversation memory that survives application restarts
• Automatic storage of conversation summaries in Supabase interactions table
• User profile updates based on conversation history
• Long-term memory for personalized DevRel assistance
• Database-backed checkpointing instead of InMemorySaver()

Record

• I agree to follow this project's Code of Conduct
• I want to work on implementing this feature

[ShivamMenda]

Hi Can I work on this issue if this Is open to take up?

[smokeyScraper]

Hi @ShivamMenda,
Thanks for taking an interest in this issue! Yupp, you can definitely give it a go.

Please make sure to refer to #73 for a bit more details on this part.
The current system only uses InMemorySaver() to store users' memory threads, which ends up as a bottleneck once the number of users increases. To mitigate this, we planned on a hybrid system where, after a timeout of a discord thread, the whole conversation context gets summarized and gets stored in a persistent DB, preferably Supabase, which can later be fetched once a user seeks DevRel help via a DB call. This way, context isn't lost and the agent responds as per the user's preferences.

Refer to AsyncPostgresSaver in LangGraph for this part. Explore out Memory and Persistence in LangGraph docs to read more this part. lg-postgres lg-postgres-persistence-doc

Workflow Planned

Memory Persistence:

Memory → Context + Chat Convos + WebSearch (Merges to Chat Convos)

Storage in the DB after summarization and timeouts of the user thread.

Active conversation (InMemorySaver)

1. User message → LangGraph with InMemorySaver
2. Every N interactions → Summarize + Update context to this
3. Continue the conversation in memory

Session Timeout

1. Store the final summary in PostgreSQL
2. Clear from InMemorySaver()
3. User returns later → gather_context_node loads from PostgreSQL

[ShivamMenda]

Hi @smokeyScraper Sure will work on this, pretty new to langgraph and looks like an interesting problem to solve, will research and stark taking It up

[DhruvK278]

Hello, can I also work on the same issue. I am new to open source contribution but would like to give it a try.

[smokeyScraper]

hey @DhruvK278 ,
yepp, you can surely give it a go!

Have you tried running the codebase? and testing around? make sure to populate the .env by referring to .env.example
The issue isn't very much complex, testing of issue might end-up a bit of problem for you as it requires setting up Supabase database. You can backend/app/database/supabase for this part btw, but just dm me if you face any problem while running the application.

[DhruvK278]

@smokeyScraper Hi I was having an issue with supabase setup and regarding that I have sent u a dm on your discord id.

[smokeyScraper]

I have assigned the issue to you @DhruvK278, please make sure to refer the docs referenced in the thread.
All the things associated with temporary storage of threads work, we just need to implement store_summary_to_database function as detailed in the issue and check it's connection throughout the codebase if required.