Hi folks, are there file-system based stores around, to replace `InMemoryStore` from `langgraph.store.memory`?
I would like to use such a system to save a memory within a local folder, and be able to edit it easily. I guess this would work fine with SQLite too, but I’d be really eager to craft an agent that works just like a coworker, that would take notes as text files within the proper shared folder.
Hi @eric-burel
I guess sqlite-vec would be what you are actually seeking SQLiteVec - Docs by LangChain
Plain text files would be useful for tools such as claude-code/Cursor etc., which would make the tool configurable for users. But still, they would be configuration files, not the agent store itself
Not exactly, this is useful for using the file system as a storage and I use this pattern for instance to have durability in Google Colab, using Google Drive integration. Here I’d need to use folders as tables and text files as rows (data plane), and maybe sqlite for config (control plane). But to be fair that’s something that should be pretty easy to implement in user land if it doesn’t exist yet, I am just trying to not reinvent the wheel.
Ok, I get it. Deep Agents seems to have it implemented Deep Agents, you could look into the codebase and draw some inspiration
Here is the implementation deepagents/libs/deepagents/deepagents/middleware/filesystem.py at master · langchain-ai/deepagents · GitHub
1 Like