Long Term Memory - Mem0/Zep/LangMem - what made you choose it?

Posted by nicoloboschi@reddit | LocalLLaMA | View on Reddit | 2 comments

I'm evaluating memory solutions for AI agents and curious about real-world experiences.

For those using Mem0, Zep, or similar tools:

- What initially attracted you to it?

- What's working well?

- What pain points remain?

- What would make you switch to something else?

[-]

sandstone-oli@reddit

Mem0 for ease. Zep for self-hosted. LangMem if you're already in that ecosystem.

All three solve storage and retrieval. None solve relevance governance: what's still current vs what's stale after months of use. That's the gap that makes agents feel dumber over time even though the model hasn't changed.

Building that governance layer at KAPEX (getkapex.ai). Different problem than the three above. Complementary, not competing.

[-]

secopsml@reddit

Postgresql, S3, structured output generation for metadata extraction and you have memory layer that works ❤️‍🩹