Two Competing Answers to Agent Amnesia Sit at 91,319 and 58,484 Stars: claude-mem Compresses Whole Sessions, MemPalace Claims the Best Benchmarks

GitHub·medium signalXBlueskyLinkedInCopy link

thedotmack/claude-mem (91,319 stars) captures everything an agent does during a session, compresses it with an LLM, and injects the relevant slice back into future sessions. MemPalace (58,484 stars) takes the head-on route, advertising itself as 'the best-benchmarked open-source AI memory system' built on ChromaDB with an MCP interface. The two make opposite bets — session-transcript compression versus a benchmarked retrieval store — and neither has published a comparison against the other, so the benchmark claim is currently unrefereed.