← Back to feed

supermemory

GitHub Repo Pretty sure · benchmarks are real, but limited
https://github.com/supermemoryai/supermemory

Persistent memory layer that actually ships with eval wins and API polish — the rare case where 'AI context engine' isn't pure marketing repackaging.

22%
50%
28%
Slop 22%Signal 50%Science 28%

Supermemory solves a concrete problem: LLMs don't retain context across sessions. The benchmark victories (LongMemEval, LoCoMo, ConvoMem) are measurable, not marketing fiction. The API surface is tight — one call for profile + search, no vector DB configuration theater. MCP integration is functional, not vaporware. The slop: 'state-of-the-art memory engine' oversells what's fundamentally RAG + temporal fact extraction. The benchmarks are niche (designed by the research community, not producti...

18495 stars TypeScript 2026-03-24 756 days old

Become a MFer to rate — log in