YourMemory applies Ebbinghaus forgetting curve decay to agentic AI memory, reporting +16pp recall over Mem0 on the LoCoMo benchmark.
Key Takeaways
Core mechanism: memories decay over time using the Ebbinghaus curve unless reinforced, modeling human memory consolidation rather than flat RAG retrieval.
Benchmarked against Mem0 on LoCoMo, claiming a 16 percentage point recall improvement at 52% absolute recall.
Claims 84% reduction in token usage versus baseline memory approaches.
Open-source on GitHub under sachitrafa/YourMemory; framed as a drop-in agentic memory layer.
Hacker News Comment Review
Skeptics call the “biological memory” framing marketing language on top of standard cache or chunked RAG mechanics, not a novel architecture.
The 84% token reduction claim is contested: commenters argue any typical chunked RAG system achieves comparable compression, so the baseline choice matters.
LoCoMo dataset selection is questioned directly; critics note it has known issues and is considered easy to overfit, which weakens the benchmark comparison.
Notable Comments
@larrydakhissi: “you just make Alzheimer a feature” – sharp framing of the intentional forgetting mechanic.
