memweave
A zero-infrastructure, async-first Python library that gives AI agents persistent memory stored as plain Markdown files and indexed by SQLite, so every memory can be opened, searched, edited and tracked with git diff rather than living in an opaque database.
memweave stores an AI agent's memories as ordinary Markdown files rather than in an external database or vector store, with SQLite used only to index them for retrieval. Retrieval combines three lightweight, non-neural post-processors implemented as plugins: a confidence-adaptive entity boost, a corpus-relative keyword boost, and an additive confidence-adaptive temporal boost, on top of a base search.
On the LongMemEval-S benchmark, a 500-question test of multi-session memory, temporal reasoning and preference tracking, the project reports recall and ranking scores measured with a local embedding model and no LLM or API calls at any stage, and publishes a comparison against a competing memory system along with 5-seed cross-validated results and full reproduction scripts.
It is a Python library, licensed under MIT.