Memweave CLI lets AI agents store searchable memory as plain text files instead of proprietary databases

A developer released memweave, a Python library that gives AI agents persistent memory stored as readable Markdown files indexed by SQLite—no external services, no cloud vector database to pay for. The tool combines keyword search (BM25) with semantic search (vector embeddings) and automatically degrades to keyword-only mode if the embedding API goes down.

Unlike black-box memory systems, every memory lives in a file you can open, edit with any text editor, search with grep, and version-control with git. The SQLite index is purely derived; losing the database is not data loss. Embeddings are cached by content hash so identical facts never get embedded twice, eliminating redundant API calls.

Engineers building AI agents now have a way to inspect exactly what their agents learned between runs, edit memories directly on disk, and run the entire system offline or without paying for vector database hosting. Developers working on multi-agent systems can isolate namespaces per agent while sharing foundational knowledge across teams.