memsearch

A markdown-first memory system that provides AI agents with persistent, human-readable semantic search across local files.

Memsearch treats your markdown directory as the single source of truth, converting plain text files into a high-performance vector index via Milvus. It eliminates vendor lock-in by keeping data in standard .md format, allowing you to edit, grep, or version-control memories with Git. The system features smart SHA-256 deduplication to prevent redundant embedding calls and includes a live file watcher for real-time synchronization. Whether you are using OpenAI, Gemini, or local models like Ollama, memsearch integrates into any Python agent framework to provide reliable, long-term context without the overhead of complex database management.

https://github.com/zilliztech/memsearch