# memsearch

A cross-platform semantic memory layer for AI coding agents that stores conversations as Markdown files and indexes them in Milvus for hybrid vector and keyword search, so memory carries over between Claude Code, Codex, OpenClaw and other agents.

memsearch gives AI coding agents a shared, persistent memory. It captures conversations automatically through per-agent plugins, storing them as human-readable Markdown files that stay the source of truth, while Milvus holds a rebuildable vector index built from content hashes for change detection.

Retrieval combines dense vector search, BM25 keyword search and reciprocal rank fusion across a three-layer recall path of search, expand and full transcript. A background process also distills repeated workflows into installable agent skills and keeps project and user notes current across sessions.

It works across Claude Code, Codex, DeepSeek Harness, OpenClaw and OpenCode, and exposes a command-line tool and a Python API for building custom memory or harness integrations.

Website: https://zilliztech.github.io/memsearch/