Give your coding agents a shared, searchable past

What it does

memsearch is a cross-platform memory layer for AI coding agents. It captures conversation turns from supported agents—Claude Code, Codex CLI, OpenClaw, and OpenCode—and stores them as daily Markdown files. A local Milvus instance (or Zilliz Cloud) indexes these files so any agent can semantically recall context from sessions that happened in another tool.

The interesting bit

The design treats Markdown as the source of truth and Milvus as a disposable “shadow index.” Because your memories are just .md files, they remain human-readable, editable, and version-controllable; if the vector database disappears, you can rebuild it from the text. The retrieval pipeline is also deliberately layered: it runs dense vector search, BM25 sparse search, and RRF reranking, then expands results and pulls raw transcript chunks.

Key highlights

• Memories flow across agents without extra setup; a conversation in Claude Code becomes searchable context in Codex CLI.
• Runs locally by default using ONNX bge-m3 on CPU—no API key required for embeddings.
• Auto-indexes via a file watcher and uses SHA-256 hashing to skip unchanged content.
• Offers a CLI and Python API for developers building memory into custom agents.
• Optional background tasks can maintain higher-level project notes and user profiles, though these are disabled by default.

Verdict

Worth a look if you juggle multiple AI coding agents and want one long-term memory store you can actually read and diff. Skip it if you only use a single agent with built-in history you already trust.