MemSearch 🤖 Agent Open source

What it does

memsearch gives AI coding agents a persistent, cross-platform memory layer. Each supported agent installs a plugin that automatically captures conversation turns, stores them as Markdown files, and lets the agent recall relevant history later, either through a /memory-recall command or by asking naturally when a question needs past context. Because memories are plain .md files, they are human-readable, editable, and version-controllable; a Milvus vector index sits alongside them as a rebuildable "shadow" cache for semantic search. The same project memory can be shared across different agents.

What is inside

The retrieval stack does progressive three-layer recall (search, expand, transcript), hybrid dense-vector plus BM25 sparse search with reciprocal-rank-fusion reranking, SHA-256 content hashing to skip unchanged content, and a file watcher that re-indexes in real time. Beyond the daily episodic journals, optional background maintenance keeps a durable PROJECT.md (active threads, decisions, risks) and USER.md (preferences, working style) current, and a "skills from memory" layer distils repeated workflows into installable agent skills that stay inert until you choose to install one. Embeddings default to a local ONNX bge-m3 model (no API key, no cost); OpenAI and Ollama providers are also supported. Beyond the plugins there is a full CLI and Python API for building memory into your own agents.

Works with

Claude Code, Codex, and OpenCode (the project also supports DeepSeek Harness and OpenClaw). All plugins share one backend, configured once.

How to install

/plugin marketplace add zilliztech/memsearch
/plugin install memsearch

Maintenance and safety

MIT licensed and very actively developed by Zilliz. Default embedding and the Milvus Lite backend run entirely locally; switching to Zilliz Cloud or a self-hosted Milvus sends memory data to that service. On first launch it downloads a roughly 558 MB model from HuggingFace.

Who should use it

Developers who want their agents to remember decisions, debugging threads, and project history across sessions and across tools.