memsearch

What it is and what it does

Memsearch is a semantic memory layer for AI coding agents that indexes markdown files as a searchable knowledge base. It sits between agent platforms (Claude Code, OpenClaw, Codex CLI, OpenCode) and a vector database (Milvus Lite by default, or Zilliz Cloud), capturing conversation turns automatically and making them retrievable through natural language queries. Memories are stored as human-readable markdown files in `.memsearch/memory/`, with Milvus serving as a derived, rebuildable search index.

The package provides both plugin integrations (zero-config for end users) and a full Python API and CLI for developers building custom agents. It uses hybrid search combining dense vector embeddings (via ONNX, OpenAI, or Ollama), BM25 sparse retrieval, and reciprocal rank fusion reranking. A file watcher auto-indexes new or changed files in real time, and optional background tasks keep project and user notes current across sessions.

Use it for:

• Persistent context across multiple agent conversations: capture decisions and code patterns once, recall them automatically in future sessions.
• Team memory for multi-agent workflows: shared markdown knowledge base indexed and searchable by all agents in an OpenClaw or self-hosted setup.
• Distilling repeated workflows into reusable agent skills: extract procedural patterns from conversation history and package them as installable skills.
• Local-first semantic search without API costs: use the default ONNX embedding provider to index and search markdown without sending data to external services.
• Version-controlled knowledge management: store all memories as plain markdown, enabling git-based collaboration and audit trails.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Memsearch builds a searchable semantic memory index from markdown files, enabling AI coding agents to recall and reuse conversation context across sessions and platforms through vector search and hybrid retrieval.

Yes, if you use one of the supported agent platforms (Claude Code, OpenClaw, Codex CLI, OpenCode) and want persistent, searchable conversation memory with zero configuration. The plugin model is genuinely zero-setup for end users. If you're building a custom agent, the Python API and CLI are available but would benefit from clearer stability guarantees. No known security vulnerabilities and active maintenance make it safe to adopt.

Install

memsearch on PyPI

pip install memsearch

uv add memsearch

poetry add memsearch

Installing memsearch

Before you install

Low friction: pure Python wheel with nine runtime dependencies including click, milvus-lite, and openai. Active maintenance with a release 14 days ago. Supports Python 3.10–3.13.

License in practice

MIT license permits unrestricted use, modification, and distribution with minimal attribution requirements.

Quickstart

pip install memsearch

from memsearch import MemSearch
mem = MemSearch()
mem.add_memory("project.md", "content here")
results = mem.search("what did we discuss about caching?")

Requires Python ≥3.10. First-time embedding setup downloads ~558 MB ONNX model from HuggingFace Hub unless configured otherwise.

Verify before relying

• Whether the Python API is fully documented and stable for direct programmatic use outside plugin contexts.
• Performance characteristics when memory grows to thousands of markdown files or millions of indexed tokens.
• Whether background sync and skill distillation features are production-ready or still experimental.

Package facts

Evidence: memsearch-0.4.17-py3-none-any.whl

Tags

More Artificial Intelligence packages

Further reading