memsearch plugin
Shared semantic memory for agents: conversations are saved as markdown and recalled by search across Claude Code, Codex and other agents
- 2,862 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Writes conversation content into local markdown files and a vector index
- By default sends text to an external embedding service unless a local model is chosen, and runs background memory maintenance
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/memsearchDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Writes conversation content into local markdown files and a vector index
- By default sends text to an external embedding service unless a local model is chosen, and runs background memory maintenance
This is third-party code. Review the repository files before installing.
What it does
memsearch gives agents long-term memory that is shared across tools. The plugin automatically records the flow of a conversation into markdown files and returns the relevant piece on request through semantic search. Plain .md files are the source of truth, while the Milvus vector index is treated as a derived cache that can be rebuilt at any time. Search is hybrid: dense vectors, sparse BM25 and reranking, with hash-based dedup and real-time file watching. It installs as a plugin for Claude Code, Codex, DeepSeek Harness, OpenClaw and OpenCode, and for custom use there is a CLI and a Python API from the memsearch package. Embeddings can come from a cloud provider or a local model with no key.
Who it is for. For developers and engineers who want memory that survives sessions and is shared between different agents.
Good fit when
- You want the agent to remember decisions and context between sessions
- You want one shared memory across several agents, for example Claude Code and Codex
- You need semantic search over your own markdown notes from a CLI or code
Not a fit when
- Manual notes in a project file are enough, without an index or background processes
- You cannot send conversation content to an external embedding service and cannot run a local model
Example request
We discussed Redis caching before, what TTL did we choose back thenLimitations
It needs Python 3.10 or newer. By default embeddings run through OpenAI, which is not reachable from Russia without a VPN, but there are local providers with no key: ONNX, Ollama and sentence-transformers, which work offline. The default store is a local Milvus Lite, and a remote Milvus is also supported. The plugin writes conversation content into files and an index, and background memory maintenance is a separate set of tasks that you enable deliberately.
How to disable. Remove the memsearch plugin via /plugin and restart the agent. For Codex and other platforms remove the plugin from the profile with the matching command. The index and markdown memory live in the .memsearch directory and are deleted separately.
FAQ
Where is memory stored?
In plain markdown files in the .memsearch directory, which are the source of truth. The Milvus index is a derived cache rebuilt from those files.
Can I avoid a cloud key?
Yes. There are local embedding providers, ONNX, Ollama and sentence-transformers, and the Milvus Lite store, all running on your own machine.
Discussion
No questions or reviews yet.
Related
PluginsA long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Persistent memory for coding agents: insights are stored as markdown cards and recalled at the start of the next session.
Coding memory across sessions, machines and teammates: decisions and dead ends flow into later prompts on their own.
Long-term memory for agents: new facts supersede old ones, every answer traces to its source, and recall makes no LLM calls
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| memsearch (this card) | no | 2,862 | |
| memory-lancedb-pro | no | 4,450 | a long-term memory plugin for the OpenClaw agent on LanceDB |
| Uteke | no | 286 | local memory for agents in a single Rust binary |
README in short
The README presents memsearch as cross-platform semantic memory for agents: markdown is the source of truth and Milvus serves as the index. Users install a plugin and get automatic conversation capture and recall, while developers use a CLI and Python API. It documents installation for Claude Code, Codex, DeepSeek Harness, OpenClaw and OpenCode, the choice of embedding provider including keyless local ones, and a keyless quick start on Milvus Lite. It also mentions three-layer recall, hybrid search and hash-based dedup, plus distilling repeated workflows into installable skills. MIT license, by Zilliz.