Memoose plugin
Dual-path memory for agents: a knowledge graph on your machine, search on demand and hints before you ask, across sessions.
- 58 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Installation registers hooks in agent settings and adds a subagent.
- The agent writes project facts and procedures to long-term memory.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/memooseDetects 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.
- Installation registers hooks in agent settings and adds a subagent.
- The agent writes project facts and procedures to long-term memory.
This is third-party code. Review the repository files before installing.
What it does
Memoose keeps project knowledge so that it survives a session and a change of agent. An engine holds a typed knowledge graph on your machine, while skills, hooks and tools teach the connected model to use it. Every fact has an evidence pointer: the user's words, a file and lines. Knowledge is found two ways: by `memoose recall` search on request and by recommendation, which surfaces what the agent did not ask about before each prompt. Procedures are stored as a graph of steps with transition conditions, advice and pitfalls, and a session outcome counts against every transition taken so the next run learns from the last. `memoose maintain` gathers contradictions and duplicates into one worklist and a model decides on a small subagent. It has 26 MCP tools, four skills and a CLI: `memoose remember`, `recall`, `view` and more. Install with `memoose install claude`.
Who it is for. Teams on long projects where decisions, conventions and service ownership must stay correct for months.
Good fit when
- The agent forgets past session decisions and asks the same things.
- Project knowledge must be passed to another agent, Claude Code or Codex.
- Docs and the git log should become project memory once for a new agent.
- The agent breaks the order of procedure steps and you want to hint the next step.
Not a fit when
- One notes file in the repository is enough.
- You want cloud memory with no local install.
- You do not want hooks editing `~/.claude/settings.json`.
Example request
Remember: Bao owns the auth service and auth uses PostgreSQL. Then find who owns auth and what it runs on.Limitations
It needs Python 3.11 or newer. `memoose install claude` copies skills, registers hooks in `~/.claude/settings.json` and puts a subagent into `~/.claude/agents/`; for Codex hooks are registered in `~/.codex/hooks.json` and must be trusted in `/hooks`. Scope is set with `--project .` and `uninstall <host>` reverses it. An MCP server is only for agents without a shell: `install <host> --mcp` adds it and needs `uv`. Without the `fastembed` package a keyless hash fallback replaces local embeddings. The last commit was on 2 October 2026.
How to disable. Run `memoose uninstall claude` (or `codex`, `opencode`, `cursor`): it removes skills, hooks and the subagent. Delete graph data with `memoose forget` or by removing the store folder.
FAQ
How do I store and find a fact?
A fact looks like `source:Type --relation--> target:Type`: `memoose remember "bao:Person --owns--> auth:System"`. Search with `memoose recall "who owns billing"`.
What does memoose view show?
The knowledge graph in a browser as one HTML file, with nothing uploaded. The `--superseded` flag draws history dashed.
How do I fill memory for a new project?
Ask the agent to onboard Memoose: the `memoose-onboard` skill checks the host, installs what is missing and fills memory from the README, docs and git log.
Which hosts are supported?
Claude Code, Codex, OpenCode and Cursor: `memoose install claude`, `codex`, `opencode` or `cursor`. Hooks are supported on Claude Code and Codex.
What does memoose maintain do?
It gathers contradictions, duplicates and stale summaries into one worklist for judgment. The command decides nothing itself: a model decides on a small subagent.
Discussion
No questions or reviews yet.
Related
PluginsOne local binary that gives Claude Code, Codex and Cursor shared memory across sessions, with a knowledge graph on disk and no API keys
Long-term memory for agents: new facts supersede old ones, every answer traces to its source, and recall makes no LLM calls
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
An MCP server that gives an agent memory across sessions, projects and machines: decisions, dead ends, preferences and skills
README in short
A knowledge graph engine plus skills, hooks and a subagent for Claude Code, Codex and other hosts. Memory is found on demand and recommended before it.