Repolode
Submit
mnemon-dev

mnemon-dev/mnemon

Local, single-binary persistent memory for AI agents, using a four-graph knowledge store and LLM-supervised recall and writes across many coding agent runtimes.

View on GitHubWebsite
Stars
652
Forks
77
Open issues
14
Last commit
3 days ago

What is it?

LLM agents lose context between sessions, and long conversations push earlier decisions out of the window. Mnemon gives an agent a persistent memory store that survives across sessions. It keeps memories in a four-graph model with temporal, entity, causal, and semantic edges, and supports intent-aware recall, importance decay, and automatic deduplication.

Mnemon follows an LLM-supervised design. The host LLM decides what to remember, how to link items, and what to forget, while the mnemon binary handles storage, graph indexing, search, and decay. The core memory path needs no API keys, and optional embeddings can use a local Ollama or an OpenAI-compatible server for hybrid vector and keyword search.

The project ships setup integrations for many agent runtimes, including Claude Code, Codex, Cursor, OpenClaw, Pi, Hermes Agent, and DeepSeek Harness. Named stores isolate memory per project or agent, and a local web interface lets you browse memories and the relationship graph.

Key features

  • 01Four-graph storage with temporal, entity, causal, and semantic edges
  • 02Intent-aware recall combining graph traversal with optional vector search
  • 03Built-in deduplication and importance decay with garbage collection
  • 04Works without API keys; optional local or OpenAI-compatible embeddings
  • 05One-command setup for multiple agent runtimes
  • 06Named memory stores for isolating memory per project or agent

Use cases

  • Giving a coding agent like Claude Code memory of past architectural decisions across sessions
  • Keeping separate memory stores for different projects or agents
  • Browsing what agents have remembered through the local web interface
  • Adding persistent memory to an agent runtime that supports skills or hooks

Quick start

  1. 1

    Install the CLI with npm (requires Node.js 22+)

    npm install --global @mnemon-dev/mnemon
  2. 2

    Deploy the Mnemon setup for Claude Code

    mnemon setup
  3. 3

    Start a new Claude Code session; memory works automatically

Things to know

  • Agency is in preview and is not available on Windows.
  • The npm install path requires Node.js 22 or newer.

Topics

#agent-memory#ai-agent#knowledge-graph#llm-memory#cli#claude-code#persistent-memory#go#agent-framework#ai-tools

FAQ

What is mnemon?
Local, single-binary persistent memory for AI agents, using a four-graph knowledge store and LLM-supervised recall and writes across many coding agent runtimes. LLM agents lose context between sessions, and long conversations push earlier decisions out of the window. Mnemon gives an agent a persistent memory store that survives across sessions. It keeps memories in a four-graph model with temporal, entity, causal, and semantic edges, and supports intent-aware recall, importance decay, and automatic deduplication.
Is mnemon free to use? What license does it have?
mnemon is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is mnemon still maintained?
Actively maintained. The last commit to mnemon was on Oct 19, 2026.
What can I use mnemon for?
Giving a coding agent like Claude Code memory of past architectural decisions across sessions Keeping separate memory stores for different projects or agents Browsing what agents have remembered through the local web interface Adding persistent memory to an agent runtime that supports skills or hooks
How do I install mnemon?
Install the CLI with npm (requires Node.js 22+): npm install --global @mnemon-dev/mnemon Deploy the Mnemon setup for Claude Code: mnemon setup Start a new Claude Code session; memory works automatically
Where can I get mnemon?
The source code and installation instructions are on GitHub: https://github.com/mnemon-dev/mnemon

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects