OMEM
Open-source, self-hosted memory for AI agents built on belief revision. Keeps both sides of a contradiction instead of overwriting, tracks what was believed and when, answers why with a provenance chain, and gates risky actions behind human approval. MIT, pip install, dashboard included.
Cost / License
- Free
- Open Source (MIT)
Platforms
- Windows
- Mac
- Linux
OMEM
Features
Properties
- Lightweight
- Privacy focused
- AI-Powered
Features
- Works Offline
- Dark Mode
- No registration required
- No Tracking
- Ad-free
- Audit Trail
- Model Context Protocol (MCP) Support
OMEM information
Developed by
Omem Licensing
Open Source (MIT) and Free product.Written in
Alternatives
0 alternatives listedSupported Languages
- English
AlternativeTo Categories
Development, AI Tools & ServicesGitHub repository
- 0 Stars
- 0 Forks
- 0 Open Issues
- Updated
What is OMEM?
OMEM is memory for AI agents that refuses to decide what is true. Most agent memory overwrites a fact the moment a new one conflicts with it, and the history is gone. OMEM keeps both sides of every contradiction, marks which one is currently believed, and can reconstruct what was believed at any past moment.
Every belief answers "why": the source, the time, the evidence chain, and anything on record that contradicts it. Risky actions do not run on a model's say-so: only registered action types can execute, and high-risk actions wait for a named human approver, with approvals and refusals both recorded. That makes OMEM an audit trail for agents, not just a store, which matters for teams shipping agents to clients who ask "why did it do that" in security reviews.
Runs entirely on your own machine: one pip install, zero runtime dependencies, SQLite by default (Postgres optional), API and dashboard on one port, MCP server included for Claude and other MCP clients. A CI test fails the build if it ever makes a network connection that is not loopback. MIT licensed, with a written commitment that the core stays MIT.