Universal Memory for AI agents
Portable context for every coding agent.
Universal Memory stores durable preferences, project facts, operating rules, and reusable skills so agents can continue work across sessions, hosts, and models without the same setup conversation every time.
$ uv tool install universal-memory
$ umem init
$ umem remember "Prefer typed Python and clean architecture." --scope project
$ umem context --scope project --format json
{
"ok": true,
"operation": "context",
"scope": "project",
"data": { "summary": "..." }
}Choose your path
Users
Install UMEM, initialize a project, and let your agent retrieve context and persist durable learnings safely.
Contributors
Understand the clean architecture, validation commands, release readiness, and alpha validation flow.
Agents and LLMs
Follow the operating protocol for context retrieval, durable memory, skills, and safe mutation.
Designed for controlled persistence
CLI and MCP parity
The CLI is canonical. MCP exposes equivalent capabilities for agents and hosts.
Snapshots and audit
Automatic mutations go through validation, secret scanning, snapshots, and audit events.
Reusable skills
Draft, validate, publish, and maintain canonical skills under .umem/skills/, then sync them into supported runtimes explicitly.
How this site relates to the README
README.md remains the package and repository entrypoint. This
MkDocs site is the curated documentation surface: task-oriented for users,
architectural for contributors, and operational for agents.