Memoria is a persistent memory layer for AI agents that applies Git-style version control to what an agent remembers. Every memory change is tracked and reversible: you can take snapshots, open branches, merge them back, and roll back to an earlier point in time. Version control runs on the copy-on-write storage engine of the MatrixOne database, so branches and snapshots are zero-copy.
Agents reach Memoria over the Model Context Protocol (MCP), so it works with any MCP-compatible agent, including Kiro, Cursor, Claude Code, Codex, and Gemini CLI, plus a native OpenClaw plugin. It runs as a hosted cloud service or self-hosted next to a MatrixOne database, and it is also the memory component of MatrixOrigin's Astra agent platform, where it stores facts, preferences, decisions, and session summaries.
Retrieval combines vector and full-text search. A governance layer detects contradictions between memories, quarantines low-confidence entries, and records a snapshot and provenance chain for every mutation. Steering rules installed into each coding agent tell it when to store, retrieve, correct, and purge memories, how to bootstrap and clean up a session, and when to branch for an experiment.
| Memory type | Used for |
|---|---|
| semantic | Project facts and decisions |
| profile | User preferences |
| procedural | Workflows and how-to steps |
| working | Temporary task context |
| episodic | Session summaries |
Features
- Memory version control: snapshots, branches, merges, diffs, and point-in-time rollback for agent memory
- Hybrid retrieval: vector and full-text search find memories by meaning as well as keywords
- Self-governance: contradiction detection, quarantine of low-confidence memories, and periodic reflection to synthesize insights
- Audit trail: every memory mutation carries a snapshot and provenance chain
- MCP tool set: store, retrieve, search, correct, purge, list, profile, and relevance-feedback tools exposed to the agent
- Steering rules: per-agent rules for Kiro, Cursor, Claude Code, Codex, and Gemini CLI covering memory hygiene, session lifecycle, branching, and goal tracking
- Goal tracking: goals, steps, and lessons persist across conversations so an agent can resume unfinished work
- Local embeddings: an optional local embedding model keeps memory data on your machine
- Deployment modes: a hosted cloud service or a self-hosted MatrixOne-backed server, plus a REST API
