Memoria provides a persistent memory layer for AI agents, offering Git-level version control for memory operations. It addresses the challenges of data integrity, hallucination reduction, and maintaining consistent long-term context for AI agents. The project benefits developers building AI agents by making memory management auditable, reversible, and robust, akin to version control for code.

How It Works

Memoria implements Git-like functionalities—snapshot, branch, merge, and rollback—for AI agent memory, powered by MatrixOne's native Copy-on-Write engine. This enables zero-copy branching and instant snapshots, facilitating isolated experimentation and point-in-time recovery. It also incorporates hybrid semantic search (vector + full-text) for efficient retrieval, self-governance features for contradiction detection and memory quarantine, and a privacy-first approach with local embedding options.

Quick Start & Requirements

• Cloud Mode (Recommended): Sign up at thememoria.ai for a token. Install the CLI via curl -sSL https://raw.githubusercontent.com/matrixorigin/Memoria/main/scripts/install.sh | bash. Initialize with memoria init -i, selecting "Remote" mode and providing the token.
• Self-Hosted Mode (Docker): Clone the repository, start MatrixOne and the API using docker compose up -d. Install the CLI as above. Initialize with memoria init -i, selecting "Embedded" mode.
• Prerequisites: Internet access for Cloud mode; Docker and MatrixOne for Self-Hosted.

Highlighted Details

• Native Git-level version control for AI memory, including zero-copy snapshots, branches, merges, and rollbacks.
• Hybrid semantic retrieval combining vector and full-text search capabilities.
• Self-governance features for automatic contradiction detection and quarantining of low-confidence memories.
• Privacy-focused design with an option for local embedding models, ensuring data stays on the user's machine.
• Cross-conversation persistence of user preferences, agent facts, and decisions.
• Comprehensive audit trail tracking every memory mutation with snapshot and provenance.

Maintenance & Community

The project is powered by MatrixOne. While specific community channels or contributor details are not explicitly detailed, the project references its research foundation in an arXiv paper detailing the underlying data version control design.

Licensing & Compatibility

Memoria is released under the Apache-2.0 license, which is permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

The provided README does not explicitly detail limitations, alpha status, or known bugs. Setup for self-hosted deployment requires Docker and MatrixOne. Cloud deployment necessitates account registration and token acquisition.