MindMemOS is an open-source long-term memory system designed to equip AI agents with persistent, searchable, and evolving memory. It transforms conversations, files, tool traces, and feedback into structured, project-isolated memory assets, significantly enhancing agent capabilities and enabling reusable user profiles and skills across different frameworks.

How It Works

The system centralizes agent experiences into a structured memory store. Its core innovation lies in self-evolving memory mechanisms, including schema learning, "dreaming" for consolidation, and feedback integration to continuously refine memory quality and retrieval. A key feature is the Memory-Skills loop, which converts operational experiences into skill candidates, fostering skill evolution. Memory is designed to be portable across various agent frameworks.

Quick Start & Requirements

Local deployment utilizes uv for dependency management. The primary command to start the full Docker stack is make dev, while make dev-core starts essential services like Qdrant, Neo4j, and Kafka. Configuration requires setting API keys in config/mindmemos/api_keys.yaml and defining model routers (chat_model_router, embed_model_router, rerank_model_router) in config/mindmemos/dev.yaml. The Qdrant vector size must match the embedding dimension. Cloud deployment uses a hosted FastAPI service with an API key obtained from the MindMemOS website. Detailed instructions are available in docs/deploy/instruction.md.

Highlighted Details

• Achieves state-of-the-art (SOTA) performance on the LoCoMo benchmark for conversational memory, scoring 93.64 overall.
• Leads the PersonaMem benchmark for persona-centric memory by approximately 2 points in overall accuracy.
• The "Dreaming" feature demonstrably improves memory quality and reduces memory count during QA tasks.
• Enables the creation and evolution of reusable skills from agent experiences via the Memory-Skills loop.

Maintenance & Community

The project maintains a community presence via a Feishu group for updates, discussions, and support.

Licensing & Compatibility

MindMemOS is released under the permissive MIT License, generally allowing for commercial use and integration into closed-source projects.

Limitations & Caveats

The provided README does not detail specific limitations, alpha status, or known bugs, focusing primarily on features and performance benchmarks.