MindMemOS - Huawei Noah's Ark's Open-Source AI Agent Memory Operating System

What is MindMemOS

MindMemOS is an AI Agent memory operating system open-sourced by Huawei Noah's Ark Lab, solving the pain point of Agents forgetting as soon as they finish. The system decouples memory from individual Agents into an independent asset that can be reused across frameworks, using an entity-attribute-time three-dimensional structure to preserve the latest state and the complete evolution trajectory. MindMemOS continuously evolves skills through Dreaming offline memory organization, Feedback mining of user correction signals, and Skill Evolution based on real execution trajectories, achieving co-evolution of memory and the system.

MindMemOS's Main Features

• 3D Memory Structure: Organizes memory using the three-dimensional coordinates of entity, attribute, and time, tracking attribute evolution trajectories and semantic associations between entities.
• Dual-Mode Memory Extraction: MindVanilla provides open-domain, template-free rapid ingestion; MindSchema presets entity and attribute rules for different business scenarios.
• Compact Search Retrieval: The outer Agentic module plans retrieval paths, while the inner layer performs multi-path searches such as entity reverse lookup, multi-hop relationship expansion, and timeline tracking.
• Dreaming Offline Organization: The offline stage identifies duplication, conflicts, and evolutionary relationships, merges redundancy, archives invalidated versions, and supplements higher-order patterns.
• Feedback Closed Loop: Supports explicit and implicit feedback, and after determining the signal type, performs add, update, archive, delete, or strengthen operations.
• Skill Evolution: Automatically extracts success strategies and failure patterns based on real execution trajectories, generates Skill modification plans, and iterates versions.

Follow WeChat and reply "Open Source" to join theAI Open Source Project Communication Group

How to Use MindMemOS

• Clone the repository and prepare the configuration：cp .env.example .env and cp config/mindmemos/dev.example.yaml config/mindmemos/dev.yaml, and set chat_model_router, embed_model_router, rerank_model_router, and API key in the configuration file.
• Start the service: Run make dev to start the complete Docker dependency stack and FastAPI service (default port 8000), or make dev-core to start only the core dependencies.
• Install and configure the SDK：pip install mindmemos-sdk, run mindmemos auth to configure base_url, api_key, and user_id.
• Add and retrieve memories: Call client.memory.add() through the SDK to write conversations or facts, and use client.memory.search() to perform retrieval.
• Submit feedback: Call client.memory.feedback() or CLI mindmemos memory feedback to submit explicit or implicit correction signals.
• Trigger Dreaming: Run mindmemos memory dreaming or call the dreaming API for offline memory consolidation.
• Register and evolve Skills: Register local Skills via mindmemos skill register , and use mindmemos skill evolve to trigger version evolution based on execution trajectories.

MindMemOS's Project Address

• Project Website：https://mindmemos.cn/
• GitHub Repository：https://github.com/mindscale-noah/MindMemOS

MindMemOS's Core Advantages

• Transferable Memory: Decouples the Agent access layer, memory algorithm layer, and memory structure layer, so memory can be reused across applications as an independent long-term asset.
• Leading Structured Recall: Achieves an overall accuracy of 94.03% on LoCoMo (SOTA) and 70.63% on PersonaMem, significantly outperforming solutions such as Mem0, MemU, Zep, and EverOS.
• Less is more in memory quality: While compressing active memory by 19.4%-23.5%, Dreaming improves QA accuracy by up to 10.3 percentage points.
• Skill self-evolution: An unoptimized initial Skill actually performs worse than a no-Skill baseline, while MindEvolve-Sup raises the task success rate from 51.3% to 57.2%, proving the value of distilling experience into rules.
• Feedback-driven evolution: The Feedback mechanism applies user correction signals back to memory extraction and retrieval strategies.
• Multiple access methods: Supports local deployment (Docker + FastAPI), cloud API, Python SDK, CLI, and the OpenClaw plugin.

Comparison of MindMemOS with similar competitors

Application scenarios of MindMemOS

• Personalized AI assistant: Retains user preferences, project background, and historical pitfalls across sessions, avoiding the need to retrain the Agent each time.
• Multi-Agent collaboration platform: The same memory repository is shared and reused by different Agents such as OpenClaw, Hermes, Claude Code, and OpenHands.
• Automatic algorithm design (LLM4AD): Systematizes effective experience and pitfall-avoidance guides from algorithm search into a retrievable, reusable three-layer memory.
• Complex office automation: Based on Skill Evolution, continuously optimizes execution strategies for repetitive tasks such as spreadsheet processing and document editing.
• Long-term user profile construction: Tracks the trajectories of user attributes over time, supporting personalized recommendations and social planning.