memU

One-Minute Overview#

MemU is a future-oriented agentic memory system that receives multimodal inputs (conversations, documents, images), extracts them into structured memory, and organizes them into a hierarchical file system supporting both embedding-based (RAG) and non-embedding (LLM) retrieval.

Core Value: Transforms unstructured data into retrievable, evolving structured memory, enabling AI systems to better understand and leverage historical information.

Quick Start#

Installation Difficulty: Medium - Requires Python 3.13+ and OpenAI API key

# Basic test (no database required)
export OPENAI_API_KEY=your_api_key
cd tests
python test_inmemory.py

# PostgreSQL test (requires pgvector)
docker run -d --name memu-postgres -e POSTGRES_USER=postgres -e POSTGRES_PASSWORD=postgres -e POSTGRES_DB=memu -p 5432:5432 pgvector/pgvector:pg16
export OPENAI_API_KEY=your_api_key
cd tests
python test_postgres.py

Is this suitable for me?

• ✅ AI Assistant Development: Intelligent assistants needing to remember user preferences, habits, and conversation history
• ✅ Customer Service Bots: Systems requiring memory of customer information, issues, and resolutions
• ❌ Simple Information Storage: Applications needing only basic CRUD operations without complex semantic understanding

Core Capabilities#

1. Hierarchical File System - Structured Memory Management#

Three-layer architecture (Resource → Item → Category) provides full traceability with increasingly abstracted views at each level.

Real Value: Memory evolves from raw data to structured information to categorized summaries, making it easy to retrieve and analyze.

2. Dual Retrieval Methods - Flexible Memory Querying#

Supports both RAG (embedding vector search) and LLM (deep semantic understanding) retrieval strategies.

Real Value: Flexibly choose between speed and precision based on your application's needs, balancing performance with depth of semantic understanding.

3. Multimodal Support - Unified Content Processing#

Unified processing of conversations, documents, images, videos, and audio inputs.

Real Value: AI systems can learn and remember from various sources, building comprehensive knowledge bases.

4. Self-Evolving Memory - Continuous Memory Optimization#

Memory structure automatically adapts and improves based on usage patterns.

Real Value: The memory system becomes more intelligent and efficient over time, reducing maintenance costs.

Technology Stack & Integration#

Development Language: Python 3.13+ Key Dependencies: OpenAI API (with configuration support for custom providers) Integration Method: API / SDK

Ecosystem & Extensions#

• Multiple LLM Providers: Supports OpenAI, Qwen, OpenRouter, and other LLM services
• Deployment Options: Both self-hosted and cloud service options
• Custom Extensions: Flexible configuration to add custom LLM and embedding model providers

Maintenance Status#

• Development Activity: Actively developed with continuous updates and feature enhancements
• Recent Updates: Recent major feature updates and performance optimizations
• Community Response: Active development community with Discord and Twitter support channels

Commercial & License#

License: Apache 2.0

• ✅ Commercial: Commercial use allowed
• ✅ Modification: Modification and distribution allowed
• ⚠️ Restrictions: Must retain original license and copyright notices

Documentation & Learning Resources#

• Documentation Quality: Comprehensive with full API documentation and multiple practical examples
• Official Documentation: SERVICE_API.md (includes complete API reference)
• Example Code: Complete example code for multiple use cases