MemVault: Long-Term Memory Server
MemVault provides a production-grade API designed to equip AI agents with long-term memory capabilities without the complexities of traditional setups. By abstracting the Retrieval-Augmented Generation (RAG) pipeline into a single API endpoint, it eliminates the need for extensive configurations involving Pinecone, embedding pipelines, and chunking logic, allowing for a seamless development experience on your own infrastructure using PostgreSQL and pgvector.

✨ Key Features

• Hybrid Search: Efficiently retrieves memories by assessing a weighted score based on Semantic Similarity, Recency, and Importance.
• Auto-Embedding: Automatically manages text chunking and embedding generation, with support for OpenAI and upcoming local models.
• Self-Hostable: Operates on standard PostgreSQL, ensuring no vendor lock-in.
• Visualizer Dashboard: A built-in frontend tool provides real-time debugging capabilities, allowing users to see the reasoning behind memory retrievals.
• Prisma ORM: Ensures type-safe database access for enhanced reliability.

👁️ Visualizer (The "Debugger" for RAG)

Debugging data retrieval can be complex. The MemVault visualizer simplifies this process by allowing developers to validate their retrieval pipelines in real-time.

• Live Demo

🚀 API Usage Examples

MemVault enables straightforward API interactions that facilitate memory management for AI agents. For instance, to store a memory and automatically manage chunking and vectorization, the following request can be executed:

curl -X POST http://localhost:4000/api/memory/store  
  -H "Content-Type: application/json"  
  -d '{  
    "sessionId": "agent-007",  
    "text": "The user prefers strictly typed languages like TypeScript.",  
    "importanceHint": "high",  
    "metadata": { "source": "user_chat" }  
  }'  

To retrieve relevant memories based on a specific query, use the following command:

curl -X POST http://localhost:4000/api/memory/retrieve  
  -H "Content-Type: application/json"  
  -d '{  
    "sessionId": "agent-007",  
    "query": "What language should I use for the backend?",  
    "limit": 3  
  }'  

Additionally, to clear history for a session, the following command can be executed:

curl -X POST http://localhost:4000/api/memory/clear  
  -H "Content-Type: application/json"  
  -d '{ "sessionId": "agent-007" }'  

🤝 Contributing

Contributions to MemVault are encouraged, with particular interest in support for Ollama embeddings (local) and integration with other vector databases like Chroma.
MemVault aims to enhance the capabilities of AI agents by providing robust long-term memory solutions, making it an essential tool for developers in the AI space.