Download
Memobase is an open source backend system that enables long-term user memory functionality for AI applications by capturing and structuring information about users across interactions. Its design centers on creating user profiles and recording event timelines, allowing AI systems to remember, understand, and evolve in their behaviour toward individual users over time. Instead of relying purely on traditional embedding-based retrieval or RAG systems, Memobase uses profile and timeline structures to deliver memory that reflects user context efficiently and meaningfully. The system focuses on three principal performance metrics: high search performance, reduced large language model (LLM) costs through batch processing techniques, and low latency with minimal SQL operations. Memobase supports integration with existing LLM workflows via APIs and SDKs (including Python, Node, and Go), making it easy to adopt within diverse application stacks.
Features
- Structured long-term memory through user profiles and event timelines
- High performance search and retrieval without embedding-only reliance
- Batch processing buffer per user to lower token and LLM invocation costs
- Easy integration via API plus SDKs for Python, Node/JS, and Go
- Flexible memory configuration to include only relevant user data
- Designed for low latency access (~<100 ms) with few SQL operations
Project Samples
