hsuanshao/agentmemory is a self-hosted AI agent memory service written in Rust. The public materials list no version, and the repository description names its main features as L0-to-L3 layered memory, asynchronous LLM summarization extraction, and retrieval that combines BM25 with vector search. This describes the composition of the service the repository introduces, not the changes of a new version.
The service focuses on handling agent memory in tiers and separating it per user, team, and agent. It retrieves memory using BM25 and vector search together. The repository lets you choose SQLite or PostgreSQL as the store, and states that the coordination layer can run as a single node or use Redis. An OpenAI-compatible LLM can be connected, and the project also includes a Rust core service and a TypeScript monitoring panel. The license is MIT.
The README covers workspace tests and integration tests that use real PostgreSQL and Redis. However, the provided materials present no performance benchmarks or real operational results, so it is hard to judge what quality this feature set delivers in a production environment. It is more reasonable to view it as a service that combines existing elements such as the L0–L3 tiers and hybrid retrieval; it has not demonstrated an original algorithm or operational performance.