Rekal, an MCP server providing long-term memory for LLMs. It stores memories in SQLite and retrieves them using hybrid search (BM25 + vectors + recency decay).

The stateless nature of current LLM interactions presents a significant friction point for users, requiring constant re-contextualization. Rekal directly addresses this by implementing a local, long-term memory solution for LLMs, stored in a single SQLite file. Its hybrid retrieval mechanism (BM25, vectors, recency decay) is critical for efficient and relevant memory recall, enhancing conversational continuity and reducing token waste. The "local embeddings" and "no API keys" approach prioritizes privacy and cost-efficiency, appealing to developers and users concerned about data egress and recurring cloud expenses. This positions Rekal as an essential component for building more intelligent, persistent, and user-friendly AI agents, enabling a new class of personalized and context-aware applications.