Provides long-term memory for LLMs using a hybrid search mechanism backed by a single SQLite file.
Sponsored
Rekal solves the common problem of LLMs forgetting past interactions by offering a robust, persistent memory system. It stores conversational context, user preferences, facts, and procedures within a single SQLite file, allowing LLMs to recall relevant information across sessions. The tool employs a sophisticated hybrid search algorithm combining keyword matching, semantic similarity, and recency to retrieve the most pertinent memories. Rekal operates entirely locally, utilizing local embeddings and requiring no cloud services or API keys, enhancing privacy and ease of use for developers building more capable AI agents.
主な機能
01Local embeddings via fastembed for privacy
024 GitHub stars
03Comprehensive suite of 16 MCP tools for memory interaction
04Stores all memory in a single, local SQLite file
05Hybrid search combining BM25, semantic similarity, and recency
06Conversation management including history, conflicts, and linking
ユースケース
01Help LLMs manage and resolve conflicting or outdated information over time
02Enable LLMs to remember user preferences and facts across multiple conversations
03Allow LLMs to recall project-specific procedures and context for consistent assistance
