gbrain is an agent framework and retrieval-augmented generation system that combines a durable task queue, a git-synced vector store, and a knowledge graph engine. It provides a foundation for building AI agents that interact with structured knowledge bases using the Model Context Protocol.
The system synchronizes markdown files from a git repository into a database for high-performance semantic retrieval and creates typed edges between data pages by extracting entity references and wikilinks. It uses a database-backed queue to execute persistent background jobs and tool loops, ensuring reliability and preventing data loss during system failures.
Information retrieval is handled through a hybrid search approach that combines vector embeddings with keyword matching to synthesize cited answers and perform gap analysis. The framework supports the organization of information into a structured knowledge graph using custom schemas derived from filesystem structures.
The project includes tools for benchmarking retrieval quality against standard datasets to evaluate hybrid search performance.