AI agents often struggle with recalling past context, solutions, and project specifics, restarting 'cold' with each session. MahoRAGa addresses this by offering a local, graph-powered memory system that allows agents to build a persistent knowledge base over time. It stores errors, solutions, concepts, and project details in an embedded Kuzu graph database, enabling semantic search and cross-project knowledge transfer. This ensures agents learn from past interactions, never solving the same problem twice, and operate with a rich, cumulative understanding, all while keeping your sensitive data entirely on your local machine with no cloud dependencies.
Características Principales
01Kuzu embedded graph database with strict referential integrity
02Hybrid semantic search with `all-MiniLM-L6-v2` embeddings for enhanced recall
033 GitHub stars
0425+ tools exposed via FastMCP over stdio/SSE for client integration
05Persistent artifact storage for datasheets, configs, logs, and code snippets
06Automatic aggregation of sessions into daily summaries with garbage collection
Casos de Uso
01Equipping AI coding agents with long-term semantic memory to recall past errors, solutions, and project context
02Facilitating cross-project knowledge transfer, allowing concepts learned in one project to be instantly available in others
03Providing queryable activity intelligence, including daily summaries, session histories, and project timelines for agents and users
