FusionPact is a groundbreaking retrieval engine purpose-built for AI agents, addressing the limitations of traditional vector databases by integrating three powerful retrieval paradigms: HNSW vector search, LLM-driven reasoning-based tree retrieval, and keyword search, all intelligently fused through Reciprocal Rank Fusion. Beyond just retrieval, it offers a sophisticated, multi-type agent memory system (episodic, semantic, procedural, shared), robust multi-agent orchestration for seamless coordination, and functions as an MCP server, enabling direct integration as persistent memory for various AI agents. This comprehensive, local-first solution empowers developers to build more intelligent, context-aware, and collaborative AI applications.
Key Features
01Hybrid Retrieval Engine (Vector, Reasoning-based Tree, Keyword search with Reciprocal Rank Fusion)
02Reasoning-Based Tree Index for structured document navigation using LLMs
03Agent Memory System with episodic, semantic, procedural, and shared memory types
04Multi-Agent Orchestration for coordinating and communicating between AI agents
05MCP (Model Context Protocol) Server for agent-native persistent memory integration
060 GitHub stars
Use Cases
01Developing AI agents that require advanced, context-aware memory and retrieval capabilities.
02Orchestrating complex multi-agent systems with shared knowledge and inter-agent communication.
03Building sophisticated Retrieval Augmented Generation (RAG) pipelines for large language models that need precise information retrieval.
