01Customizable Embedding Models: Supports overriding the default embedding model for multilingual corpora and specific use cases, with automatic prompt format adjustment.
020 GitHub stars
03MCP Server Integration: Exposes a Model Context Protocol (MCP) server for seamless tool-call integration with any MCP-compatible agent runtime.
04Deterministic Privacy: All inference (embedding, reranking, query expansion) runs locally on GGUF models via node-llama-cpp, ensuring data never leaves the edge.
05Hybrid Precision Retrieval: Combines BM25 symbolic retrieval, neural vector similarity, and LLM cross-encoder reranking with Reciprocal Rank Fusion (RRF) for high accuracy.
06Agent-Native Design: Architected for child_process invocation with structured output (JSON, files, CSV, XML) for autonomous agent consumption.
