# GraphRAG-rs

A modular Rust implementation of GraphRAG that builds knowledge graphs from documents and answers natural-language questions over them, with server-only, WASM-only and hybrid deployment options and optional GPU acceleration through WebGPU.

GraphRAG-rs builds a knowledge graph from a document or set of documents and answers natural-language questions against that graph, either through a command-line tool or as a Rust library. A quick-start CLI can index a text file and then ask questions against the resulting graph, and an Ollama flag switches from pattern-based to language-model-quality entity extraction.

It supports three deployment architectures: a traditional server-only setup, a WASM-only build that runs entirely client-side in the browser, and a hybrid of the two, with GPU acceleration available through WebGPU. Default configuration uses hash-fallback embeddings and pattern-based entity extraction with a persistent workspace, and Ollama, GLiNER or custom chunking can be enabled through the library's builder for higher-quality extraction.

Building it needs Rust 1.85 or newer with the WebAssembly target, Node.js for WASM builds, and platform-specific build tools; an optional Qdrant vector database can be run through Docker Compose for containerized vector storage.

Website: https://automataia.github.io/graphrag-rs/