GraphRAG in Rust with PDF vision, MCP, and six query modes

What it does EdgeQuake implements the LightRAG algorithm in Rust. It ingests documents, uses LLMs to extract entities and relationships, and stores the result in PostgreSQL AGE and pgvector. A React frontend and an OpenAPI REST API let users query the graph with modes that mix vector similarity and graph traversal.

The interesting bit For a pre-1.0 project, it bundles an unusually complete stack: embedded pdfium for PDF text extraction, an optional vision mode that sends pages to GPT-4o or Claude as images, multi-tenant workspace isolation, and an MCP server so agents can query the graph directly. It is essentially trying to be backend, frontend, and agent integration layer all at once.

Key highlights

• Six query modes from naive vector search to hybrid graph traversal; its own benchmark table claims hybrid latency under 200 ms.
• PDF pipeline handles standard text extraction or routes scanned documents through multimodal LLMs, with fallback to text if vision fails.
• Knowledge Injection accepts domain glossaries and synonym mappings that expand queries without appearing as result citations.
• Per-workspace custom entity types (up to 50) with presets for healthcare, legal, manufacturing, and other domains.
• Ships with a React 19 frontend using Sigma.js for interactive graph visualization.

Caveats

• Performance benchmarks compare against an unnamed “Traditional RAG” baseline, so treat the improvement multiples as self-reported.
• The README is truncated in the provided sources, leaving some setup details unclear.
• Version 0.12.x moves fast enough that the maintainers document a multi-step local release gate and separate repos for the backend, WebUI, and MCP server.

Verdict A solid candidate if you want a full-stack GraphRAG prototype with PDF ingestion and a UI out of the box. Look elsewhere if you need a minimal, composable library to drop into an existing pipeline.