Graph-RAG

runic.rag is a thin Graph-RAG SDK built on runic.ogm and runic.migrate. It extracts entities and relations from your documents into a knowledge graph, then answers questions over that graph with hybrid (vector + fulltext + graph expansion) retrieval and inline citations. The whole pipeline sits behind one GraphRAG facade and runs on FalkorDB or Neo4j out of the box.

Graph-RAG

• What is Graph-RAG? — The big picture: how documents become a graph and how that graph answers questions.
• Quickstart — Install runic.rag, ingest a document, and ask your first question — all on one page.
• Ingesting documents — How the graph is built: chunking, extraction, embedding, and entity resolution.
• Retrieval & answers — How questions are answered: the local, hybrid, and auto modes, and the Answer shape.
• Designing & optimizing ontologies — Tune the entity vocabulary to your domain for sharper extraction and retrieval.
• Evaluating quality — Measure faithfulness, relevancy, and recall with deepeval before you ship.
• Configuration & deployment — Every RagSettings knob, backend selection, and the dev-to-prod schema lifecycle.
• Writing custom ports — Swap any pipeline stage — chunker, extractor, retriever, synthesizer — with your own adapter, and what to keep in mind when you do.
• Document parsing with Docling — Optional structure-aware parsing of PDF/DOCX/PPTX/XLSX/HTML and scanned images, in-process or via docling-serve.
• API Reference — runic.rag — the facade, domain objects, ports, and default adapters.

---