GraphRAG is a collection of RAG access patterns which target different layers of a knowledge graph.
It is available on https://graphrag.com
The documentation is structured in accordance with Diátaxis principles.
- "Concepts" explain high-level ideas
- "Guides" are how-to guides that provide problem solving directions
- "Reference" are technical descriptions of what things are
- "Tutorials" are complete examples that bring everything together
- "Glossary" defines relevant terminology
Consider this priority order of contribution:
- Glossary - everything gets a glossary entry
- terse definitions of terminology and names
- reference citations and attribution where appropriate
- for example, define "vector embedding"
- add the entry to
/src/content/docs/appendices/glossary.mdx
- Reference - some glossary entries are expanded on with reference material
- describe what something is
- provide an illustration
- reference related reference material :)
- for example, describe what a vector embedding is
- add a page under
/src/content/docs/reference - use the template
/src/content/docs/reference/template.md
- Guide - some reference material gets a how-to guide
- process-oriented, how to achieve a specific goal
- try to be implementation independent
- reference relevant reference material
- for example, how to perform vector similarity search
- add a page under
/src/content/docs/guides - use the template
/src/content/docs/guides/template.md
- Tutorial - some how-to guides can be combined into tutorials
- walk through a complete example, including data
- still try to be implementation independent
- for example, chat with a PDF
- add a page under
/src/content/docs/tutorials - use the template
/src/content/docs/tutorials/template.md
All diagrams are created using Excalidraw. Original source files
are saved under /media and exported SVG images in /src/assets/images.
Thanks to Neo4j for sponsoring this work.
GraphRAG Pattern Catalog by Neo4j, Inc is licensed under CC BY 4.0
