Skip to content

Get started

These pages take a clean clone to a running knowledge base, then point at the reference or guide you need next.

QuickstartInstall the kit, ingest the synthetic CC0 corpus, inspect retrieval, and expose the same service over REST and MCP. About 10 minutes.

Tour the repositorySee what you specialize, what you run, and why this is a live template repository rather than a generator.

Choosing Sci RAG KitCompare the kit honestly with LightRAG, PaperQA2, LlamaIndex, and Microsoft GraphRAG.

TroubleshootingStart from the symptom, run sci-rag doctor, and follow the specific check or recovery path.

  1. Run the quickstart with the offline embedder.
  2. Read the repository tour before replacing the demo profile.
  3. Follow Bring your own domain with a small, well-understood corpus, or use a corpus campaign to discover candidates and build a fail-closed open-access manifest.
  4. Review the campaign's rights distribution before ingestion, then run an evaluation before changing retrieval weights or enabling the reranker.

Still deciding? The FAQ answers what this is, who it is for, and why each design decision went the way it did, without sending you to a decision record first.

If Postgres, credentials, or parsing gets in the way, do not guess from an empty result. Run uv run sci-rag doctor; the kit is designed to tell you which layer is missing.