Quickstart¶
From install to a cited answer in about five minutes, using the document_qa
recipe — a ready-to-run project you point at your own documents. It ships a
filled-in schema, extractor, and pipeline, so there's no code to write for a first
answer. For the full command surface and customization, see the
usage guide.
1. Install¶
openai pulls the SDK for the recipe's default ingestion model; ingestion pulls
the document loaders. Prefer Claude or Gemini? Swap the extra (anthropic /
gemini) and change the model line in step 4.
2. Scaffold a project¶
3. Add your API key¶
Reigner auto-loads this .env from the project root — no export needed.
4. Drop in a corpus¶
Put a document or two in library/raw/. For a throwaway first run, one tiny
markdown file is enough:
The recipe's default ingestion model is openai:gpt-5.5. To use another provider,
edit the model line in extractors/my_extractor.py.
5. Compile the corpus¶
$ reigner ingest
# representative output
✓ loaded 1 document
✓ extracted 1 entity → library/artifacts/
✓ built BM25 index → search-index/documents.json
Want to see the plan first without spending tokens? reigner ingest --dry-run
lists the file → loader mapping and exits.
6. Ask a question¶
$ reigner chat
› What does Orbit cost?
# representative output
✓ Retrieved 2 calls · 6.1k tok · $0.004 · 1.8s
Sources
[1] faq.md section=pricing
╭─ Answer ──────────────────────────────╮
│ Orbit is $8 per user per month. [1] │
╰────────────────────────────────────────╯
The answer is grounded in a retrieved source and cited inline as [1]. Swap in
your own documents and re-run ingest — new files compile, unchanged ones are
skipped.
Where to go next¶
- Usage guide — every command and flag, plus the full configuration reference.
- Large or mixed corpora — map-reduce extraction for big PDFs and documents that don't share a shape.
- The
code_navigatorrecipe — one agent reading across several repositories, with no ingestion step. - Observability — export the agent loop as OpenTelemetry spans.