Quickstart

Get from zero to a first related-issue analysis. Assumes the prerequisites are met (uv, authenticated yt, reachable Ollama).

1. Check connectivity

uv run yt-issue-reviewer doctor --ollama-host http://<host>:11434

Expect YouTrack (yt): OK and Ollama embeddings (nomic-embed-text): OK.

2. Run an analysis

uv run yt-issue-reviewer analyze \
  --project PROJ \
  --state open \
  --min-score 0.6 \
  --ollama-host http://<host>:11434 \
  --db ./yir.db

This ingests issues (cached in ./yir.db), embeds them, scores relatedness (semantic + structural), groups the high-scoring pairs, and prints ranked groups with evidence. Existing YouTrack links are excluded from new findings.

Add --label to get a generated one-line theme label per group (marked (generated) — it never affects scores or membership).

3. Re-display or browse results

uv run yt-issue-reviewer show --db ./yir.db     # re-render the latest run, offline
uv run yt-issue-reviewer runs --db ./yir.db     # list runs + settings + staleness
datasette ./yir.db                              # browse all tables in a web UI

Repeat runs over unchanged issues are near-instant thanks to the embedding cache.

What next

• Every command and flag: CLI reference.

• Tune models, weights, threshold: configuration.

• How scoring works: architecture.

• What leaves your infrastructure (nothing): privacy & security.

For the full validated walkthrough (including degraded-mode behavior), see specs/001-related-issue-finder/quickstart.md.