Changelog
What shipped, and when. NautAI is in private preview, so releases land every two to three weeks.
v0.14
Merge checks for eval scores
- Block a merge request when any attached eval drops by more than a threshold you set.
- Merge request pages now show the score change for each suite next to the diff.
- naut diff --stat is about 4× faster on datasets over 10M rows.
v0.13
Feedback grouping
- Production feedback is now grouped by similarity, so one review covers many near-identical failures.
- New webhook source for feedback, alongside the Python SDK.
- Fixed: redaction counts were missing from the commit summary for audio transcripts.
v0.12
Coverage reports and synthetic fill
- Coverage reports show how rows are distributed across any field, with gaps highlighted.
- naut synth --fill generates seeded examples for under-represented values and opens them as a review queue.
- Synthetic rows carry a permanent origin tag that survives merges and exports.
v0.11
Consensus labeling
- Queues can require 2–5 reviewers per row, with adjudication when they disagree.
- Per-reviewer agreement stats in queue settings.
- Guideline files are versioned with the labels they produced.