v0.122.3 v0.3.29
medium risk
parruda/swarmOct 7, 2026 → Oct 7, 20260 days between releases
What's New
Adds evidence inspection and new training flags, alongside documentation corrections.
New Features
- Evidence inspection over checkpoint and eval artifacts
- New training flags for --eval and abstention threshold fitting
- Three-seed RLCD vs soft-CE default-loss panel
Improvements
- Cap NumPy below 2 in the onnx extra
- Corrected examples and API descriptions to match implementation
Bug Fixes
- Fix state predict_long's questions reading
- Guard scroll-fade ResizeObserver
- Update stale drawer tests
Upgrade Notes
Review training flags and configuration; ensure documentation examples match current implementation.
The Two Releases
v0.122.3
v0.122.3
Oct 7, 2026· 15 days ago
Features a complete redesign of the Settings page with a new Analytics tab, walkthrough videos, and improved onboarding, along with instant chat queue actions.
Full release notesv0.3.29
v0.3.29
Oct 7, 2026· 15 days ago
A documentation-heavy release that corrects examples and API descriptions to match implementation, alongside new training capabilities and evidence inspection features.
Full release notes