v0.7.0a5
Pre-release
Pre-release
What's Changed
- Add skill support for RealtimeVoiceChannel, fix release blockers
- Add guardrails example with multi-layer safety hooks
- Update Gemini Live default model to gemini-3.1-flash-live-preview
- Reduce subtitle font size to fit 640px frame
- Fix: call start_capture() to begin webcam frame capture
- Replace snapshot with MJPEG HTTP server for live video viewing
- Fix: replace cv2.imshow with file snapshot (headless-safe)
- Add translation debug logging and clearer instructions
- Fix: add missing attach_channel for voice (again)
- Fix subtitle example: add cv2 display window and video join
- Fix translate function: use AIContext, not keyword args
- Fix: restore attach_channel for voice — join() requires it
- Fix subtitle example: add VAD_MODEL env var, remove duplicate session
- Rewrite subtitle example using shared helpers
- Rewrite subtitle example: real webcam + mic + STT + translation
- Final review fixes: BGR color bug, missing tests, docs completeness
- Add overlay example, changelog entry, and docs guide
- Production hardening: overlay DRY, thread safety, bug fixes, tests
- Add rich video overlays: text, images, tables on live video frames
- Fix async_delivery docstring to match implementation
- docs: update changelog for delivery backend
- Add delivery backend docs, examples, and API reference
- Production hardening: fix critical bugs, eliminate DRY, expand tests
- Fix review issues: hook tests, Redis close() cleanup, log levels
- Add Redis delivery backend (Phase 2) and worker-side delivery hooks
- Add persistent delivery backend (Phase 1) and auto-delegate test coverage
Full Changelog: v0.7.0a4...v0.7.0a5