Fullscreen Video, Custom Vocabulary & Localization
This release brings a proper fullscreen video experience, transcription hints to improve accuracy on domain-specific content, and complete localization coverage for everything added in recent releases.
New Features
Fullscreen Video Mode lets you double-click any video or hit the expand button to enter a dedicated fullscreen player. Controls auto-hide after a few seconds so you can focus on the content. Live subtitles appear at the bottom with speaker names and colors from your diarization. All the usual keyboard shortcuts work: space to pause, arrow keys to seek, escape to exit.
Custom Vocabulary (Hotwords) gives you a way to feed domain-specific terms to the transcription model so it doesn't fumble proper nouns and technical jargon. Add comma-separated words at the user level as a default, or override them per tag or folder. Useful when you have recurring terms the model consistently gets wrong.
Initial Prompt works alongside hotwords to steer the model's output style. Provide a sentence of context like "This is a medical consultation between a doctor and patient" and the transcription model will lean into that domain's vocabulary and conventions.
Video Passthrough adds the VIDEO_PASSTHROUGH_ASR=true option for ASR backends that can handle video files natively. When enabled, the raw video file goes straight to the backend without extracting audio first. Useful if your ASR service does its own video processing.
Upload Disclaimer Modal lets admins configure a disclaimer that appears before every upload. There's also a custom banner text option in admin settings for site-wide announcements.
Tag and Folder Modal Redesign moves settings into a tabbed layout so hotwords, initial prompt, and ASR configuration each get their own space without cluttering the main form.
Localization
Every string added in the last several releases is now properly localized. This covers incognito mode UI, hotwords and initial prompt labels, the upload disclaimer, fullscreen video controls, group management tables, SSO account linking, auto-summarize toggle, color scheme names, and all the toast messages and error states in the JS composables. All six languages (English, Spanish, French, German, Chinese, Russian) have the new keys with English as placeholder for the non-English locales.
Bug Fixes
Upload notifications now appear in the correct order. Speaker snippet extraction works properly for video files with AAC audio streams. The chat textarea stays enabled and focused while the AI is streaming a response. Adding files to the upload queue while others are already processing no longer blocks the queue. Duplicate detection hashing now runs before format conversion so it catches actual duplicates. Markdown list formatting in the docs has been corrected.
Compatibility
Backwards compatible with v0.8.x. The new VIDEO_PASSTHROUGH_ASR option is opt-in. No breaking changes.