You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reduced peak memory during post-meeting refinement: the recognition pass now reads audio window-by-window from disk instead of holding whole tracks in memory, and native embedding extractors are released between tracks.
Raised the refinement duration limit to 4 hours, covering the vast majority of meetings.
Microphone speakers are now matched against the voiceprint library during refinement instead of always being labelled "local user", so local participants get real names too.
More detailed meeting summaries: the English and Chinese notes prompts now develop every substantive topic and preserve more of the discussion rather than compressing it away.
Model library capability tags now sit inline next to the model name for a tighter layout.
Fixes
Fixed the model download queue card becoming unclickable while progress was updating; the card now updates progress in place and only rebuilds when items or button states change.
Fixed speaker labels from multiple tracks (mic and system) being merged into one person after sorting by timestamp; unmatched speakers are now namespaced per track and shown with readable "Mic / Remote Speaker N" labels.
Masked API key placeholders now reflect the stored key length.
Smoothed transcript scroll restoration when reopening a meeting detail.
Downloads
macOS (Apple Silicon):Brevia-1.0.7-arm64.dmg, signed with Developer ID and notarized by Apple.