Repository navigation
Connect dictation in the composer #38
Description
Add dictation in the composer. useMic is unused.
Tracked by #6. Migrated from TODO.md — Port from old Ace.
- Channel composer dictation through
webkitSpeechRecognition, desktop usage strings and audio-input entitlement (feat: transcribe dictation into the channel composer #473) - Verify end to end in the desktop app: first-run mic/speech prompts, glow alongside recognition
- Dictation in the dashboard's new-channel composer (feat(ui): dictate into a composer that already has text, and in the dashboard #477)
- Append dictation to existing text (mic affordance when the composer isn't empty) (feat(ui): dictate into a composer that already has text, and in the dashboard #477)
Activity
iamnbutler commentedon Oct 15, 2026
Current state on main:
useMic(packages/ui/src/hooks/use-mic.ts) only opens the microphone and returns anAnalyserNodefor the composer's recording glow;ChatComposer'smicprop defaults to false and the app never enables it (apps/app/src/dashboard.tsxpassesmic={false}). Nothing transcribes audio. Dictation still needs a speech-to-text source (WebKit's speech recognition in the desktop shell, or a provider model) and the desktop app's microphone permission.iamnbutler commentedon Oct 22, 2026
From the 2026-10-22 map study on Canary 0.0.51.85: Record raised both macOS prompts with Ace's reasons ("Ace turns your dictation into message text.", "Ace listens while you dictate a message."), and the level meter and Stop worked. The Mac's output was muted, so no speech reached it, and stopping after silence gave no "No speech detected" or similar; the composer just went back to how it was. Transcription itself is still unverified (^511).