# Voice dictation in every prompt box, with smart cleanup

> Every prompt box has a mic. Speech streams to live transcription, then a cleanup pass removes filler and applies your spoken edits.

Jul 8, 2026. Topics: Specialists. https://www.generalinput.com/changelog/2026-07-08-voice-dictation

Describing a workflow out loud is faster than typing it, but raw transcripts are full of false starts. Every prompt box in the dashboard now has a mic button: words appear as you speak, and when you stop, a cleanup pass turns the ramble into the prompt you meant and puts it in the box. It never submits for you. Read [why talking to agents is faster than typing](/blog/voice-dictation-faster-than-typing-ai-agents).

### How it works

- **Streaming transcription.** Your browser streams audio directly to AssemblyAI with a short-lived token; audio never passes through our servers, and a live strip shows the partial transcript.
- **Cleanup that edits.** On stop, your workspace's utility model removes filler, keeps only the corrected version when you change your mind mid-sentence ("make it a Google Sheet, no wait, a Slack message"), and applies spoken instructions like "delete the part about revenue charts" to what's already in the box.
- **Never loses words.** If cleanup fails, the raw transcript is added instead.
- **Everywhere you type.** Chat, workflow, app, and skill building, onboarding, and the new-resource dialogs.
- **Six languages.** English, Spanish, German, French, Portuguese, and Italian, detected automatically.

Transcription costs about $0.19 per audio hour, prorated per second.