Voice dictation: push-to-talk into any input

atrium dictates into any focused input — a terminal, the composer, Monaco, or the markdown editor — on macOS and Windows.

Two modes

Push-to-talk — hold a key, speak, release. The text is finalized and pasted into whatever has focus. The default is right Option on macOS. On Windows it is right Alt, or right Ctrl when a keyboard layout uses right Alt as AltGr. Rebind it in Settings; each platform keeps its own binding. Windows reserves left Ctrl for shortcut hints.

While you hold, atrium suppresses silence and punctuation endpointing so the whole hold is treated as one utterance, and a 500 ms tail-capture grace keeps a key released a touch early from clipping your last word. The overlay shows a release-to-insert hint.

Hands-free — Cmd+Shift+M (Ctrl+Shift+M on Windows) toggles dictation on and off. In this mode atrium segments on silence, so you can speak in several passes without touching a key. A mic button sits in the composer, left of send, with the bound push-to-talk key in its tooltip.

Voice submits

While dictating into any input:

• Saying "Full stop." submits immediately (equivalent to Enter).
• Saying "Send" or "Enter" alone, with no surrounding text, also submits.

Settings

Settings → Voice Dictation:

• Microphone device picker and an input volume slider with a live level meter.
• Push-to-talk key — rebindable, captured natively. A bare modifier (like right Option) or a full combo both work.
• Recognition engine.
• macOS permission status for Microphone, Accessibility, and Speech Recognition, plus the macOS Dictation setting and on-device model readiness. The associated buttons open the relevant system settings.
• Windows model controls for live transcription and the basic dictation fallback, with download status and recovery actions.

macOS recognition

Apple Speech Recognition uses on-device processing by default. Process speech on-device in Settings controls this choice. Keep macOS Dictation enabled and let its language model finish downloading before starting.

Turning on-device processing off allows macOS to use Apple's server path, which can hit a shared request limit across running instances. If dictation stops working, check the permission and model status in Settings → Voice Dictation.

Windows live transcription

Live transcription uses a local Nemotron speech model to show words while you speak and insert the transcript when you release push-to-talk. Choose Download in Settings → Voice Dictation → Live transcription, or accept the one-time download offered when you first dictate. The prompt shows the required size.

The download shows progress, can be cancelled and resumed, and checks its files before using them. Repair replaces damaged model files. Once it is ready, your next dictation uses live transcription; an already-running dictation is not interrupted.

Until that model is ready, basic dictation uses Windows Speech Recognition for live text and a local Whisper model for the final transcript. The Whisper model downloads on first use. Basic live text uses the system default microphone, while the final transcript uses the input device selected in atrium. The Nemotron path uses the selected input for both.