Push-to-Talk voice input

Basic flow:

1. Hold the Push-to-Talk shortcut.
2. Speak naturally.
3. Optionally draw on screen while holding the shortcut.
4. Release the shortcut.
5. Picky transcribes speech, captures context, and sends the request to Pi.
6. Replies appear in the Picky cursor bubble, Hub → Conversation, or the Pickle HUD depending on routing.
7. If TTS is enabled, Picky also reads replies aloud.

Clamshell mode: Apple silicon and T2-equipped Mac laptops hardware-disconnect the built-in microphone when the lid closes (Apple Platform Security). Select an external microphone in System Settings → Sound → Input before using Push-to-Talk with the lid closed. If Azure STT receives only digital silence, Picky skips the network request and reports that no speech was detected instead of accepting a fabricated transcript.

Voice interruption

Starting a new voice input interrupts an in-progress spoken response. This lets you quickly correct, redirect, or continue without waiting for TTS to finish.

Dictating into a Pickle

Push-to-Talk goes to the main agent unless a Pickle is armed as the input target (see 4.3). Hovering over a Pickle card does not change the target.

To speak to one Pickle, use the microphone button on the left of that Pickle’s composer toolbar:

1. Click the microphone, or press Cmd + D while the Pickle’s composer has keyboard focus. The line above the composer shows Listening and the elapsed time.
2. Click it again or press Cmd + D again to stop. Picky transcribes the recording and adds the text to the end of the composer, after anything you already typed. Nothing is sent yet.
3. Review or edit the text, then send it as usual. While the Pickle is running, Send queues it as a follow-up.

Press esc while listening to cancel without changing the composer. If nothing was recognized, the composer stays as it was and the status line says so. Only one Pickle can dictate at a time, and Push-to-Talk is ignored while a composer is dictating.

Screen-context target to a Pickle

You can explicitly arm a Pickle as the target for the next Picky screen-context input:

1. Open a Pickle card.
2. Click the Pickle/Pi badge in the header, or press Cmd + K.
3. The composer shows: “The next Picky voice or quick text input goes to this Pickle”.
4. To keep targeting the Pickle until you clear it, press and hold that badge for about one second, or right-click / Control-click its Dock icon and choose Pin Picky Input to This Pickle.
5. Use Push-to-Talk or Quick Input.
6. The next screen-context input is sent directly to that Pickle.
7. By default, the input is delivered as a follow-up so it waits for the Pickle’s current turn to finish. Change Hub → Settings → Accounts and agents → Main Agent → Armed Pickle delivery to Steer if you want armed Push-to-Talk and Quick Input to interrupt the current turn instead.
8. A one-shot target clears after delivery. A pinned target stays armed until you click the badge again, choose Unpin Picky Input from its Dock context menu, or pin another Pickle.
9. When that armed delivery includes at least one screenshot, the Pickle receives the visual annotation DSL for that response only and may draw grounded RECT, LINE, or PATH annotations on the captured screen. Text-only deliveries never enable the DSL.
10. Visual Pickle replies use the same narration pipeline as the main agent: when TTS is enabled, each completed sentence is read in source order and its annotation is revealed when that sentence starts. Providers without incremental playback read the clean final reply once; with TTS disabled, sentences and annotations still appear progressively without audio. The DSL tags themselves never appear in the Pickle conversation.
11. Clearing the one-shot armed badge does not cancel the response already in flight. Its turn-scoped visual capability remains valid unless a newer screen-context submission supersedes it, the turn is cancelled, or the captured scene no longer matches.