Dictation

Dictation records your voice while you hold or press a shortcut, turns it into text on your Mac, and pastes the text where you were typing.

Turn it on in Settings → Dictation → Enable Dictation. It's off by default. When you turn it on, Tinycast explains what it needs, then macOS asks for microphone access.

Two more steps before your first dictation, both in the same pane:

1. Under Model, click Download. No model comes with Tinycast.
2. Under Commands, record a Shortcut. Dictation has none until you do.

Dictating

Shortcut behavior decides how the shortcut works:

Press esc to cancel. The recording is discarded.

Hold to talk works with a key combination or a single modifier key, like Right ⌥ or 🌐. A single modifier starts recording after you hold it alone for 250 ms. While you hold it, pressing another key or clicking cancels the recording. Double-tap shortcuts only work with Press to start or stop.

You can also use the 🎤 key on your keyboard as the shortcut. See Hotkeys.

While Tinycast listens, a small capsule near the bottom of the screen shows bars that move with your voice. While it transcribes, a wave runs across the capsule. A recording stops by itself after 5 minutes and is transcribed as usual.

Tinycast records from the System default microphone unless you choose another under Microphone.

Play sounds (on by default) plays the macOS dictation sounds when recording starts and stops. Cancelling is silent.

While you record, Tinycast turns your Mac's output volume down to a tenth, and turns it back up when you finish or cancel. If you change the volume yourself in the meantime, Tinycast leaves it as you set it.

Where the text goes

When finished, in Settings → Dictation → Output, picks the destination:

When it pastes, Tinycast reads the text around the cursor and adds a space before or after the dictation where one is needed. Adapt capitalization (on by default) capitalizes the first letter at the start of a field, paragraph or sentence, and lowercases it in the middle of a sentence. If an app doesn't share the text around its cursor, the text goes in as the model wrote it.

Insert while recording is off by default. Turn it on to paste the text in pieces while you speak: at each pause, or after 8 seconds of speech without one. Text already pasted stays, even if you cancel, and isn't corrected later. It has no effect with Copy to clipboard.

Password fields and other protected fields can refuse the paste.

Models

Choose an Engine, then a Model. Models download from Hugging Face. After that, transcription runs entirely on your Mac.

Parakeet is faster and uses less memory. Qwen covers more languages but is slower.

A download shows its progress and a Cancel button. If you switch to another model, the download keeps going. Remove deletes an installed model.

With a Qwen model installed, a Language picker appears. Auto lets the model detect the language. Parakeet has no language setting.

Memory

The model runs in a separate process, Tinycast Dictation, which you can see in Activity Monitor. Release model from memory quits it after the time you choose: 1 minute by default, up to 1 hour, or Never. A released model takes a moment longer to load the next time you dictate.

In AI Chat

While Dictation is on, the AI Chat message field has a microphone button. Click it to start, and click it again to stop and insert the text. It always inserts into the message, whatever When finished says, and the shortcut mode doesn't apply. If no model is installed, clicking it opens Settings → Dictation.

Privacy and permissions

Audio stays in memory. It's never saved to disk or sent over the network.

• Microphone. Asked when you turn Dictation on. If it isn't granted, the pane shows Microphone access required with a button to grant it or open System Settings. Settings → Permissions shows it too.
• Accessibility. Needed to paste into other apps and read the text around the cursor. Tinycast asks for it when you turn Dictation on. Copy to clipboard works without it.

Dictation's settings and shortcut aren't included in backups, so an import can't turn on the microphone.

Messages you might see