Voice input in the AI assistantProElite

In the “Voice input” section, choose how speech is recognized:

• Off — recognition is not used.
• Browser recognition — the browser recognizes speech. An internet connection is usually required.
• Cloud (your key) — recognition through a connected provider. Requires internet and uses the provider’s resources.
• On this device — recognition runs on your computer and does not require an internet connection. Your connected model turns the speech into tasks. A signed-in account is required.

For cloud recognition, you can choose a processing model: fast, accurate, or the most precise one. The list of models depends on what the provider offers. The hint under each model explains the difference and the cost.

If you choose “On this device,” download the model you need:

1. The catalog lists available models with language, size, and status.
2. Click “Download” to start. While it downloads, you see “Queued” or “Installing...”. If the download fails, click “Retry.”
3. Downloaded models are grouped under “Downloaded.” Next to them you can see how much disk space they take.
4. Remove a model you no longer need with “Delete.”

If the catalog is unavailable, the app says: “Voice model catalog is unavailable — check your connection.”

The “Model that creates tasks from text” setting chooses the AI that turns dictated speech into tasks. The list contains the models enabled in the “AI chat models” section: your own provider models and local models. The recommended model is selected by default, and you can restore it with the reset button.

If the selected model is unavailable, the default model handles the task, and a note says so. In Agent mode, the agent handles lists itself, so a separate model is not needed.

When you list several errands in one phrase, the app can arrange them in different ways. In “How to split a list of one errand,” choose:

• Auto — the AI decides: a shopping list becomes a task with a checklist, short errands stay in the title, and errands in different places become separate tasks.
• As a checklist — one task with a checklist of the listed errands.
• As separate tasks — one task per errand.
• As one task — all errands in one task without a checklist.

The “Model reasoning level” setting affects speed and accuracy. There are five levels: the higher the level, the more accurate the result and the later the tasks appear. The hint under each level explains the difference.

The “Dictation sounds” toggle plays a signal a few seconds before recording ends and when it stops.

• In the AI Assistant chat — the microphone button next to the input field.
• In quick task adding — to create tasks by voice.
• In task AI actions — the “Dictate changes to the task” item.
• In the tray widget — the “Dictate tasks” button.
• With the global dictation hotkey. Turn it on or set your own combination in Settings → “Tray and menu bar” → “Dictation hotkey.”