Skip to main content
A floating always-on-top chat you open with a hotkey. Ask by voice or by typing, get streaming answers, and have them read back aloud without leaving the app you are in.

Shortcuts

Both are rebindable in Settings → Assistant → Shortcuts.
The assistant hotkey follows the same Push to talk setting as dictation. It is on by default, so you hold the hotkey while you talk.Add Shift for the opposite mode. Or tap the assistant’s lock key mid-recording (Space on Windows, Shift elsewhere) to go hands-free and let go of the hotkey. See Hands-free.

What you can do

Images in the conversation, both screen captures and files you attached, appear as inline thumbnails you can click to enlarge.
Type /summarize to compact the whole conversation into a summary that replaces the transcript. It is the one slash command.

Header controls

The header carries a persona switcher (see Profiles) plus quick toggles for screen vision, web search, and voice replies. Each toggle applies to the conversation you are in, so you can turn one on without opening Settings. The screen-vision toggle is sticky. Once it is on, your screen is attached to every message until you turn it off. It is hidden entirely when screen access is set to Off.

Appearance

Three knobs in Settings → Assistant → Panel appearance, with a live preview as you change them. The panel is dark only. There is no accent color and no light option, so those three settings are the whole surface. App-wide theme and text size for the settings window live in General settings.
You can drag the panel’s edges to resize it. A manual resize wins for the rest of the session, and the next app start goes back to the preset.Whichever size you pick is clamped to the monitor the panel is on, so a large preset shrinks to fit a small display instead of running off the edge or hiding under the taskbar.
At 100% the panel is fully solid. Lower values let the desktop show through, which helps when you are reading an answer against the thing you asked about.

Models and providers

Run it offline, on your own server, or on a cloud key.

Behavior and memory

Conversation context and reply length.
Last modified on August 7, 2026