1.3.2
- Typing in the chat box now works on sites that steal focus: some pages close their menus on every click and remove focus from anything they don't recognise as a text field, which made the question box lose focus as soon as it was clicked. The panel is now isolated from the page's scripts (clicks, keys and focus from the panel no longer reach them), and keyboard shortcuts of the page no longer swallow what you type.
- Settings, Model field (from 1.3.1): the dropdown now lists every model loaded from the API, not just one to three.

1.3.1
- Settings, Model field: after "Load models from the API", the dropdown showed only one to three models, because the browser's native suggestion list filtered them by the text already in the field. It is replaced by a dropdown that always lists every model when opened (click in the field, the arrow, or the Down key), highlights the current one, filters only as you type, and works with the keyboard (arrows, Enter, Escape).

1.3.0
- New provider on Chrome and Edge: Chrome's built-in AI (Gemini Nano, on-device, no API key). Not available on Firefox, where this version behaves like 1.2.1: the provider is hidden and the usual API-key providers remain.
- Welcome page: the engine step groups the cloud providers under "For full power".
- Distraction-free reading tab shows the article title and byline; the model line under the panel title follows your settings live; suggested questions in the chat (experimental); new default summary prompt with key quotes.