Skales - AI Agent & Tasksv2.9.46
Released October 1, 2026Major
iOS Android
What changed
- UXAdded global voice setting to prevent microphone from opening while the agent is speaking.
- UXImproved UI for 'Thinking Process' by folding it away from the main reply area.
- PerformanceEnsured useful answers are retained in chat even if the agent corrects a tool name.
- New featureStudio visuals can now use specific, selectable providers and models with backup/restore support.
- LocalizationStudio settings and interface now support twelve languages.
- UXAdded three new easing transitions and stillness controls for motion elements.
- New featureIntroduced a free code lane for timelines powered by an internal animation library.
- PerformanceEnhanced reports to include specific provider and model metadata for better tracking.
- UXImproved backup error reporting to specify if failures are due to archives, secrets, or providers.
- PlatformOptimized file and music tool integration for better folder paging and state retention.
Original release notes
Skales Mend repairs the loose ends of the last release. Voice stops answering its own echo, a finished answer stays where you can read it, and Studio brings its own model and its craft. VOICE A recording that only repeats the end of what Skales has just said is dropped instead of being sent as your turn. A short answer to a spoken question - a yes, a no, a name - is always kept, and a stop, a permission answer or a longer question keeps its own path whatever else is playing. The switch that holds the microphone closed while Skales speaks now sits in the general voice settings and covers every voice surface, so Voice and Iris open on the same setting you saved. ANSWERS THAT STAY An answer that begins with a "Thinking Process" now folds that part away where the rest of the thinking lives, instead of standing at the top of the reply. A quoted example, or an ordinary sentence about thinking, still reads as text. And a useful answer written just before Skales corrected the name of a tool it was about to use stays in the chat rather than disappearing with the correction - kept alongside the final answer, and counted once, not twice. STUDIO A visual can now run on a provider and model of its own, chosen in the same picker as everywhere else and remembered through a fallback and through a backup restore. When nothing is chosen, a visual brings the full design craft with it - layout, print and motion - instead of none, the card names what it used, and a retry starts from the same choices. The setting speaks all twelve languages, and a job that cannot be stored says so before it starts rather than after. MOTION Scenes can now let an element leave on purpose, with three new eases to leave on, and a scene may hold an element still where the beat asks for stillness rather than making everything enter. A cut hands one scene to the next without an overlap, and stepping forward, back or over the same frame twice lands on exactly the same picture. Alongside the standard engine there is now a free code lane for timelines, drawn by an animation library that ships inside the app. REPORTS AND BACKUPS A reported answer now carries the provider and the model that wrote it, so a report says where the content came from; where an older conversation never recorded it, it stays unknown rather than guessed. A backup import that stops now says what stopped it - the archive, a saved …