Personal AI agent for macOS, Windows, Linux, Android & iOS. Set a goal, it works alone: coding (Skales Code), desktop + browser automation, autonomous scheduled Tasks. Teams of agents and humans across paired desktops. Agent2Agent, MCP, computer use, voice, local LLMs (Ollama), Agent Skills (SKILL.md). 26 AI providers, BYOK. No Docker, no terminal.
HTML agentic-aiai-agentai-assistantandroidautomationautonomous-agentdesktop-appelectronioslinuxllmlocal-ailocal-firstmacosmcpollamaprivacyreact-nativetypescriptwindows
12.9.31: the two fixes of the rebuilt binaries are in the changelog
Mario Simic84475fa53ef2
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 18def57..0f3c454 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -30,6 +30,10 @@ The other half of the release is the plugin floor: three plugins that do a job t### Fixed+- **Browsing asks the decision model the question it is actually asked.** With a decision model switched on, Skales was meant to let it take the small browsing steps, and in an ordinary conversation it almost never got that far: the model doing the talking kept reaching for the single steps it already knew, so the fast path sat there switched on and unused. Two things changed. A click described in words - "the link to the article about Vienna" - now goes to the decision model first, which picks the element on the page that the description means, and only if it is not sure does the click fall back to the older text match and the picture reader, exactly as before. And the multi-step run is now named as the way to work towards a goal on a page that is already open, so it is taken rather than passed over. Everything it does still stops and asks in the same places, and an installation with no decision model browses exactly the way it did.++- **The decision model list only offers names the service answers to.** One of the two entries in the model list was spelled a way the service refuses outright, so choosing it left every decision unanswered - and because an unanswered decision quietly falls back to the ordinary way, nothing looked broken, it was just switched on and doing nothing. The list now names only spellings that work, and a setting you saved earlier keeps working without you having to touch it.+- **The line under the name on the start screen is quieter now.** It was set almost as large as the app name and pressed up against it; it is smaller now and stands a little further below the name and the gecko, so the name reads first and the line reads as what it is, a caption.- **The menu bar icon speaks your language, and calls the Cockpit by its name.** Everything in the menu behind the Skales icon was written in English whatever language the rest of the app was set to - the status line, the provider, the count of calls this hour, the planner and goal lines, the waiting-approvals warning and every destination in the list. All of it is translated now, in all twelve languages, and the counted lines are written the way each language writes them rather than an English sentence with a number dropped into it. The menu is built fresh each time you open it, so changing the language in Settings shows up on the very next click without restarting. A machine with nothing configured yet still reads English. The entry that opens the autonomous surface is also named Cockpit now, the same name the page and the sidebar carry, instead of the older Autopilot - and so is the line that appears when the cost cap has paused it.