Launching today
Skales - AI Desktop Agent for Win & Mac
Local AI that acts. Not chats. Installs in 30 seconds. 🦎
28 followers
Local AI that acts. Not chats. Installs in 30 seconds. 🦎
28 followers
The Desktop Buddy sits persistently in the corner of your screen as a small animated gecko. Click it once, describe what you need, and it handles it - without you switching windows, opening new tabs, or navigating menus. Skales is a local first AI desktop agent for Windows and macOS. It manages emails, calendar, browser tasks and research autonomously. Asks before anything destructive. Works with any LLM: OpenRouter, Anthropic, Gemini or fully offline via Ollama. No Docker - No terminal.
Hey Product Hunt. Mario from Vienna here.
I'm a web developer and marketing guy. I built Skales because I spent two hours wrestling with Docker just to get a CLI-based AI agent running, it ate all my RAM, and I thought: my wife could never set this up. My 60+ year old mom could never set this up. There had to be a better way.
So I built a desktop AI that installs like a normal app. Double-click, done. No terminal. No Docker. No YAML files.
What it actually does (not just chats):
It manages your inbox and calendar, controls your browser via Playwright, runs multi-step research and reports back, and stays on your screen as a small desktop buddy so you never have to switch windows or open a new tab.
Every action that matters requires your approval before it executes. You stay in control.
A few things I'm proud of:
My 6-year-old used the built-in code builder to make a Snake game in 12 minutes. He named the app. It works with any LLM: OpenRouter, Claude, Gemini, Groq, or fully local via Ollama (meaning zero data ever leaves your machine). Keyboard accessible and screen reader compatible from day one.
Built solo from Vienna. Architecture feedback genuinely welcome.
What would you want a local AI to handle for you?
Fun idea. A lot of “AI agents” still mostly chat, so something that actually takes actions on the computer is a step closer to what people expect. Running locally and letting users plug in any model is also a sensible choice, especially for privacy.
How reliable the automation is in practice? Desktop agents often sound great in demos but can get messy once real workflows start piling up. Still impressive that this was built solo.
@steffen_rehmann Thanks Steffen! For single tasks - calendar events, emails, file management, web research.. it's solid and reliable. The agent knows its own capabilities and picks the right tools automatically, or fix himself or develops custom skills.
For multi-step workflows, that's where the approval system comes in. Skales shows you the full plan, you approve each step, it executes (as long as you activate safe mode instead of unrestricted). You stay in control without micromanaging. The stronger the model you connect (Claude, Gemini, GPT-4), the better it handles complex chains.
What makes it different: like mentioned before Skales can even build its own custom skills. If it recognizes something it can't do yet, it can generate a new skill for it. So it gets more capable the more you use it.
Being solo-built means I ship fast based on real feedback. The approval system itself came from Reddit users asking for more safety - shipped it within a week.
Soofte
Looks interesting Mario. Do you have linux support in your roadmap?
@mateusz_jacniacki
Hey Mateusz, appreciate it! 🙏
Short answer: Yes, absolutely!
Long answer: Since Skales is built with Electron, bringing it to Linux is definitely doable without reinventing the wheel. Right now, as a solo dev, I'm focusing all my energy on ironing out the initial launch bugs and smoothing out the Windows and macOS versions.
But I know how huge the Linux community is in the local LLM space, so it's very high up on the roadmap!
Out of curiosity: What format would you personally prefer to see first? AppImage, .deb, or something else?
Cheers,
Mario 🦎
btw.. you can take a closer look at the Roadmap here: https://doc.skales.app/roadmap.html
Soofte
@yaboymare thanks for the response! Completely understand your focus, you should protect it. For me personally Appimage would be enough.
@mateusz_jacniacki AppImage it is, noted! Honestly that's the most practical format for this use case anyway (no root, no system dependencies, just runs). I'll make sure that's the first Linux build when we get there. Thanks for the input, genuinely helpful to know what people actually want before building it 🦎
Quick update:
Already got a YouTube comment from someone downloading Skales and asking about LM Studio support - custom OpenAI-compatible endpoints are confirmed for the next update.
Keep the feedback coming! 🦎