The wake word runs on your device
You teach it your voice once, on your machine, and it stays there. No always-open line to a server, nothing to sign into before the agent will listen to you.
Skales Voice · Iris
Iris listens, she works, and she speaks. She lives in Skales, in her own place in the sidebar, and she is the agent itself - not a voice bolted onto a chat window. Ask out loud, and the thing that answers can actually go and do it.
Sidebar → Iris · Free for personal use · No account
You teach it your voice once, on your machine, and it stays there. No always-open line to a server, nothing to sign into before the agent will listen to you.
over 280 of them, not a curated shortlist: files, browser, calendar, mail, messengers, WordPress, code, Studio - plus the skills you switched on for the chat, which used to stop at the door of a spoken turn. A spoken turn and a typed turn reach the same agent, and a test in the codebase exists purely to keep it that way.
When she has something to show you, one window opens for it rather than five. The shape follows what you asked for: a short answer, a picture or a countdown comes as a circle with the text along the curve, a web page comes as the page. Escape closes it again and she is back to being a small thing at the edge of your screen.
What it feels like
The point of talking to an agent is not that talking is faster than typing. It is that you can hand over a whole task without sitting down first.
01 · You say it, from across the room
"Find the invoice from Kellner, rename it properly and put it in the tax folder."
02 · She takes it as work, not as a question
Searching, reading, renaming, moving - the same tools, the same approval gates as in chat.
03 · One window opens with the result
Not five notifications. One thing to look at, for as long as you want to look at it.
04 · Escape, and she is small again
Back to the edge of the screen, still listening if you want her to be.
Today
She already lives here. Install Skales, open the sidebar, and she is there - with the full tool set of the machine she is running on. Nothing to join, nothing to wait for.
Download Skales →Q4 2026
The standalone product, with its own site and its own timeline. Worth watching if you want the voice agent on its own terms rather than inside a desktop suite.
irisorbit.com ↗Voice questions
Iris is the voice surface inside Skales - her own place in the sidebar of the desktop app. You speak, she works, she answers out loud, and she can do everything the agent can do: files, browser, calendar, email, code, media. The skills you switched on for the chat reach her too, and a call is the same mind in the same conversation rather than a smaller one of its own. She is not a separate product you install.
Tool parity. Most voice assistants can reach a small, fixed list of actions - a timer, a song, the weather. An Iris turn sends the same tools a chat turn sends, in the same order, with the same approval gates. The Skales test suite asserts exactly that, because it is the whole point of the surface.
Yes, for the ordinary things: she says what she wants to do and waits, and yes or no in your language decides it. Say something else and she repeats the question rather than leaving you in silence, and after three tries the card comes back so there is always a way through. A permission nobody answers is refused, never granted. Anything that sends, deletes, runs or deploys deliberately stays out of that: that class always wants your eyes on the screen.
No. The wake word runs on your device. Skales is local-first everywhere else too: conversations, memory and files live on your machine, and with Skales Local even the model, the speech recognition and the speech output run on your own hardware.
Iris Orbit is being built as a product of its own, planned for Q4 2026. Iris in Skales is not a preview of it - she already lives here and works today, with this machine's full tool set behind her. If you want the voice agent now, you want Skales.
The Skales interface is fully translated into 12 languages, and voice in and out follow the model and speech engine you pick. With cloud speech engines that is most major languages; with the offline engines it is the voices you have installed locally.
Sidebar → Iris
Windows, macOS, Linux. Free for personal use, no account, and she works with whatever model you already brought.