SOLARIA full desktop to drive.
A real GUI machine your agent controls with mouse and keyboard, for the desktop apps built for people, not APIs. Pause it mid-task and resume it, state intact, in 0.78 s.
Drive it exactly like a human would.
A screen you can only watch is useless to an agent. Solari gives full interactive control. Every click and keystroke works, over a secure connection.
const d = await pt.desktops.create({ template: "office", record: true });
await d.connect();
await d.health(); // wait for ready
// real input: mouse, keyboard, hotkeys
await d.mouse.doubleClick(100, 100);
await d.keyboard.type("hello world");
await d.open("libreoffice", ["--writer"]);
// watch it live, or grab a frame
console.log("stream:", d.streamUrl);
const png = await d.screenshot();d = await pt.desktops.create(template="office", record=True)
await d.connect()
await d.health() # wait for ready
# real input: mouse, keyboard, hotkeys
await d.mouse.double_click(100, 100)
await d.keyboard.type("hello world")
await d.open("libreoffice", ["--writer"])
# watch it live, or grab a frame
print("stream:", d.stream_url)
png = await d.screenshot()- →Real display. A genuine Linux desktop, not a simulated one.
- →Real input. Mouse, keyboard, hotkeys, drag & scroll, with optional human-like motion.
- →Open any app. LibreOffice, GIMP, Firefox, or your own image.
- →Session replay. Record everything and play it back later.
Pause it mid-task. Resume it an hour later.
Enterprise workflows run for minutes or hours, with long idle gaps. You shouldn't pay for a machine that's just sitting there while the model thinks.
When a session goes idle, it pauses instead of dying, and stops billing. An open connection keeps it awake.
Pick up exactly where you left off: files, processes, and the size you chose, all intact, in under a second.
Sessions survive hardware failure. A paused machine can come back on a different host. Nothing is lost.
Capture any machine and relaunch it, or fork it into many. Set up an environment once and clone it cheaply for every run.
What teams point it at.
Desktop automation
Point an agent at legacy software built for humans: an ERP, an EHR, an office suite.
Long-horizon agents
Runs that pause on idle and resume hours later on any host, with no compute burned while thinking.
End-to-end flows
Drive multi-app workflows across a real desktop, with a recording of every session.
Your toolchain
Bring your own desktop image and get it as a fast, warm-pooled machine.
Give your agent a computer to drive.
One API key. A full desktop with real input, resuming in 0.78 s.