AI Agents

An OS built for AI Agents.

Keep control of your computer, deploy up to 10 agents, and let AI jump in when you didn’t know you needed help.

Three reasons

Why did we redesign an OS for AI Agents?

We work differently today than we did five years ago. Why? AI. Getting its full value into your day meant rebuilding the computer itself.

AI on Novastart gives you:

Ten windows on one canvas, each with its own coloured cursor labelled AI, motion trails behind every cursor as they work.

Up to 10x faster output

Ten agents, ten cursors, one canvas. Deploy them across your job, or let one lead split it across up to ten sub-agents and report back all at once. You never lose control of your mouse.

A spreadsheet being filled in by hand while an agent's suggestion card offers to finish the remaining rows, a play button beside it.

More of your day, handled

With Observe enabled, AI tells you when it can take over an activity for you. One click hands it off, and you get your time back.

An agent's green cursor clicks inside a ringed window marked AI, with Pause, Take back and Steer controls and a Checked stamp beside it.

Safety

We need to see agents to trust them. On Novastart they work in front of you, every click visible. Pause, steer or take the window back at any time.

How it works

How does Novastart’s AI work?

Four key functionalities: some you will know, and some are unique capabilities of our OS.

An agent's chat panel holding two finished asks, a cabin search and its booking, with the task box waiting below.

Chat

A running conversation with every agent. It keeps your context and your history, so you pick up where you left off.

A restaurant booking site where an agent's cyan cursor clicks Reserve on the one open table, under the ask for a table for four.

Agent

One-off support. Ask for one thing done, from the grocery order to finding a reservation, and the agent pulls up browsers and does it all in front of you, one click at a time.

A still from the AI Agents film: the agent’s three-step plan beside its fanned-out helper windows, the dock centred below.

Plan

Long-horizon work. The lead agent drafts steps you can edit, then runs the project with its sub-agents, each step ticking off live.

A filled-in expense sheet beside an agent's suggestion card offering to fill the remaining rows, a round play button beside it.

Observe

With your permission, it watches the window you are working in and suggests the next chunk of work. One click hands it to a cursor of its own.

Watch some examples for yourself.

Build trust with your agents by watching them work, and gain leverage from a team that runs while you do.

A user asks one agent to plan a ski trip. The agent drafts a plan with one task per helper, opens a window for each part of the trip in a tidy grid beside the user's own notes, and names and colours a helper for each part, one per window. The user types a correction mid-run and the agents adjust without restarting. Pause freezes all of them. The helpers report back to the lead, the lead adds it up in a sheet, and a separate check looks at a fresh screenshot before the job is marked done.

Before you try

The things people ask before they trust it.

Handing your work to an agent is a big deal. Here are the questions we get all the time.

Do I have to babysit it?

No. Every agent works in the open, so you can watch when you like and step away when you like. It keeps working while you do something else, and it can ask you a question when it needs one.

Can it start work on its own?

Only if you let it. Observe is off until you turn it on; then the agent watches the window you are working in and offers to take a chunk over. Nothing moves until you hand it off, with one click.

What if it clicks the wrong thing?

Mostly it catches itself: the agent sees the result of every click, notices when one went wrong, and gets back on track. And if you are not happy, pause it and give feedback by typing. It adjusts without restarting.

Does it work in my apps?

Yes, through the real UI, with the same clicks and keys you would use. Gmail, Excel, WhatsApp, Notion and thirty more open as real windows, and agents can also arrange windows, launch apps and open URLs.

    How many agents can I run at once?

    Up to 16 on one canvas, and each can recruit up to ten sub-agents of its own. The OS gives a window to one agent at a time and lands every click whole, so they never collide, and their windows are kept in a tidy grid beside yours.

    What happens when I restart my computer?

    Nothing is lost. Agents keep your preferences, and they and their conversations are restored after a reboot, so you pick a job back up where you left it.

    Questions

    Ask us anything.

    Anything about Novastart, the app or the install. We reply by email.