Computer-use AI that works on your real desktop

Computer-use AI is a new kind of software: an AI system that can see a computer interface, understand what is on screen, and take actions like clicking, typing, navigating, and using apps.

Numa brings that idea to your desktop.

What computer-use AI does

A normal chatbot can answer a question. A computer-use agent can act on the answer.

For example:

A chatbot can tell you how to organize invoices.
A computer-use agent can open the folder, inspect the files, rename them, move them, and report back.

Numa is built for this second category. It is not only a place to ask questions. It is an agent that can operate your computer to complete work.

The basic loop

During an active task, Numa works in a loop:

1. Your computer captures the current screen state. 2. Numa sends the task context to its reasoning system. 3. The system decides the next action. 4. Your computer executes that action, such as click, type, open, read, or run code. 5. The loop repeats until the task is done or Numa needs your input.

This lets Numa work with tools that were designed for humans, not APIs.

Why this matters

Many real workflows are visual and contextual. The next step depends on what appears in the app, what file exists, what button is enabled, or what data changed since last time.

Computer-use AI is useful when:

There is no API.
A process uses multiple apps.
A task depends on what is visible on screen.
A human would normally teach the process by showing it.
The workflow changes slightly each time.

That is where Numa fits.

Numa and local desktop work

Some agents run in a cloud browser or virtual machine. That can be useful for web-only tasks. Numa focuses on your real macOS or Windows desktop.

That means it can work with:

Local files
Desktop apps
Browser sessions where you are already logged in
Spreadsheets and documents
Internal tools
Folders and project directories
Workflows you recorded yourself

Numa is not limited to a single browser tab.

Computer use with permission

Computer-use AI is powerful, so the permission model matters.

Numa needs user-granted permissions to do useful desktop work. On macOS, that includes access such as Accessibility and Screen Recording. On Windows, it needs equivalent input automation and screen capture permissions.

Numa is not always watching your screen. Screen capture is used during active tasks, recordings, or when you explicitly ask Numa to look. When idle, Numa is not recording your day.

What Numa can attempt

Numa can help with:

Clicking through desktop apps
Filling forms
Reading visible information
Moving and renaming files
Opening folders and documents
Updating spreadsheets
Using browser dashboards
Running scripts
Repeating recorded workflows

The best tasks have a clear goal and a clear done condition.

What computer-use AI should not be trusted with blindly

You should review sensitive work:

Payments
Purchases
Public posts
Emails to real people
File deletion
Account changes
Legal, financial, medical, or safety-critical decisions

Numa can assist, but you remain the final approval layer.

Benchmarks

Benchmarks are coming soon. This page does not claim benchmark superiority yet.

The short version

Numa is computer-use AI for your desktop. It can see your screen during active tasks, decide the next step, and use your computer to complete work across apps, files, and workflows.