Now in public beta

Cloud desktops for
AI agents

Spin up a virtual machine in milliseconds. Connect any AI model. Let it see, click, type, and execute. Production-grade infrastructure for computer-use agents.

quickstart.py
import corix

# Boot a desktop in <500ms
computer = corix.create(
  os="linux",
  cpu=2,
  ram="8gb"
)

# Connect Claude to drive it
computer.prompt(
  model="claude-sonnet-4-6",
  task="Open Firefox, go to HN, find the top post"
)

# Or control it directly
computer.screenshot()
computer.click(500, 300)
computer.type("Hello from Corix")
computer.bash("ls -la /home")

Everything agents need to operate

Full desktop control through a simple HTTP API. Screenshot, click, type, execute.

Sub-500ms boot

Machines initialize in under half a second. No cold starts. No waiting. Your agent fleet scales instantly.

Full desktop control

Mouse clicks, keyboard input, screenshots, bash commands, Python execution. Everything a human can do, your agent can do.

Any model

Claude, GPT-4, Gemini, open-source. OpenAI-compatible endpoint included. Bring whatever model fits your task.

Browser-native desktop

Not just VNC. A real desktop experience rendered in the browser. File manager, terminal, editor. Watch your agent work live.

Persistent or ephemeral

Spin up throwaway sandboxes for one-off tasks, or keep persistent machines running with saved state across sessions.

SDK + raw API

Python and TypeScript SDKs for quick integration. Or use the plain HTTP API from any language with an HTTP client.

The agent loop

Four steps. Infinite capability. Your agent sees the screen, decides what to do, acts, and repeats.

1

See

Capture a screenshot of the desktop

2

Decide

AI model analyzes and plans next action

3

Act

Click, type, scroll, or run a command

4

Repeat

Loop until the task is complete

Give your agents a body

Start building in minutes. Free tier included.