Sandboxes for agents.

Your agent needs a computer, not a chat window. Give it one per task, and let it stop when the work is done.

What you get

Every machine runs Mecha, an assistant that writes and runs code on the machine itself. Deployport provides the model, so there is no API key to set up.
Your own agents run on it too, on your own subscriptions, Claude and Codex included.
A machine does not stop while Mecha is working on a task, and the idle rules resume when the task ends.
The disk is fully persistent, and you have snapshots and restore.
It is a real KVM-isolated virtual machine, not a shared-kernel container, so a package it installs stays installed.
It answers on a name of its own, so what it builds is reachable when the task ends.

One command to start it, one to end it

$ deployport machinery create --prompt "add a search endpoint, with tests"24s

machine search-endpoint (mach_kxpkzlq4qf) requested
Launching. The cloud is creating the instance
Booting
Connecting. Waiting for initial connection
Online

SSH:   ssh search-endpoint.deployport.sh
HTTPS: https://search-endpoint.deployport.app/

$ deployport machinery commands run mach_kxpkzlq4qf -- npm testexit 0

Test Suites: 4 passed, 4 total
Tests:       31 passed, 31 total

The assistant picks the prompt up as soon as the machine is ready. When the task is done, deployport machinery stop mach_kxpkzlq4qf keeps the disk and stops the meter.

The cost

A machine is billed by the minute while it runs, and its storage while it is stopped. Sizes start at $0.05 a machine-hour. See Pricing.

Start free trial · Talk to a founder