What is Canto

Canto is a cloud desktop platform built for computer-use agents. Every desktop is a real, persistent Linux machine that an agent (or a human) can wake in seconds, drive over exec/files/stream, and hibernate when it's done -- state preserved, compute stopped.

The hibernation model

A Canto desktop always exists once created -- its disk, its generation history, its identity -- but it isn't always running. It moves through a small set of lifecycle states (provisioning, waking, awake, hibernating, hibernated, destroyed; see Desktops and lifecycle), and compute is only billed while it's awake. Hibernated desktops cost nothing in hourly billing and keep their full disk state, ready to wake again on the next POST /v1/desktops/{id}/wake.

That's the whole pitch: always there, pay only when it thinks. An agent that works in bursts -- a few minutes of tool use, then long stretches of waiting on a human or another system -- doesn't pay cloud-VM prices for the idle time in between.

How it's built

• A small, versioned REST API (/v1/**) is the one source of truth: create/list/get desktops, wake/hibernate/destroy them, run commands, read/write files, get a stream ticket, and query usage.
• Every mutation (wake/hibernate/destroy) returns an operation handle you can poll or long-poll, rather than blocking your connection open.
• A TypeScript SDK wraps the API with a friendlier, camelCase, promise-based client and sensible retry/wait defaults.
• An MCP server exposes the same operations as tools, so Claude Code, Claude Desktop, Cursor, and any other MCP client can drive a Canto desktop directly.

Where to go next