> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voyager.so/llms.txt
> Use this file to discover all available pages before exploring further.

# Cloud machines

> Lease a Voyager cloud GPU by the minute, run commands on it, move files in and out, and release it when you're done, from the voyager-cloud CLI, the REST API or MCP.

A **cloud machine** is a Linux computer with a GPU in Voyager's cloud that you lease by the minute. You or your coding agent start one, run commands on it, copy files in and out, and release it. It's built for GPU work your own computer can't do quickly: renders with [Vork](https://docs.voyager.so/api/machines/vork.md), simulations, CUDA jobs and big exports.

You can drive cloud machines three ways. They all do the same things on the same account:

| Way | Best for | Start here |
| - | - | - |
| **`voyager-cloud` CLI** | You, in a terminal, or a coding agent like Claude Code or Codex | [Quickstart](https://docs.voyager.so/api/machines/quickstart.md) |
| **REST API** at `https://api.voyager.so/v1/machines` | Your own code and services | [Reference](https://docs.voyager.so/api/machines/reference.md) |
| **MCP server** at `https://mcp.voyager.so/mcp/machines` | Agents that work through MCP tools, including Claude.ai | [MCP setup](https://docs.voyager.so/api/machines/mcp.md) |

> [!NOTE]
>
> **Who can use it:** a paid Voyager workspace, Starter or above, with a completed payment. On a free workspace, requests are refused with an error that carries the billing link (`details.upgrade_url`). Machines bill your workspace's Voyager credits per second while they run. See [voyager.so/pricing](https://voyager.so/pricing) for plans.

## Presets

A **preset** picks the GPU and the software image. Pick one when you lease a machine; `render` is the default.

| Preset | GPU | VRAM | Ray tracing | Runs Vork | Use it for |
| - | - | - | - | - | - |
| `render` (default) | NVIDIA RTX PRO 6000 | 96 GB | Yes | Yes | Final renders and heavy GPU work |
| `render-lite` | NVIDIA L4 | 24 GB | Yes | Yes | Drafts, previews and lighter jobs, for less |
| `compute` | NVIDIA H100 | 80 GB | No | No | CUDA compute, training and simulation that don't need ray tracing |

All three start from the same image, with CUDA, Python, FFmpeg, Blender tools and Vork already installed. List the presets, GPUs and images with their live prices:

```sh theme={"system"}
voyager-cloud presets
voyager-cloud gpus
voyager-cloud images
```

Each preset and GPU shows `credits_per_minute` and `usd_per_minute`. Prices change; read them from the API rather than hard-coding them.

## Billing and auto-sleep

* A machine bills **per second from when it's ready until you pause or release it**. Start-up time isn't charged. A paused machine keeps its disk and doesn't bill for the GPU.
* **Idle machines pause by themselves** after 15 minutes without a command or file transfer (`idle_timeout`, 900 seconds by default; set it from 30 seconds to 7 days when you lease).
* Running a command or moving files on a paused machine **resumes it** first.
* Set a `budget_usd` when you lease and the machine stops once it has spent that much.
* **Release** a machine when you're done. Release returns the final machine with its total `cost` (`credits` and `usd`).

> [!WARNING]
>
> A machine you forget keeps billing until it pauses itself. Release it at the end of every session: `voyager-cloud release <machine>`.

## How a session goes

1. **Lease** a machine and wait until its `status` is `ready`.
2. **Push** your files to it.
3. **Run** commands. A command waits and prints its output by default; long ones run as a **job** you can follow.
4. **Pull** the results back.
5. **Release** the machine, or pause it to come back later. Save a **snapshot** first if you want to start the next machine from this disk.

Machine `status` is one of `starting`, `ready`, `paused`, `released` or `failed`. Every machine and job response has a `next` field with a plain-English hint for what to do next, for example `Machine ready. Run a command: POST /v1/machines/{id}/run`.

## Next

> [card] [Quickstart](https://docs.voyager.so/api/machines/quickstart.md)
>
> Install `voyager-cloud`, sign in and run your first command on a GPU.

> [card] [Render with Vork](https://docs.voyager.so/api/machines/vork.md)
>
> The render loop for Vork scenes on Voyager cloud GPUs, with a prompt for your agent.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.