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

# Run your first task

> Pick a workflow, supply parameters, watch it execute.

This page walks you from a fresh team to a completed task.

## Do you have a GPU machine?

Cyberun runs tasks on **agents** — processes on GPU machines your
team controls. Your path depends on whether you bring the GPU
yourself:

* **Yes, I have a GPU box.** Install the agent on it (10 minutes),
  then submit a task from the dashboard. Steps 1–3 below.
* **No, I just want to try it.** Run a workflow against shared
  team agents (if someone on your team already connected one) or
  use [Comfy Cloud fallback](/cloud/guides/connect-comfy-cloud)
  (when enabled on your deployment). Skip step 1 and start at
  step 2.

## 1. Connect an agent

Your team needs at least one agent online to run tasks. If
someone else on your team already connected one, skip ahead to
step 2.

1. In Cloud, open **Access** in the sidebar. Switch to the
   **Agent** tab.
2. Click **Create agent key**. Name it after the machine —
   `home-rtx-4090`, `studio-mac-mini`. Copy the `ak-...` value
   shown; **it's displayed once.**
3. On the GPU machine, run the installer one-liner for the
   runtime you want:

   ```sh theme={null}
   curl -sL https://releases.cyberun.cloud/agent/install.sh | bash -s -- \
       --api-key ak-...
   ```

   This rootless one-liner works on Linux and macOS — no `sudo`, with
   comfyui as the default tool. For Windows and the full flag list
   (Docker mode, Nerfstudio, GPU index pinning), see
   [Connect an agent](/cloud/guides/connect-agent).
4. The installer sets up the runtime, registers a per-user service,
   and starts the agent. In Cloud, open **Agents** in the sidebar
   — the new agent appears as **idle** within a few seconds.

Labels (`gpu`, `comfyui`, `nerfstudio`, …) describe what the
agent can run. The installer derives them from `--tool` and the
detected hardware; you don't set them by hand for the common
cases. Workflows match against an agent's label set when
dispatching.

## 2. Add a workflow

A **workflow** is a reusable template. A brand-new team starts
with an **empty Workflows list** — add the first one before
there's anything to run.

1. Open **Workflows** in the sidebar.
2. If the list is empty, create or import a workflow so it
   appears here. A workflow declares its parameters and the agent
   labels it needs.
3. Open a workflow whose required labels match an agent on your
   team (the card shows the labels it needs).

## 3. Submit a task

1. On the workflow detail page, click **Run**.
2. Fill in the parameters the workflow asks for (prompt, seed,
   image, …).
3. Click **Submit**.

You land on the task detail page. The status moves through
**pending → waiting → provisioning → queued → running →
completed** (or **failed**). Live progress streams in as the agent runs. When
the task finishes, the **Outputs** panel shows the files the
workflow produced — click to preview, download icon to save.

## Next

<CardGroup cols={2}>
  <Card title="Connect an AI client via MCP" icon="plug" href="/cloud/guides/connect-mcp">
    Drive future tasks from Claude Code, Cursor, or any MCP client.
  </Card>

  <Card title="Generate an API key" icon="key" href="/cloud/guides/api-key">
    Submit tasks from your own scripts and CI.
  </Card>

  <Card title="Webhooks" icon="webhook" href="/cloud/guides/webhooks">
    Get notified when long tasks finish instead of polling.
  </Card>

  <Card title="Find your output" icon="download" href="/cloud/concepts/tasks">
    How to download results, re-run from the workflow, and
    cancel.
  </Card>
</CardGroup>