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

# Reuse a setup with Cloud agents

A Cloud agent is a saved setup for new workspaces: which repository to clone, what the agent should know, which
model to use, how big the machine is, and which tools it has. Start as many workspaces from it as you like. Each one
gets its own machine, files and chats.

Saving a Cloud agent starts nothing and costs nothing. Only the workspaces you start from it do.

## Create one

1. Open **Cloud agents** in the sidebar and choose **Create agent**.
2. Give it an **Agent name** and, if you like, a **Description**.
3. Open each section and fill in what you need. Every section is optional:

| Section | What you set |
| --- | --- |
| **Behaviour** | **Instructions** the agent gets at the start of every chat, the **Default model** and the **Reasoning effort** |
| **Repository and SSH** | The **Git repository** to clone, a **Branch or tag**, and the **SSH known hosts** for your Git host |
| **Machine** | The **Compute** size, **Disk (GB)** and the **Idle timeout (minutes)** before it sleeps |
| **Environment and setup** | **Environment variables** for secrets and settings, and a **Startup command** that runs on every boot |
| **MCP, agents, skills and permissions** | MCP servers, custom agents, skills and permission rules, as JSON |

4. Choose **Create agent**.

### Private repositories

After you save, the agent shows an **SSH public key**. Add it to your repository as a deploy key once. Every
workspace you start from this agent can then clone it.

## Start a workspace from it

- On **Home**, pick the agent under **Start from**, describe the task and press Enter.
- Or, in **Cloud agents**, choose **Start workspace** on the agent.
- To change a setting for this one workspace before it boots, choose **Customize and start** instead.

The workspace copies the agent's settings when it starts. Later changes in its **Workspace settings** never touch the
agent.

## Edit, copy or delete an agent

Open the agent in **Cloud agents** to edit it. Edits apply to workspaces you start **afterwards**. Workspaces that
already exist keep the settings they started with.

**Duplicate**, in the agent's menu, makes a copy to change. Environment variables are not copied: enter them again
in the copy.

Deleting an agent keeps every workspace started from it, with their files and chats.

## Turn a workspace into an agent

Set up a workspace by hand until it works the way you want. Then open **Workspace settings** and, under **Reuse**,
choose **Save as Cloud agent**. Cloud generates a new deploy key for the agent. Choose **Also copy encrypted
environment variables** to bring your secrets along.

## Tips

- **Keep the startup command safe to run twice.** It runs on every boot, not only the first. Check whether a
  dependency is installed before installing it.
- **Put secrets in environment variables,** not in instructions or MCP settings. In MCP settings, refer to a
  secret as `${VARIABLE}`. Saved values are encrypted and never shown again.
- **Install large tools with the startup command.** The JSON configuration is limited to 64 KiB.

## From the terminal

```sh
haarness cloud agent list
haarness cloud agent show AGENT_ID
haarness cloud workspace create --agent AGENT_ID --title "Nightly review"
```