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

# The DevOps AI Agent

> An AI system administrator that sets up and changes an application on your own VPS over SSH, while you chat with it

## What It Is

The DevOps AI Agent is an AI system administrator for your own server. It logs in to your Ubuntu VPS over SSH as root, installs an application from its GitHub repository, configures it, checks that it responds, and stays in a chat that you can come back to later. You do not need a terminal: the agent is instructed to do the work itself and never to end a message with commands or configuration steps for you to do.

> **Sets Up the App**
>
> Installs the application from its repository on a fresh Ubuntu server, sets up a web server with HTTPS on the server's domain, and reports the address

> **Stays in the Chat**
>
> Return to the conversation from the server's manage page to change settings, add features, or find out why something does not work

> **Your Own Server**
>
> It works on your dedicated VPS, with full root access, and its AI usage is not billed on top of the server's hourly price

> [!NOTE]
>
> The agent is an AI and it can be wrong. Its last message names the address of your application: open it, and check the application works the way you expect. [Safety and Control](https://repocloud.io/docs/ai-agents/safety-and-control.md) explains how to protect your work.

## Where It Appears

A DevOps AI Agent chat is created together with its server, in one of three ways:

- **[AI Deploy](https://repocloud.io/docs/deployment/ai-deploy.md).** Enter a public GitHub repository, choose a tier, and optionally add [custom instructions](https://repocloud.io/docs/ai-agents/custom-instructions.md).
- **The Deploy as VPS switch.** On the deploy form of a Container App that has a GitHub repository, the **Deploy as VPS with AI system admin agent** switch hands the same project to AI Deploy on a VPS. [Deploy as VPS](https://repocloud.io/docs/deployment/deploy-as-vps.md) lists the conditions.
- **Deploy Now on some catalog apps.** A few apps open AI Deploy directly when you press **Deploy Now**.

Afterwards the server's manage page shows a **DevOps AI Agent** card with a **Resume Chat** button (see [Resume the Chat](https://repocloud.io/docs/ai-agents/devops-ai-agent.md#resume-the-chat)). Only servers created this way have a chat. A server deployed from the Ubuntu VPS marketplace listing and a server created through the API have no chat and no card. You cannot start a chat for a server you already have.

## From Click to Chat

**1. The Server Is Created**

The status page shows three steps: **Creating Environment** (provisioning a fresh Ubuntu VPS of the tier you chose), **Researching Project** (crawling the project's documentation and creating a deployment plan), and **Ready to Deploy** (connecting to the AI deployment agent).

**2. The Chat Opens**

When the steps finish, the page counts down ("Redirecting in 3 seconds...") and offers **Continue to Deployment**. The chat starts itself: a first message that names the repository and your server's tier is sent for you, so there is nothing to type.

**3. You Watch It Work**

The agent writes what it is doing and runs its commands. The form says that an AI-assisted deployment typically takes 5–15 minutes.

**4. It Reports**

The agent is instructed to finish by reporting the HTTPS address of the application, any credentials it created, and whether the application works. From then on the chat is open for your next request.

## What You See in the Chat

- **Messages.** Your messages and the agent's replies, as in any chat. The chat takes text only: it has no attachments.
- **Tool rows.** Each action the agent takes on your server appears as a row named after the tool, such as `ssh_execute`, with a status icon. Click a row to see its arguments (the command or the file) and its result. This is how you check exactly what the agent did.
- **The message box.** It reads "Write your message..." and **Send** sends it (or press Enter; Shift+Enter starts a new line). While the agent is working the box reads "Running..." and the button becomes **Stop**.

## What It Does on Your Server

The agent starts from a fresh Ubuntu 24.04 LTS server and has full root SSH access. Its instructions, in the order it works through them:

| Stage | What the agent is instructed to do |
|---|---|
| Prepare | Update the server's packages |
| Fetch | Clone the repository, then read its README, Dockerfile, and compose files |
| Install | Install the dependencies the project needs; for a Docker-based project it prefers Docker Compose over a manual installation |
| Configure | Configure the application as its documentation says; a web application is made reachable on port 80 or 443, with a reverse proxy (nginx or Caddy) answering for the server's domain when the application runs on another port |
| Secure | For a web application, set up HTTPS for the server's domain with Certbot |
| Start and verify | Start the application, then check the logs and the HTTP responses |
| Finish | Complete the post-deployment configuration, such as the firewall, cron jobs, log rotation, and environment variables, then report the address and any credentials it created |

The agent uses the server's domain name, not its IP address, for HTTPS and for every address it gives you. That name is the default domain RepoCloud gives the server: a free name that points at the server's IPv4 address and has HTTPS only if it is set up, which is one of the agent's jobs ([Ubuntu VPS](https://repocloud.io/docs/deployment/ubuntu.md) explains the name).

## How It Works With You

- **It asks for what only you can give.** API keys, OAuth credentials, SMTP settings, and license keys are not things it can invent. It is instructed to ask for them and wait for your answer.
- **It keeps going when a step fails.** It is instructed to work out the cause and fix it, and to try again while it is making progress. When it is truly stuck, it tells you what it tried, shows the key error output, and asks for what it needs.
- **It sticks to the documentation.** It is instructed not to install or run what it does not understand, and to follow what the project's documentation says.
- **It reports briefly.** It reports progress and not every command; the tool rows show the commands. It is instructed never to end a message with commands for you to run: if more work remains, it does the work or asks whether you want it done.
- **It suggests extras.** After the deployment it may offer SSL, backups, monitoring, security hardening, or optional features of the application. Accept the ones you want and ignore the rest.
- **You can steer it at any time.** Write what you want in plain language: change a setting, add a plugin, explain what went wrong. [Customizing Applications](https://repocloud.io/docs/ai-agents/customizing-applications.md) has recipes.

### Stopping the Agent

**Stop** ends the run and closes its connection to your server. A command that had already started may or may not have finished, so ask the agent to check the server's state before it carries on.

## Resume the Chat

The conversation belongs to the server. On the server's manage page, the **DevOps AI Agent** card ("Resume your deployment chat session") shows the repository and the chat's status. **Resume Chat** continues the conversation with its history, and any custom instructions you gave at deployment stay in force. The card appears while the chat's status is ready, active, or completed.

While the server is paused for a low balance, the button is disabled with the hint "Add funds to resume". Add funds, and the button works again once the server is running. See [Managing VPS Apps](https://repocloud.io/docs/manage/vps-apps.md) for the rest of the manage page.

## What It Does Not Do

- **It works through SSH on your server.** Its tools run commands, and read and write files, on the server. It has no tool for your RepoCloud account, your balance, your DNS provider, or your other servers.
- **It does not take snapshots, resize the server, or change tiers.** Those are buttons on the manage page, and they are yours to press.
- **It is not an expert on your project.** It works from what it reads in the repository's documentation and on your server, and from its general training, and it can misjudge a step.

## Frequently Asked Questions

> **Does the agent cost extra?**
>
> No. The server is billed by the hour at the price of its tier, like any other VPS App, and the agent's AI usage is not billed separately. See [VPS Apps](https://repocloud.io/docs/deployment/vps-apps.md) for the tiers.

> **Can I choose the AI model?**
>
> No. RepoCloud chooses the model the agent uses.

> **Can I send the agent a file or a screenshot?**
>
> No, the chat takes text only. Give the agent a link that the server can download the file from, or copy the file to the server yourself with `scp`, using the server's IPv4 address from its manage page and its root password (the **Email Me** button on the Root Password row sends it to you again).

> **Can I use the agent on a server I already have?**
>
> No. The chat is created together with the server, through AI Deploy, the Deploy as VPS switch, or an app that opens AI Deploy. For a server you already have, you can log in and set it up yourself, or let [your own AI agent](https://repocloud.io/docs/ai-agents/your-own-agent.md) do the work.

> **Can the agent deploy a private repository?**
>
> No. AI Deploy takes public GitHub repositories. To deploy a private one, use your own AI agent with the AI Agent MCP: see [Your Own Agent](https://repocloud.io/docs/ai-agents/your-own-agent.md).

> **What if it does something I did not want?**
>
> Tell it to undo the change, or restore a snapshot if you took one. The snapshot is the dependable way back, because it returns the whole disk. See [Safety and Control](https://repocloud.io/docs/ai-agents/safety-and-control.md).

> **The chat is not working**
>
> See [Troubleshooting the Chat](https://repocloud.io/docs/ai-agents/troubleshooting.md).

## Next Steps

> **[Custom Instructions](https://repocloud.io/docs/ai-agents/custom-instructions.md)**
>
> Tell the agent how to set up your deployment

> **[Customizing Applications](https://repocloud.io/docs/ai-agents/customizing-applications.md)**
>
> Recipes for changing an application with the agent

> **[Safety and Control](https://repocloud.io/docs/ai-agents/safety-and-control.md)**
>
> Snapshots, Stop, what the agent can reach, and who sees what

> **[Troubleshooting the Chat](https://repocloud.io/docs/ai-agents/troubleshooting.md)**
>
> What to do when the chat or the agent does not behave