# Remote Agents (https://depot.dev/docs/agents/overview)

Depot's remote agent sandboxes provide a secure, isolated environment for running AI coding agents like Claude Code in the cloud. Allowing you to move your agent coding sessions off of your local machine and into fast remote environments where you can easily launch, resume, and share sessions.

Current agent sandboxes support Claude Code, with more agents coming soon. By default, running `depot claude` will start a new session in a remote sandbox.

<DocsCTASecondary buttonId="agents-docs-cta-to-quickstart">
  <a href="/docs/agents/claude-code/quickstart" className="font-semibold text-radix-mauve12">
    To dive into using remote agents on Depot, check out our Claude Code quickstart guide →
  </a>
</DocsCTASecondary>

## Key features

### Isolated environments

Each agent session runs in its own isolated container, providing a clean and secure environment for your development work. Sessions are completely isolated from each other, ensuring your work remains private and secure.

### Persistent file system

Agent sandboxes work directly with your git repositories and persists files automatically across agents sessions. Allowing you to resume your work exactly where you left off, whether you're picking up a session from last week, sharing with a teammate, or starting a new session from an existing sandbox.

### Pre-configured development tools

Agent sandboxes come pre-installed with popular programming languages, package managers, and development tools.

### Session management

Every agent sandbox not only persists your filesystem, but also the entire context and conversation you have built up with your coding agent in the remote sandbox.

### Git integration

Work directly with Git repositories in your sandbox. Clone public or private repositories (using secrets for authentication), make changes, and push updates - all within the isolated environment.

### High performance

Agent sandboxes run on Depot's optimized infrastructure with plans to provide automatic integrations with our existing Depot services like accelerated container builds, Depot Cache, and more. Every sandbox launches with **2 vCPUs and 4 GB RAM** by default, providing ample resources for most development tasks.

### Web UI for sessions

Manage your agent sessions through the Depot dashboard:

* **Session overview**: View all your Claude Code sessions with their status, last updated time, and whether they used a remote agent
* **Resume sessions**: Pick up exactly where you left off by resuming any session with a new prompt
* **Start new sessions**: Launch fresh sandboxes by entering a prompt and selecting a repository
* **Session details**: View the session details, full conversation, and sandbox execution history

## How it works

To demonstrate how remote agents work in Depot, we will use the `depot claude` command to demonstrate how remote Claude Code agents are launched in Depot.

**Note:** To run the command below, you should complete the [Quickstart guide for Claude Code](/docs/agents/claude-code/quickstart) first.

```shell
depot claude \
  --session-id feature-auth \
  --repository https://github.com/user/repo.git \
  --branch main \
  "Implement authentication flow"
```

This command will fire a request to the Depot control plane to start a new remote agent sandbox and return a url to the web UI where that Claude Code session can be monitored and managed.

Behind the scenes, Depot will do all of the following:

1. **Session creation**: A new isolated container is provisioned for your session, named after the `--session-id` you provided or generate a session ID if not specified
2. **Environment setup**: The sandbox comes pre-configured with development tools, languages, and libraries
3. **Load filesystem**: The sandbox can be prepopulated with a filesystem from a previous session via the `--resume` flag. If no previous session is passed in, a brand new file system is provisioned for the session.
4. **Git repository cloning**: If you specified a `--repository`, Depot will clone the repository into the sandbox and checkout the specified branch (if no branch is specified, it defaults to `main`)
5. **Session saving**: When Claude Code has finished it's work and exits, the session state is preserved for later resumption
6. **Easy resumption**: Use `--resume <session-id>` to continue from any environment

The following examples start a new session with one prompt, then resume it later to continue working.

**Using the Depot CLI:**

```shell
# Start a new session with a custom ID
depot claude \
  --session-id feature-auth \
  --repository https://github.com/user/repo.git \
  "Create a new branch called `feature-auth` and lets implement authentication flow for this new feature. Once you're happy with the initial implementation, commit your changes and push the branch to the remote repository."

# Later, resume the session to continue working
depot claude --resume feature-auth "This looks good, but we need to add the concept of a user profile now."
```

**Using the Depot dashboard:**

1. Navigate to the [**Claude Code** section](https://depot.dev/orgs/_/claude) in your dashboard
2. Click **New sandbox** to start a fresh session, selecting your repository and providing an initial prompt
3. Later, click on any session to view its details and use the prompt input to resume it with additional instructions

## Pricing

Depot remote agent sandboxes are available on **all plans*&#x2A; and are billed at a usage rate of **$0.01/minute** with no included usage for remote agents.

<DocsCTA buttonId="agents-docs-cta">
  <a href="/sign-up" className="font-semibold text-radix-grass12">
    Start your 7-day free trial to try remote agents on Depot →
  </a>
</DocsCTA>

## For AI Agents

The full site index is at [llms.txt](https://depot.dev/llms.txt). Append `.md` to any documentation, blog, changelog, or customer URL to fetch its markdown source directly.