## Managed agents Run Claude Managed Agents for your application or automated workflow with tool execution in [Depot sandboxes](/content/docs/sandboxes/overview/index.html). Anthropic manages the agent loop, and Depot runs your agent's commands and file operations. View run status, timing, and failure details in your Depot dashboard. ## How it works A [self-hosted environment for Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes) runs tool calls on compute you choose while Anthropic manages the agent. You configure an agent's model, instructions, and tools in the Claude Console or API. Start a session with that agent and a self-hosted environment connected to Depot. Anthropic queues the session. Depot checks the queue and starts an environment worker in a sandbox. Anthropic runs the agent loop and requests tool calls. The worker executes commands and file operations, then sends the results back to Anthropic. How Claude Managed Agents uses Depot sandboxes: Start a session through the Claude Console or API with your agent and self-hosted environment. Anthropic queues the session and runs the agent loop. Depot polls the queue, claims work, and starts an environment worker in a sandbox. The worker connects to Anthropic over outbound HTTPS, receives tool calls, executes commands and file operations, and sends results back to the agent loop. Tool inputs and outputs flow to Anthropic so Claude can decide what to do next. See Anthropic's [self-hosted sandboxes guide](https://platform.claude.com/docs/en/managed-agents/self-hosted-sandboxes) for details. ### What's the difference between Claude Code and Claude Managed Agents on Depot? - **Claude Code**: Use Claude's web, desktop, or mobile interface for coding tasks. Depot runs the agent loop and tools. Follow the [Claude Code guide](/content/docs/managed-agents/claude-code-self-hosted/index.html). - **Claude Managed Agents**: Build agents for applications or automated workflows. Anthropic runs the agent loop and Depot executes commands and file operations. ## Get started ### Prerequisites Before you start, you need: - A [Depot account](/content/sign-up/index.html) with the **Owner** role in your Depot organization. - A [Claude workspace](https://platform.claude.com/) with Managed Agents access and an agent configured with command and file tools. If you don't have an agent, follow Anthropic's [Managed Agents quickstart](https://platform.claude.com/docs/en/managed-agents/quickstart). ### Create your environment and get credentials Get an **API key**, an **Environment key**, and an **Environment ID** from the same Anthropic workspace. 1. In the [Claude Console](https://platform.claude.com/), select the workspace that contains your agent. 2. In [**Environments**](https://platform.claude.com/environments), create an environment with a **Name** and **Hosting** set to **Self-hosted**. 3. Save the environment's **Environment ID**, which starts with `env_`. 4. Click **Generate environment key** and save the key. 5. In [**API keys**](https://platform.claude.com/settings/keys), create a key scoped to the same workspace. Under **Linked account**, select a service account. Save the API key. A service account keeps the connection independent of any one person's account. Ask your Anthropic administrator to provision one if none is available. Depot keeps the API key outside the sandbox. The worker uses the environment key to connect to Anthropic. ### Connect Depot to your Claude Managed Agents environment 1. Open [**Settings** \> **Connections** \> **Agent providers**](/content/orgs/_/settings/connections#agent-providers/index.html) in Depot and select **Anthropic**. 2. Enter your API key in **API key**. 3. Enter your environment key in **Environment key**. 4. Enter your environment ID in **Environment ID**. 5. Click **Connect**. Depot shows **Connected to Anthropic** when the connection succeeds. ### Choose a sandbox size After connecting, select a **Sandbox size** for the connection and click **Save**. The default is `4x16` (4 CPUs, 16 GB memory). See [sandbox sizes](/content/docs/managed-agents/overview#sandbox-sizes/index.html) for available sizes, billing, and when changes take effect. ### Start a session The examples below use `curl`. You can also use Anthropic's CLI or SDKs for Python, TypeScript, and other languages. See Anthropic's [Managed Agents quickstart](https://platform.claude.com/docs/en/managed-agents/quickstart) for installation and session examples. Use your existing agent and the self-hosted environment you connected to Depot. Replace `` and `` with your agent ID and environment ID. Set `ANTHROPIC_API_KEY` to an API key from the same Anthropic workspace. ``` curl --fail-with-body https://api.anthropic.com/v1/sessions \ -H "x-api-key: $ANTHROPIC_API_KEY" \ -H "anthropic-version: 2023-06-01" \ -H "anthropic-beta: managed-agents-2026-04-01" \ -H "content-type: application/json" \ -d '{"agent":"","environment_id":""}' ``` Copy code Copy the session's `id` from the response. Send a first message, replacing `` with the session ID: ``` curl --fail-with-body 'https://api.anthropic.com/v1/sessions//events' \ -H "x-api-key: $ANTHROPIC_API_KEY" \ -H "anthropic-version: 2023-06-01" \ -H "anthropic-beta: managed-agents-2026-04-01" \ -H "content-type: application/json" \ -d '{"events":[{"type":"user.message","content":[{"type":"text","text":"Create a hello.txt file containing Hello from Depot."}]}]}' ``` Copy code The agent creates `hello.txt` in the Depot sandbox. View the agent's response in the Claude Console, or use the API to [stream session events](https://platform.claude.com/docs/en/managed-agents/quickstart#send-a-message-and-stream-the-response). ## Verify your run in Depot Open [**Agent runs**](/content/orgs/_/agent-runs/index.html) in your Depot dashboard and select **Claude Managed Agents** to see your runs, newest first. Select a run to see its status, timing, duration, and sandbox ID. Use **Open in Claude Platform** to view the session in the Claude Console when the link is available. If a run fails, check **Why it failed** for details. ## What's next Give your agent access to a repository and its dependencies, then use it to investigate a failing test and propose a fix. Your application starts the session and supplies the task; Depot runs the commands and file operations. Continue working by sending another message to the same Claude Managed Agents session. If the session needs a new sandbox, Depot restores the latest available workspace snapshot. See [Session continuity](/content/docs/managed-agents/overview#session-continuity/index.html) for details and limitations.