Claude Code, Codex and other coding agents can work for hours on one task. On a laptop they stop when the lid closes, the battery runs out or the Wi-Fi drops. This guide shows how to keep them running on a server you own, around the clock.
Why a laptop is the wrong place
An agent is a process on the machine where you start it. Close the laptop and the process sleeps; reopen it and the agent has lost its connection, often in the middle of a change. Long tasks such as a refactor across a codebase, a test suite that needs fixing or a migration to a new framework version take longer than most people want to keep a laptop open.
A server never sleeps. Start the agent there and it keeps going while you are offline.
What the server needs
Running an agent on a bare server is only half of the job. It also needs the things it works with:
- the language runtimes and package managers of your projects;
- databases, Redis and the other services your code expects;
- a way for you to open the sites it builds in your browser;
- a terminal session that survives when your SSH connection drops;
- a way to tell you when it is done or waiting for your answer.
Setting this up by hand takes a day and has to be kept up to date. rdev does all of it in one install.
Running agents with rdev
- Get a server with at least 4 GB of RAM. The panel can create one at Hetzner, DigitalOcean and other providers for you.
- Install rdev. The server gets the whole development environment in a container, with your projects, services and the agent profile.
- Open a terminal on the server from the laptop app or any SSH app and start the agent in a project, for example
claudeorcodex. - Give it the task and close your laptop.
Every terminal on the server is a persistent session: when you come back, from the same laptop or another device, you are in the same session with the agent's full output. Notifications tell you on your phone when the agent finishes, needs your approval or hits an error, and the panel's Agents page shows which sessions are working or waiting.
What the agent can do on its own
Agents on rdev know the environment from the first day. Through MCP tools they create sites, add databases and services, install PHP extensions, start queue workers and read logs themselves instead of asking you. Checkpoints let you undo their changes in one step. See AI agents for the supported agents and what each plan includes.
Costs
You pay three separate bills: the server to your provider (from about €4 a month), your rdev plan, and the agent's own usage to Anthropic, OpenAI or another model provider. rdev never sees your prompts or your code; see Your server, your root.