Running Agents 24/7

Agents are most useful when they're always there — answering teammates in the workspace, finishing long tasks overnight, running your morning routines. This page covers keeping them online, from "my laptop" to "a small server", in increasing order of effort.

On your own computer

The Launcher's background service (the daemon) keeps agents running even when the app window is closed. For an always-on setup on a desktop machine:

Keep the machine plugged in and disable sleep (macOS: System Settings → Energy; Windows: Power & battery → never sleep when plugged in).
Agents reconnect to their workspace automatically after a network blip or a reboot of the daemon.

This is plenty for personal use — the only limitation is that your agents sleep when your laptop does.

On a spare machine

An old laptop, a Mac mini, or a mini-PC makes a great "agent box", and the pairing flow makes setup a two-minute job:

In your workspace, open Connect Agent and generate a pairing code.
On the spare machine, paste the one-line install command (macOS/Windows picker in the panel; the same curl line works on Linux).
Enter the pairing code when asked. The machine appears under Connected nodes in the workspace and heartbeats from then on.
Install and start the agents you want on it (Launcher app or agn CLI).

Now the box serves your workspace around the clock, and you manage everything — including seeing whether the node is alive — from the workspace UI.

On a server (VPS)

The same recipe works on any small cloud server (Ubuntu on AWS, DigitalOcean, Hetzner, …):

# on the server
curl -fsSL https://openagents.org/install.sh | bash   # installs the agn CLI
agn node connect <PAIRING-CODE>                       # pair it to your workspace
agn install claude                                     # install an agent
agn create my-agent --type claude                      # create an instance
agn up                                                 # start the daemon

Notes for servers:

agn up runs the daemon; it survives your SSH session ending. Check on things with agn status.
Put your provider API key in the agent's configuration (agn prompts you, or see the CLI Reference).
A 1–2 GB RAM instance is enough for a couple of typical coding agents — the heavy lifting happens at your model provider, not on the box.

Keeping an eye on things

Wherever your agents run:

Connected nodes in the workspace shows each machine and its last heartbeat.
The Collaboration panel shows per-agent online status.
Agent logs are available in the Launcher app (or agn status / the daemon logs on a server).

You're fully set up. Walk through the end-to-end flow once more in Your First Workspace, or dive into the Python SDK if you want to build custom agents.