blog

Keep coding agents running when you close your laptop

The run is not on your laptop. It is on your pad, a machine that stays up whether or not you are looking at it. Closing the lid closes your view of the run.

Where the run actually happens

When you start an agent on reachpad, the agent process starts on the pad. Your browser is a terminal onto that machine, the same way an ssh client is. The work, the file writes, the test runs and the network calls all happen there.

This is the difference from running an agent locally. A local run is a process on your machine, so it is subject to your machine: sleep, a closed lid, a dropped wifi connection, a reboot for an update. A run on the pad is subject to the pad, and the pad is not going to lunch.

Your pad has its own hostname on reachpad.dev. That address is the pad, not a session, so it means the same thing tomorrow as it does now.

What survives when you disconnect

Everything on the machine, because the machine is still running:

Agent sessions that were mid-run. They keep going and finish without you.
Files and working directories, including uncommitted edits an agent made while you were gone.
Git state. Branches, stashes and remotes are where you left them.
Your setup. Dotfiles, shell history, and anything you installed stay installed.

The pad is provisioned once for your account and kept, rather than built fresh each time you open it. That is why the second visit looks like the first one ended.

Getting back in

The browser terminal. Sign in and press “Open terminal in browser” on your dashboard, or go to your pad’s own address; both land you in the same terminal. Nothing to install, which is what makes a phone or a borrowed machine work as well as your own laptop.

You land back on the same sessions. Reconnecting is a reattach rather than a restart, so an agent that has been working for an hour is an hour in when you open it, not starting over.

What still waits for you

An agent that needs an answer stops and waits for one. That is not a failure mode, it is the agent doing the right thing: it will not guess at a question it should ask you. What changes on a pad is that the wait is cheap. The run has not died, the machine has not slept, and the twenty minutes you spent at lunch cost you twenty minutes rather than the whole afternoon.

Every session has a tab in the terminal, and each tab carries its state: running, your turn to answer, or stopped on a permission it needs. Coming back is a matter of answering the ones that stopped rather than opening each one to check.

The agents stay current

Claude Code, Codex, OpenCode and Kimi come preinstalled. The pad updates them between your visits, so the version you get is the current one without you running an upgrade.

You sign in to each agent with your own accounts. Your keys, your config, your usage. reachpad runs the agents, it does not resell them or sit in the middle of your agent’s auth.

What this does not do

It does not make an agent correct, and it does not make a long run shorter. A refactor that takes three hours takes three hours; the difference is that you are not required to be present for them.

The free session behaves differently from a paid pad. It sleeps when you step away and resumes where you left off, which is right for a session you open and close, but it is not the same as a pad that stays up through a long unattended run. See pricing for the difference.

Pricing: the free session, the pad, and enterprise terms.
Common questions: agents, accounts, and how code gets in and out.
Something here not answered? Write to us and a person answers.