comparison
reachpad vs Claude Code on the web
Both keep a run going after you close the laptop. They disagree about what a session is: Anthropic's is a task against a repository, reachpad's is a machine you keep.
Side by side
| Claude Code on the web | reachpad | |
|---|---|---|
| Where code runs | A cloud VM, Anthropic-managed by default | A Linux VM on the reachpad fleet, where you are root |
| Needs a GitHub repo | Yes, or a local repo bundled up with --cloud | No. Your own git remotes, or no repository at all |
| State between sessions | The repo is cloned fresh each session; your local config does not come with it | Disk and memory are sealed to object storage, so a paused workspace resumes mid-session |
| Which agents | Claude | Claude Code, Codex and OpenCode, or anything you install |
| Whose account the agent bills | Your Anthropic plan | Yours, signed in inside the workspace; reachpad is not in the middle |
| Network from the session | Set per environment; the default level reaches allowlisted registries | The machine's own network |
| Keeps running when you disconnect | Yes | Yes |
| How you drive it | claude.ai or the Claude mobile app | The reachpad CLI, or a browser terminal on a pad |
Anthropic's own comparison of its four surfaces is worth reading before ours: Claude Code on the web quickstart. The rows above describe it as documented on 12 August 2026, while it is in research preview for Pro, Max and Team.
The difference that matters
A cloud session that clones your repo is built for work that starts from a repository and ends in a branch. That covers a great deal of real work, and when it fits, it fits with no setup at all.
It does not fit work whose value is in the machine rather than the diff: a half-finished migration with a database running next to it, a long-lived service you keep poking, a scratch environment with twenty things installed that you would rather not install again. Cloning a repo cannot carry any of that, because none of it is in the repo. A workspace whose disk and memory survive can, which is why reachpad seals both and why an attach picks up where the last one stopped rather than at a fresh boot.
The second difference is plurality. reachpad does not know or care which agent you run: several can run at once in one workspace, each in its own terminal, planning in one and implementing in another.
When Claude Code on the web is the right answer
When the task is well-defined, starts from a GitHub repository, and what you want back is a pull request. It is included in a Claude subscription you probably already pay for, there is nothing to install, and it will be ready before a reachpad workspace has finished its first apt install. If that describes most of your work, use it, and come back when you hit the day it does not.
Trying reachpad
1,000 one-time trial credits, 16 hours 40 minutes of active workspace time are shared across up to 10 workspaces, with no card required. The quickstart is five steps and the fifth one is closing your laptop.
Checked against published documentation on 12 August 2026. Something out of date or unfair? The other comparisons, or this page as raw markdown: /compare/claude-code-on-the-web.md