Running Claude Code in the cloud

Claude Code is a terminal agent. It reads your files, runs your tools and edits your code, all on the computer where you launched it. That locality is the reason it is good — it has your actual project, not a copy — and it is also the reason a long job is awkward. The work stops being about the code and starts being about the laptop: leaving it open, keeping it plugged in, not closing the lid on the train.

Anthropic shipped Channels in March 2026, which lets you talk to a running session from Telegram, Discord or iMessage. It is genuinely useful and it solves the where am I typing problem. It does not solve the where is it running problem: your machine still has to be on, and if the agent hits a prompt that needs your permission, it waits there until you are back at the desk.

The zero-cost answer. Turn off sleep, leave it plugged in, and use Channels or SSH to reach the session from your phone.

It works, and for a single afternoon job it is the right call. What it costs you is the machine: it cannot be shut, moved, or updated while the work is in flight, and a laptop that sleeps at the wrong moment takes the session with it. It also means the thing your agent depends on is the same thing you carry through airports.

A few dollars a month gets you a small Linux box that never sleeps. Install Node, install Claude Code, add your API key or sign into your subscription, clone your repository, start a persistent session in tmux, and reach it over SSH.

This is the most flexible option and the one most developers reach for first. The honest cost is that you now run a server. Something has to restart the session after a reboot, keep the CLI updated, hold your credentials safely, and be debugged at the exact moment you wanted to be doing something else. It is not hard. It is simply never finished.

Several services will run an agent on a container for you, from a few dollars a month, and hand you a browser terminal to reach it. You skip the server administration and keep most of the flexibility.

The catch is usually the model access: you bring your own API key or your own Claude subscription, so the hosting fee is on top of what you already pay Anthropic. You are also back in a terminal, which is fine if a terminal is what you wanted.

The last option turns the arrangement around. Instead of you having a machine and putting an agent on it, the agent has a machine and you have a chat.

That is what we build. You message a bot in Telegram; it creates an engineer with its own name, its own chat, and its own cloud computer. Claude Code is already installed and already signed in — there is no API key to supply and no separate Anthropic subscription to hold, because the AI comes with it. The machine stays awake when you close your laptop, and the work continues while you are asleep.

The tradeoff is real and worth stating: you are talking to it through a chat window rather than a terminal, and the machine belongs to the agent rather than to you. If what you wanted was a shell on a box you control, option 2 is the honest answer and you should take it.

Leave the laptop on if the job finishes today and you are not going anywhere.

Rent a VPS if you want a shell, you enjoy owning the setup, and you would rather pay in attention than in money.

Use managed hosting if you want the box without the administration and you already have model access you are happy to point at it.

Give the agent its own machine if you would rather describe the job than run the tooling — and especially if you want the answer waiting for you in the morning rather than a session you have to go back and check on.