Claude Code on a shedix machine's full desktop, opened full screen, next to the browser it shares with you.
The same desktop inside the machine's page in your browser: open it there, in a new tab, or full screen.
Phone view. The same desktop on a phone: tap the screen, type with the keys, steer with the trackpad.
Both turn off by themselves when nobody uses them, so they don't hold memory the agent needs for a build. ## What do you still decide? The skill is written so the agent asks before anything that's yours to decide: - **Private files:** before copying a project, the agent lists the files that don't travel with git, such as `.env` files and keys, without opening them, and asks which may go. Nothing private is copied without a yes. - **Public links:** it asks before sharing a link to your app. - **Your machines:** it never sleeps, stops or deletes a machine you're using unless you ask. - **The key:** the agent's access is a key of its own, which you can revoke on your account page at any time. How machines are kept apart from each other is on our [security page](/security). ## How does this compare with services you start from a dashboard or SDK? Each of these is built for its own job, and each can be started by software. The difference is who writes that software. Facts as of 2026-11-10: | Service | How a machine starts | Built for | What's inside | |---|---|---|---| | [E2B](https://docs.e2b.dev/) | `Sandbox.create()` from its Python or JavaScript SDK, with an API key | Apps that run model-written code in fresh sandboxes | A template you define | | [Daytona](https://www.daytona.io/docs/en/sandboxes/) | Its dashboard, SDKs, `daytona create` or its API; it also offers an MCP server | Long-lived sandboxes created from code | A snapshot or image you choose | | [Modal](https://modal.com/docs/guide/sandboxes) | `modal.Sandbox.create()` from its Python SDK | Python and GPU workloads at scale, with sandboxes beside them | An image defined in code | | [Codespaces](https://docs.github.com/en/codespaces/developing-in-a-codespace/creating-a-codespace-for-a-repository) | The repository's Code button, VS Code or `gh codespace create` | A person at an editor | A dev container from the repository | | [Fly.io Sprites](https://docs.fly.io/sprites) | Fly.io's CLI | Bursty work that mostly waits | A persistent Linux environment with its own URL | | shedix | One line to your agent, or one click | Your coding agents and you, on one machine | Ubuntu with root, Docker, Claude Code, Codex, a desktop, and a shared browser on the Web, Python and Go & Rust stacks | With an SDK or a CLI, an agent can start a sandbox too, once someone has given it the key, the template and the steps. The shedix skill is those steps, written for the agent: which machine to reuse, how to wait for it, how to copy a project without private files, how to hand over a session. ## What is each one built for? - **E2B** is built for an app that runs its model's code in a fresh sandbox from an SDK, often many times a day. - **Daytona** is built for sandboxes that last, created from code or a dashboard, with a shell your team can open. - **Modal** is built for Python and GPU work at scale, with sandboxes started from the same SDK. - **Codespaces** is built for a developer in an editor, with a machine made from the repository's dev container. - **Sprites** are built for bursty, mostly idle work, billed on what's used, from Fly.io's CLI. - **shedix** is built as a real computer for your coding agents: desktop, root, Docker, its own address, every way in. Your agent can start it and use a browser it shares with you; Claude Code and Codex come ready; you pay by the second while it runs, and a fraction of a cent an hour while it sleeps. ## How is it billed while the agent isn't using it? By the second while it runs, from $0.09/hr for 2 vCPU, 4 GB RAM, 30 GB disk. A machine that nobody uses goes to sleep by itself and keeps its files and running programs, for under a cent an hour: $0.006/hr for that machine. Claude Code or Codex working on it counts as use, so it stays awake while the agent works. A sleeping machine wakes in under a minute. Machines run in the United States. If you want to work with an agent's machine yourself, the [close the laptop](/use-cases/close-the-laptop) page shows the handoff from the person's side, and [give your coding agent its own machine](/blog/give-your-coding-agent-its-own-machine) covers why it helps. ## Which should you pick? If your product runs model-written code for its users, E2B, Daytona and Modal are built for that. If you want a machine made from a repository for yourself in an editor, Codespaces fits. If the machine is for your own coding agent, to start when it needs one and keep for days, a machine the agent starts itself saves the glue. Prices and features change, so check each provider's own pricing and docs before you choose. ## What to do next Read [who makes shedix](/about), then install the [shedix skill](/skills) and ask your agent to start a machine, or [start one yourself](/). Your first 12 hours of compute are on us.