Keep the coding environment you already have.

A remote control is useful when it brings you back to a working environment—not when it asks you to rebuild one.

A development server accumulates useful context: your repositories, build tools, authentication, and the small configuration choices that keep work moving. For many developers, the starting point is already a working EC2 box. The missing piece is access when the laptop is closed.

Keep execution where it belongs.

ccr-cloud runs coding sessions on infrastructure you control. In host mode, it uses regular Claude Code and the server account’s existing configuration. There is no required model router, extra LLM key, or new hosted workspace.

That choice also creates responsibility. A host-mode task has the account’s access. A separate Git worktree prevents two tasks from sharing the same checkout; it is not a security sandbox.

The phone should control the work. It shouldn’t become the machine doing the work.

Give the owner useful decisions.

A small screen needs more than a transcript. It needs a clear state: working, ready, waiting for approval, or stopped. An approval must identify the exact tool request. A follow-up sent during a turn should not be mistaken for a new parallel instruction.

Readable Markdown and exact-source Copy buttons help with the other half of the workflow: taking an answer or command back into your own tools without losing its formatting or quoting.

Separate connection from inference.

SSH forwarding is a useful first test. Tailscale and a protected Cloudflare hostname are existing access choices. We are also designing an optional paid managed connection, so users can pair their server once and receive a stable HTTPS address.

That service will not require our gateway or a coding plan. Your existing Claude login, Git setup, and model choice remain yours. It is a proposed managed HTTP connection—not a claim of blind end-to-end encryption.

Build the next layer in the open.

Codex support and multi-agent Rooms are separate tracks. Rooms would let independent sessions coordinate through a host-owned timeline and task board. They do not require moving agent execution or room authority into a commercial cloud service.

Start with the self-hosted guide, or follow the managed-connection design. The most useful feedback is a real workflow that gets stuck.