ユースケース

You're at your desk with a powerful machine running Claude Code or a long build. You move to a sofa, kitchen, or small laptop. Open the browser URL and continue watching and interacting — no SSH, no VPN, no setup.

Claude Code and similar agents run for minutes or hours. They pause for approval. With push notifications and mobile terminal access, you get notified the instant the agent needs you, confirm from your phone, and let it keep running.

You're paged at 2am. Instead of going to a computer, you pull up the terminal session on your phone, tail the logs, and run the fix. Full keyboard and output stream, in bed.

Your secondary device is a Chromebook, iPad, or old netbook with no local tooling. The heavy work runs on your main machine; the secondary device is just a browser. No SSH client, no VPN, no config.

Start work on your desktop in the morning. Walk to a meeting with your laptop. The tmux session is exactly where you left it — open the browser and reconnect. Sessions survive across browser disconnects.

Share the URL with a colleague. They watch your terminal in real time. No screen-share lag, no Zoom overhead, no installing anything. Read-only or interactive depending on session type.

Run a command in your terminal; students or audience see it instantly in their browsers. Every attendee watches the same session simultaneously. Works on any device with a browser.

Your cloud instance or Raspberry Pi runs services in tmux/screen. You SSH once to set up; forever after you monitor from the browser on any device, including mobile.

Multiple AI agents run in parallel terminal sessions. One browser page shows all sessions. Push notifications fire when any agent pauses. You triage from mobile, approve, and the fleet keeps going.

Share a terminal session with a colleague in readonly mode. No SSH key exchange, no VPN required — just share your UID and session ID. Ideal for code reviews, live debugging, training junior developers.

Run Claude Code under approved-only mode. Pre-approve safe interactions; all other input is blocked at the agent level — not just the UI. Ideal for supervised AI in sensitive environments.