Keep a Coding Agent Running While Your Laptop Sleeps
Happy Agent on a server keeps Claude Code and Codex working while your laptop sleeps, and so, in their own ways, do tmux, Claude Code Remote Control on a server, exe.dev, and T3 Code. To set up Happy Agent, see Run Claude Code and Codex on a Remote Server.
A harness, not a remote control
- A remote control forwards a session that runs in the vendor's own CLI on your computer. Claude Code Remote Control works this way, and so does the original Happy CLI. The session lasts as long as that process and that computer.
- Happy Agent is the harness. It runs the agent loop itself, with its own tools, sandbox, and review, and calls the models directly. Put it on a server and the sessions keep going while your laptop sleeps; Desktop and the phone are windows onto them.
It reuses the subscriptions you already pay for:
-
Claude goes through Anthropic's own Claude Agent SDK, which you can use with your Claude plan. Anthropic announced a change that would have moved SDK usage onto a separate credit, then paused it. From its help center, Use the Claude Agent SDK with your Claude plan:
Update June 15, 2026: We’ve paused the previously-announced changes to Claude Agent SDK usage. For now, nothing has changed: Claude Agent SDK,
claude -p, and third-party app usage still draw from your subscription limits.Its October 7 update adds: "You can still use the Claude Agent SDK,
claude -p, and third-party apps with your subscription limits." -
Codex uses the ChatGPT sign-in from your Codex CLI. "Codex is included across ChatGPT plans," says OpenAI's help center, and OpenAI invites open-source apps to build on those plans: "your open-source app can request permission to use the user’s ChatGPT plan" (Sign in with ChatGPT).
Where Happy came from
We started Happy as the Happy CLI, happy on npm. It runs Claude Code, Codex, and other ACP-compatible agents in your terminal, unchanged, and lets you control them from the Happy app. It is still the way to keep the exact original CLIs and the terminal on your own machine, and its legacy docs are still here, in maintenance mode. Happy Agent and Happy Desktop are the current iteration, where Happy runs the harness itself, on your computer or a server.
Alternatives to a Happy Agent server
A coding agent runs on the machine where you start it. To keep one working, keep that machine awake and online, or run the agent somewhere that stays on:
| tmux + SSH or Mosh + Tailscale + Termius | Claude Code Remote Control | A VM with its own agent (exe.dev and Shelley) | T3 Code | Happy Agent on a server | |
|---|---|---|---|---|---|
| Agents | Any terminal agent | Claude Code | Shelley; Claude Code, Codex, and Pi preinstalled | Claude Code, Codex, Cursor, Grok Build, OpenCode, Pi, Muse Code, and any ACP Registry agent, on your existing subscriptions | Claude, Codex, and Grok models in one harness, reusing your subscriptions; Gemini and any ACP agent (beta) |
| Runs on | Your server | Your computer, or a server inside tmux | An exe.dev VM | Your computer, or a Linux or Apple silicon server it launches over SSH | Your server, VPS, or Mac |
| While your laptop sleeps | Keeps running | On a laptop, pauses until it wakes; on a server, keeps running | Keeps running | On a laptop, stops; on a server with t3 service install, keeps running, and a Mac host must stay logged in and awake | Keeps running |
| Steer it from | A terminal app on any device, over SSH or Mosh | The Claude app, claude.ai/code, or Claude Desktop | Shelley's web page; SSH or the browser terminal for the other agents | T3 Code desktop, the web app, and the iOS and Android app; outside agents over MCP | Happy Desktop and the Happy phone app |
| Approvals | Terminal prompts, typed by hand | Permission prompts in the app, with push notifications | Depends on the agent; the VM is the boundary | Supervised, Auto-accept edits, Auto (the provider's own review), or Full access, the default; push notifications through T3 Connect; no sandbox of its own | Auto mode reviews actions itself, inside the OS sandbox |
| With a team | Shell accounts on the server | One person's sessions; organization controls on Team and Enterprise | Share the VM with Web or Root access, per person or with the whole team | No team mode in its docs; one owner pairs devices and agents with scoped links, such as read-only | Team mode: each member signs in as themselves, in shared projects and sessions |
| In between | Tailscale's coordination and relays; traffic WireGuard-encrypted | Anthropic stores the transcript while connected | exe.dev runs the VM and its HTTPS proxy | T3 Connect, with Clerk sign-in and a Cloudflare Tunnel per machine; without it, direct LAN, Tailscale, or SSH | Tailcat relay for Desktop, WireGuard-encrypted; Happy's relay for the phone, end-to-end encrypted |
| Setup | Install and connect four tools | One command | Create a VM | One install command or the desktop app, then t3 connect to reach it from elsewhere | Ask the Chief of Staff; it follows a public recipe |
Where each is better:
- tmux + SSH or Mosh. Works with every terminal agent and tool, with nothing new to trust. Mosh survives sleep and network changes, and Termius speaks it on iOS and Android.
- Remote Control. Nothing to install if you already use Claude Code and the Claude app, Anthropic supports it, and permission prompts reach your phone as push notifications. Team and Enterprise get admin controls. See Happy vs Remote Control.
- exe.dev. A persistent VM with agents already installed, Shelley, a web agent that works on mobile, and VM sharing that follows team membership. No service to set up. You can also run Happy Agent on it; see exe.dev in the remote server guide.
- T3 Code. Works with more agents: seven built in, plus the ACP Registry. It has a polished free phone app with push notifications and Live Activities, routes work across several machines, launches servers over SSH from its desktop app, and has a much larger community (26.8k GitHub stars; T3 says over 400,000 developers use it).
- Happy Agent. One runtime for three model families under one sandbox and review, a desktop app and a phone app built for agent sessions, and a team mode where everyone keeps their own identity. The cost: more setup than the others, and you run and upgrade the server.