What’s the best way to run Claude Code or Codex in the cloud, watch it work, and take over?
By NoClick, checked November 8, 2026. We make NoClick, listed first; every other company’s facts and prices are from its own pages, linked beside them.
The short answer
NoClick runs Claude Code, Codex and open-source harnesses on your own key or plan, with the harness’s own terminal to take over. Otherwise, pick by your plan and how much you need to drive.
- Claude Code, Codex or open-source harnesses from your own codeNoClick (we make it), with the harness’s own terminal to take over and hand back
- Claude Code on a Claude planClaude Code on the web, at no charge beyond your plan
- Codex on a ChatGPT planCodex cloud, built for reviewing results rather than watching
- A live IDE or desktop to take overDevin or Cursor’s Cloud Agents, running their own agents
- Build it yourselfAn E2B, Daytona or boat sandbox with tmux
NoClick.Your harness, your key, a terminal to take over.
NoClick
About $0.24 an hour of sandboxA developer platform where a coding agent on Claude Code, Codex, OpenCode, Hermes Agent or OpenClaw runs in a sandbox of its own per conversation, driven from the SDKs, the HTTP API or NoClick’s web chat.
- Best for
- Developers who want Claude Code, Codex or open-source coding harnesses running from their own code, on their own key or plan, with a real terminal to take over and hand back.
- Falls short
- Sandbox time is billed at cost, on top of your model plan.
How it works and full pricing
You create an agent on one of the five harnesses with your API key or plan: a Claude plan runs only Claude Code, and a ChatGPT plan runs Codex, OpenCode and Hermes Agent. A conversation’s first turn starts its sandbox. agent.watch(conversation, replay=True) follows it live, with what its commands print and the diffs it makes. agent.take_over(conversation) opens the harness’s own terminal on the agent’s session: Claude Code, Codex and Hermes Agent pause while you drive, and OpenCode and OpenClaw let you join the session they run. Messages that arrive meanwhile wait; hand_back() puts what you did into the conversation, and the agent’s next turn goes on from there. Only the account’s owner takes over, and it hands itself back after 15 minutes with nobody typing or 5 with no terminal open.
Model tokens go to your own key or plan. NoClick bills the sandbox’s running time, at 1 core and 4 GB: about $0.24 an hour, the same while you’ve taken over. Free includes $1 a day of usage, up to $5 a month, with a 15% fee on funds added; Member is $25 a month with $10 included and a 5% fee.
At a glance.
We compared which agents each option runs, whether it uses your own subscription or API key, whether you can watch the agent while it works, whether you can take over and hand back, and the price, from each company’s own pages on November 8, 2026.
| Option | Agents it runs | Your own subscription or key | Watch live | Take over | Price |
|---|---|---|---|---|---|
| NoClickours | Claude Code, Codex, OpenCode, Hermes, OpenClaw | Yes: Key, Claude or ChatGPT plan | Yes: Commands, output, diffs | Yes: Its own terminal, then hand back | About $0.24 an hour |
| Claude Code on the web | Claude Code | Partly: Claude plan; no API key | Yes: Browser or app | Partly: Steer; teleport makes a copy | From $20 a month |
| Codex cloud | Codex | Partly: ChatGPT plan; no API key | Not stated: Not described | Partly: Follow-ups; apply it locally | From $20 a month |
| Cursor Cloud Agents | Cursor’s agent | Not stated: Not mentioned | Yes: Run page | Yes: Take its desktop, then release | From $30 a month |
| GitHub Copilot cloud agent | Copilot; Claude and Codex in preview | Partly: Copilot plan | Yes: Session log | Partly: Steer; resume in its CLI | From $10 a month |
| Jules | Jules | Partly: Google AI plan | No: Notified when done | Partly: Approve its plan | Free tier |
| Devin | Devin | Not stated: Not mentioned | Yes: IDE, shell, browser | Yes: Its IDE and terminal | From $20 a month |
| OpenHands Cloud | OpenHands | Yes: Pick your model | Yes: Terminal, VS Code | Yes: Type in its terminal | Not published |
| Warp cloud agents | Warp, Claude Code, Codex | Yes: Your keys on Business | Yes: Its control room | Partly: Steer; pull it local | From $20 a month |
| Factory | Droid | Partly: Keys in its CLI and app | Not stated: Not described | Partly: SSH into its computer | From $20 a month |
| Your own sandbox (E2B, Daytona or boat, plus tmux) | Any CLI | Yes: Whatever you sign in with | Yes: Attach to tmux | Yes: Same pane, then detach | From $0.036 an hour (boat) |
The options.
Each runs a coding agent away from your laptop; they differ in which agents, whose subscription, and how much of the work you can see and drive.
Claude Code sessions that run in Anthropic-managed VMs, started from the browser, the Claude app, the Desktop app or claude --cloud.
- Best for
- Anyone already on a Claude plan who wants Claude Code in the cloud with the least setup.
- Falls short
- Claude Code only, on a claude.ai sign-in (no API key, Bedrock or Vertex), cloning and PRs need GitHub, and teleport forks the session rather than handing it back.
How it works and full pricing
You open the session on claude.ai or the Claude app to check progress, steer Claude, answer its questions, and review the diff. To take over, claude --teleport checks out the session’s branch and loads the conversation into your terminal, but the terminal gets its own copy: new work there stays local and doesn’t appear in the cloud session.
Included in Claude plans: Pro $20/mo, Max $100/mo, Ultra $200/mo, Team $25 or $125 per seat (monthly). No separate charge for the cloud VM; sessions count toward plan usage limits.
Sources:code.claude.comclaude.com
Codex tasks that run in their own cloud workspace, started from the web, the Codex app, the IDE extension or the CLI.
- Best for
- ChatGPT subscribers who want to hand Codex tasks off and review the result.
- Falls short
- Its pages don’t describe watching a task’s terminal live or taking it over mid-task, and cloud tasks need a ChatGPT plan, not an API key.
How it works and full pricing
Each task runs in a workspace built from an environment you publish (repositories, setup script, network rules). You review the changed files and test results, ask for follow-ups, then commit or open a pull request; the CLI can start tasks, check their status and apply a result to your local repository.
Plus $20/mo; Pro from $100/mo; Business $25/user/mo billed monthly ($20 billed annually). The API key option has no cloud features.
Cursor Cloud Agentsby Anysphere (Cursor)
From $30 a monthCursor’s own agent running in isolated cloud VMs with full development environments (formerly Background Agents).
- Best for
- Cursor users who want a cloud agent with a desktop they can drive.
- Falls short
- It runs Cursor’s agent, not Claude Code or Codex, its page doesn’t mention bringing your own key, and it’s not on the Free or Plus plans.
How it works and full pricing
Agents run on a curated set of models and leave artifacts (screenshots, videos, logs) showing what changed. You can take control of the agent’s desktop to test the software yourself, then release control back to the agent; teammates who open a run’s URL see it read-only.
Premium $30/mo, Teams $30+/user/mo; cloud agents use your plan’s usage at the selected model’s rate.
Sources:cursor.comcursor.com
Copilot’s coding agent working on your GitHub repositories in a cloud sandbox or a GitHub Actions-based environment, plus third-party Claude and Codex agents.
- Best for
- Teams whose work starts from GitHub issues and pull requests.
- Falls short
- GitHub repositories only, Actions-based sessions stop at 59 minutes, steering isn’t available for third-party agents, and the plans page contradicts itself on whether Pro includes them.
How it works and full pricing
You monitor a session’s progress in real time from the agents panel, steer it with follow-up messages (applied after its current tool call) or stop it, and continue a cloud session in Copilot CLI with /resume. Claude and Codex can be assigned to issues too, and the Claude agent can run on your Copilot subscription.
Pro $10/mo, Pro+ $39/mo, Max $100/mo per user. Sessions draw on AI credits and the account’s compute allowance.
Sources:docs.github.comdocs.github.comdocs.github.comgithub.com
Julesby Google
Free tier; Pro $19.99 a monthGoogle’s asynchronous coding agent that clones your repo into a cloud VM and works there.
- Best for
- Plan-first background tasks on a free or Google AI plan.
- Falls short
- It runs Google’s agent only, its docs describe no live terminal view or take-over, and paid plans are for individual @gmail.com accounts.
How it works and full pricing
Jules writes a plan you review and approve before it changes code, then notifies you when the task completes or needs your input. The Jules CLI lists your sessions and pulls a completed session’s changes locally.
Free: 15 tasks a day, 3 at once. Jules in Pro (100 a day, 15 at once) comes with Google AI Pro, $19.99/mo; Jules in Ultra with Google AI Ultra.
Devinby Cognition
From $20 a monthAn autonomous software engineer with its own machine, IDE, shell and browser.
- Best for
- The fullest live view of a hosted agent’s IDE, shell and browser.
- Falls short
- It runs Devin’s own agent, and its pages don’t mention running Claude Code or Codex or bringing your own key.
How it works and full pricing
You watch Devin make edits in real time in its embedded IDE, watch its commands and browsing, and stop the session to take over the IDE, with full terminal access, or the Interactive Browser for logins and CAPTCHAs.
Pro $20/mo (includes Devin Cloud); Max $200/mo; Team $80/mo plus $40/mo per full dev seat.
OpenHands Cloudby All Hands AI
Not publishedThe hosted version of the open-source OpenHands agent, at app.openhands.dev.
- Best for
- An open-source agent with a shared terminal and editor you can type into.
- Falls short
- The hosted cloud runs the OpenHands agent; running Codex or Claude is listed for the self-hosted plans, and the cloud’s price isn’t published.
How it works and full pricing
While it works you see a Terminal tab where both OpenHands and you can run commands, an embedded VS Code to browse and edit files, a view-only browser tab and the running app. Its LLM setting lets you pick another LLM instead of the OpenHands provider.
OpenHands Cloud’s price isn’t on the pricing page (its Billing tab adds credits for the OpenHands provider). Agent Canvas, run on your laptop or a shared VM, is free; self-hosted Teams is $500/mo for up to 10 users.
Sources:docs.openhands.devdocs.openhands.devdocs.openhands.devopenhands.dev
Cloud runs of the Warp Agent, Claude Code or Codex on Warp-hosted compute or your own infrastructure, with Warp Factories for teams.
- Best for
- Teams that want Claude Code and Codex runs managed beside Warp’s own agent.
- Falls short
- Third-party harness access depends on your plan, Factories are invite-only, and its pages don’t say whether the cloud run keeps going once you pull it local.
How it works and full pricing
Runs show in Warp’s control room live and afterwards, and open on the web or a phone. You steer by sending a message or approving an action without restarting the run, and take over by pulling the in-progress session into your own terminal or IDE. With your provider credentials, the model provider bills you; Warp bills compute and platform usage.
Free with limited cloud agents; Build $20/mo with $20 of agent usage at API rates; Business $50/user/mo with your own API keys. Warp Factories are in closed beta.
Factory’s Droid agents, run locally or as cloud background agents, with persistent Droid Computers it hosts.
- Best for
- Droid users who want a persistent cloud computer they can SSH into.
- Falls short
- It runs Droid rather than Claude Code or Codex, your own keys don’t reach its hosted web or mobile surfaces, and its pages don’t describe watching or taking over a running session.
How it works and full pricing
A Droid Computer keeps its packages, files and services between sessions, pauses when idle, and opens an interactive shell with droid computer ssh <name>. Bring-your-own-key works in the Droid CLI and desktop app.
Pro $20/mo (cloud and local background agents); Plus $100/mo adds Droid Computers; Teams $60/mo plus $40/mo per seat.
Your own sandbox (E2B, Daytona or boat, plus tmux)by E2B, Daytona, boat
From $0.036 an hour (boat)A cloud sandbox you rent by the second, with Claude Code, Codex or any CLI you install running inside tmux.
- Best for
- Full control over the agent, its login and its terminal, if you’ll build the rest.
- Falls short
- You build the session handling, secret hygiene, idle shutdown, history and hand-off yourself, and a person in the terminal can read every key you put there.
How it works and full pricing
Start the agent’s CLI in a tmux session in the sandbox, then attach to watch it and type into the same pane to take over; detach to hand it back. E2B connects a terminal with e2b sandbox connect <sandbox-id>; Daytona has a browser web terminal and daytona ssh <sandboxId>; boat’s VMs take SSH and come with Claude Code’s and Codex’s dependencies, plus a desktop. tmate’s public servers shut down in July 2025, so sharing a link needs your own tmate server.
E2B: Hobby $0 plus $100 one-time credits (1-hour sessions), Pro $150/mo plus usage (24-hour sessions), $0.000014 per vCPU-second and $0.0000045 per GiB-second. Daytona: pay as you go, $0.0504 per vCPU-hour and $0.0162 per GiB-hour, $200 free compute. boat: $0.036 an hour for 4 vCPU and 8 GB, with a $20 monthly minimum.
Sources:e2b.devdocs.e2b.devdaytona.iodaytona.ioboat.devgithub.com
Start with a key.
Create an API keyOr paste one prompt into your coding agent. It saves a key to your .env.
Get the prompt in the consoleOr add NoClick’s MCP server to your client.
Questions
Yes. Claude Code on the web runs cloud sessions on Pro, Max, Ultra and Team plans in Anthropic-managed VMs, counted against your plan’s usage with no separate VM charge. It needs a claude.ai sign-in: an API key, a Console sign-in, Bedrock or Vertex can’t start cloud sessions.
Not according to OpenAI’s pricing page, which says the API key option has no cloud-based features; cloud tasks come with ChatGPT Plus ($20/mo), Pro and Business. To run Codex in a cloud sandbox on an API key, use a service that runs the Codex CLI with your key, such as Warp’s Codex harness, NoClick, or your own sandbox.
Devin and OpenHands show the agent’s shell and editor live, and Claude Code on the web and Copilot show a session’s progress as it runs. On NoClick, agent.watch(conversation, replay=True) streams a coding agent’s commands, their output and its diffs, and the web chat shows the same as a read-only terminal; on your own sandbox, run the CLI in tmux and attach.
Cursor’s Cloud Agents let you take control of the agent’s desktop and release it back, and Devin lets you stop it and use its IDE and terminal. Claude Code’s --teleport and Warp pull a session into your own terminal instead. NoClick opens the harness’s own terminal on the agent’s session and, when you hand it back, the agent’s next turn goes on from what you did, with what you typed in its history.
A terminal shows whatever keys and tokens the sandbox holds, so limit who opens it. On NoClick only the account’s owner can, signed in or with a secret key holding the agents:terminal scope (not included in *), and the browser gets a single-use token that expires in two minutes and is sent in the websocket’s first message, never its URL.
Claude Code on the web and Codex cloud are included in their plans, from $20 a month. Where you bring the model yourself, you pay for compute: NoClick bills about $0.24 an hour of sandbox time while the agent works and for about a minute after each reply, and E2B and Daytona charge by the second, about $0.05 per vCPU-hour plus memory, and boat $0.036 an hour for a 4 vCPU VM.
OpenHands Cloud runs its own open-source agent, and on a sandbox from E2B, Daytona or boat you can install any CLI. NoClick runs OpenCode, Hermes Agent and OpenClaw beside Claude Code and Codex, each on a key or plan from the model’s provider, and all five can be taken over.
On NoClick, live editing runs an app’s dev server in a container, and an agent on the sdk, claude_code or codex harness edits it there: each change shows in the open pages within a second, and the session’s end deploys only if the build and its check pass. It works for Vite, React Router 7 or 8 on the Cloudflare Vite plugin, Astro, Next.js 16.4.1 or later, and sites of plain files.