Approve the next step from anywhere.
When an agent wants to run a command or edit a file, the prompt reaches your browser and your phone. Allow it once, deny it, or stop it asking for the rest of the chat.
How approvals workClaude Code, Codex, and Grok
All in one remote
for Coding Agents
Remote control Claude Code, Codex, and Grok across your machines.
Start tasks, review changes, and approve work from your desktop or phone without any VPNs.
14-day free trial · not charged until day 15 · cancel any time
Works with the subscriptions you already have.
Termdeck drives the official CLIs with their own logins. Keep your plans, local tools, and MCP configuration.
Pick the agent per chat, not per app. All three share one transcript, one approval card, one diff view, and one usage page, so when one agent hits a limit you open another in the same folder and keep going.
Remote control
One remote for every agent, on every machine.
Claude Code has Remote Control, Codex has the ChatGPT app, and Grok Build has no remote at all. Each covers one agent on one computer. Termdeck is one remote for all three: start, steer, approve, and stop any session on any machine you own, from a browser tab or your phone.
Why one remote beats three- Claude Code, from anywhereFollow the transcript, answer prompts, and send the next message from your phone. Every session on every machine, not one at a time. Compare with Remote Control
- Codex CLI, not just the appDrive Codex CLI itself on Linux, Windows, macOS, or a server. Send a follow-up mid turn and Codex picks it up. Codex remote control
- Grok Build, finally remoteGrok Build ships with no remote of its own. Start, steer, and approve Grok on your machine from any browser. Grok remote control
- Approve from the lock screenPush notifications carry Allow and Deny buttons, so a blocked run moves again without opening anything.
- No VPN, no port forwardingA small agent on each machine dials out. Nothing to open on your router, no SSH tunnel to keep alive.
- Steer, stop, switchChange model or effort, stop a run that went wrong, or switch account when a rate limit hits.
- Read the diff on the goSee exactly what an agent changed before you let it carry on.
- Back at the deskThe chat is the CLI's own session. Carry on with
claude --resumeorcodex resume.
Terminals
Every agent on one screen, tiled like tmux.
Put Claude Code, Codex, and Grok chats side by side in panes, from any machine you connect. Each pane reads like the CLI you already know, and every action is a click.
Why Terminal mode Read the Terminals guide- Split by mouseSplit right or down, drag a line to resize, double click it to even out, and zoom any pane. Nothing to memorise.
- Drag chats inDrop a chat from the sidebar on a pane's edge to split it, or attach one that is working right now.
- New chat in any panePick the machine, folder, agent, model, and effort, then type.
/clearstarts over with the same setup. - Approvals in placeA chat that needs you shows its choices numbered in the pane. A window with one waiting gets
!on its tab. - Windows, tmux styleEach tab is a window with its own panes, marked
*-#!as in tmux. Only the one on screen is live. - Every pane at a glanceProject, branch, and machine; model, effort, context, cost, and tokens; then idle, working, or needs you.
- Claude, Codex, and GrokEach pane picks its own agent, and all three share one layout.
- Make it yoursSix themes, four fonts, text from 12 to 24 px, and your choice of borders, titles, and clock.
Features
Everything a run needs from you, in one place.
Approvals, diffs, usage, and every machine's sessions, from your desk or your phone.
Step away from the desk.
Install Termdeck on your phone to follow a run, send a prompt or a screenshot, and get a push notification when an agent needs you.
Termdeck on your phoneEvery machine on one board.
See what's working, what's finished, and which sessions are waiting on you, across every connected machine.
See the fleet boardReview changes and track usage.
Read the diff beside the conversation. See token usage and estimated API-equivalent cost by project, model, and machine.
Explore usage trackingSwitch accounts before a limit stops you.
Connect more than one Claude, OpenAI, or xAI account, watch their rate limits live, and switch in one click.
Multiple accountsYour sessions stay in your terminal.
Termdeck reads and writes the CLIs' own session files. Pick a browser chat back up with claude --resume or codex resume, and terminal sessions show up in the browser.
And the rest of the toolkit.
Tour the interface- Queue follow-upsLine up the next prompts and leave. They send in order as each turn finishes.
- CheckpointsRestore files to before any prompt, or branch the chat from there.
- Search every sessionFind any chat by prompt or project, across every machine.
- Background shellsLong commands keep running in their own panel while you keep talking.
- Sub-agentsWatch each sub-agent's work in its own panel beside the chat.
- Auto-approve rulesStop answering the same prompt. Allow what you trust, keep asking for the rest.
- Voice promptsSay the prompt instead of typing it, handy on a phone.
- Your MCP serversThe MCP servers configured on your machine work as they do in the terminal.
How it works
Your code stays on your machine.
Termdeck connects your browser to the CLIs running where your code already lives. No repository migration. No new development environment.
Read the quickstart-
Connect a machine
Create an account and run the install command provided for your machine.
-
Sign in to your CLI
Use the Claude Code, Codex, or Grok login you already have on that machine.
-
Open your browser
Start a session, send a prompt, and follow the work. Your native session history stays on disk.
The agent on your machine connects out to Termdeck. Your CLIs, repositories, credentials, and session files stay where they are.Security
Under the hood
Follow one task,
hop by hop.
Every line below is a connection Termdeck actually makes. Watch where your prompt goes, where the code runs, what reaches your phone, and what the cloud keeps: no transcripts, no code, no files.
Read how it works
Replay on
YOUR COMPUTERSCode runs here
Your AI providerThe CLI signs in with your own account
$ termdeck-agent
idle, connected
~/.claude/projects/0$ termdeck-agent
idle, connected
~/.codex/sessions/0> termdeck-agent
idle, connected
%USERPROFILE%\.grok\sessions\0Inbound
request
request
Your firewall
0 inbound ports opened
TERMDECK.IORelays, never stores
termdeck.ioRelay, scoped to your account
IN MEMORY, PASSING THROUGH
nothing in flight
- Machines online
- 0
- Frames relayed
- 0
- Transcripts stored
- 0 B
- Code or files stored
- 0 B
Keeps only what runs the service: your account, machine list (tokens as hashes), chat titles, push subscriptions.
Browser push serviceApple, Google or Mozilla delivers it
WHEREVER YOU AREYou decide here
Fix the login formno machine yet
Add validation and run the tests.
Permission to run a command
npm test
Waiting for you
9:41
Termdeck nowNeeds approval
Bash · npm test
DenyAllow01 / AGENTS DIAL OUT
Your computers call us. We never call them.
The Termdeck agent on each Mac, Linux box or Windows PC opens one outbound secure WebSocket to termdeck.io and proves itself with a machine token. The server keeps only a hash of that token. Nothing listens for inbound connections, so there is no port to open, no VPN and nothing for a scanner to find.
- Outbound only
- TLS encrypted
- Token stored as a hash
- Revoke a machine any time
02 / YOU SIGN IN
Your browser sees your machines, and only yours.
Open termdeck.io and sign in. Over HTTPS the page asks for your machines and chats, and every request and live update is scoped to your account and to one named machine. Another account's machine is never in the list and never in the stream.
- Sign-in required
- Scoped to your account
- Routed per machine
03 / YOUR PROMPT TRAVELS
Straight through, in memory, to one machine.
Pick a computer, a project and an engine, then send. The prompt crosses termdeck.io in server memory and goes down that machine's own connection. It is not written to our disk on the way. Watch the meters: frames go up, stored transcripts stay at zero.
- Passes through memory
- 0 B transcripts stored
- One machine, one route
04 / THE WORK HAPPENS AT HOME
Your CLI, your login, your files.
Claude Code, Codex or Grok starts on that computer as your user, with your tools, MCP servers and configuration. It talks to its AI provider directly with the subscription you already have; Termdeck never holds that login. Every step is written to the CLI's own session file on your disk, line by line.
- Runs as your user
- Your subscription, direct
- Session file on your disk
05 / IT APPEARS IN YOUR BROWSER
A live view of the real session.
Words, tool calls and edits stream back up the same connection and render in your browser as they happen. History comes from the session file on your machine; the stream only adds what is still arriving. Open a file or a diff and it is fetched from your machine on request.
- Rendered from your disk
- Files only when you open them
- Encrypted both ways
06 / IT ASKS BEFORE ACTING
The command waits until you say so.
In a mode that asks for approval, anything your rules do not already allow stops on your machine and waits. The browser shows exactly which tool and command it wants to run. Allow it once, deny it, or add a rule so you are not asked again. Full access modes do not stop at every step, so choose them on purpose.
- Permission modes
- Your allow rules
- Nothing runs while it waits
07 / YOUR PHONE BUZZES
Walk away. It will find you.
Turn on push in Settings, grant permission on each device you want, and pick which alerts you care about: approvals, questions, finished turns, errors. The alert goes through your browser vendor's push service to your lock screen. On iPhone and iPad, add Termdeck to your Home Screen first.
- Opt in per device
- You pick the alert types
- Delivered by your browser's push service
08 / ANSWER FROM ANYWHERE
One tap, one answer, one use.
Tap Allow or Deny on the notification, or open the chat on any signed-in device. A notification's buttons carry a single use token that answers exactly one pending request and expires after 5 minutes, the same moment an unanswered request is denied anyway. Your answer goes back down to the machine and the run continues.
- Single use token
- Expires in 5 minutes
- Unanswered means denied
09 / SAVED ON YOUR DISK, NOT OURS
Close the tab. The history stays home.
When the turn ends, the conversation already lives in the CLI's own session file on your computer. Close the tab, come back from another device, and Termdeck reads it from your machine again. Or skip us and resume in your terminal. If a chat is open in another process, Termdeck shows it view-only so two writers never fork it.
- Disk is the source of truth
- Resume in your terminal
- View-only when in use elsewhere
Pricing
Two plans. 14 days free on either.
Pick Starter or Pro and enter a card to start a 14-day free trial. Nothing is charged until day 15, and if you cancel before then you are never charged. Bring your own Claude, OpenAI Codex, and xAI (SuperGrok) subscriptions. Termdeck is the control plane.
Starter
One machine, a couple of projects. Auto-approve, queue work, and let runs finish while you're somewhere else.
$9.99
per month · 14-day free trial
- Auto-approve rules, so you stop answering the same prompt
- Queue follow-up turns and walk away
- Unlimited browser turns
- Checkpoints, MCP, voice, search, and the last 7 days of spend
- 1 connected machine, unlimited sessions at once
- 2 projects, 2 saved logins per engine, a 2x2 terminal grid
Most popular
Pro
Run a fleet. Every machine and every project on one board, all of it unattended.
$19.99
per month · 14-day free trial
- Everything in Starter, with none of its limits
- 10 connected machines, unlimited sessions at once
- Unlimited projects, and as many saved logins as you have, so a rate limit never stops a run
- Your whole spend history, and up to 9 terminals in a window
- Early access to new engines and features
Cancel any time from your account page. A cancelled plan keeps your machines, chats and rules; nothing runs until you subscribe again. Have a creator code? Redeem it on your account page. Model tokens and API usage bill to Anthropic, OpenAI, or xAI, not to Termdeck. Need more than 10 machines? Email [email protected] for custom fleet pricing.
Is Termdeck a remote control for Claude Code?
Yes. Termdeck is a remote control for Claude Code, Codex, and Grok that runs in any browser and installs on your phone. Unlike Claude Code Remote Control, it covers every session on every machine you connect, not one session at a time.
Can I remote control Codex CLI and Grok Build?
Yes. Termdeck drives Codex CLI and Grok Build on your own machines, on Linux, Windows, or macOS. Watch the transcript, approve tool calls, send follow-ups, and stop runs from anywhere.
Do I need a VPN, SSH, or port forwarding?
No. A small agent on each machine connects out to Termdeck, so there is nothing to open on your router and no tunnel to keep alive.
Is there a web UI for Claude Code?
Yes. Termdeck is a browser UI for the Claude Code CLI running on your machine. It streams the local transcript and adds tool approvals, diffs, project usage, session search, and phone access without moving the agent into a separate runtime. Read the Claude Code web UI guide →
Does OpenAI Codex CLI have a web interface?
Termdeck provides one by reading Codex CLI rollout files from disk and rendering the transcript, tool calls, and changed files in the browser, beside your Claude Code sessions. Read the Codex web UI guide →
Does Grok CLI have a web UI?
Yes. Termdeck runs xAI's Grok CLI as a third engine over the Agent Client Protocol. Start a Grok Build session, stream the transcript, approve tool calls, and switch model or reasoning effort mid-run. Grok sessions use grok's own session files, so a terminal and the browser can share one session. Read the Grok CLI guide →
How is this different from Claude Code on the web?
Claude Code on the web runs tasks on Anthropic-managed cloud infrastructure. Termdeck drives Claude Code on machines you own, so it can use your local repository, services, environment variables, credentials, and MCP servers. Compare Claude Code on the web vs Termdeck →
Can I use more than one Claude or OpenAI account?
Yes. Connect multiple accounts per engine, see each account's live rate limits side by side, and switch in one click when one hits a limit. The session keeps going under the account you switched to.
Can I run Claude Code, Codex, or Grok from my phone?
Yes. Install Termdeck as a PWA to follow Claude Code, Codex, and Grok sessions, send prompts and screenshots, and approve or deny tool calls. Push notifications surface waiting approvals when you are away from the computer.
Is Termdeck self-hosted or cloud-hosted?
Termdeck is cloud-hosted. A small agent on each of your machines connects out to it. Your coding CLIs, repositories, credentials, and native session files stay on your machines; the browser connects through Termdeck.
Will the sessions still resume in my terminal?
Yes. Termdeck reads and writes the same session files as the CLIs, so you can continue with claude --resume or codex resume. Terminal-started sessions also appear in the browser.
Can one dashboard manage multiple machines?
Yes. Connect macOS, Linux, or Windows machines and browse their active and historical sessions from one fleet board. Useful for a laptop, desktop, VM, home server, or remote development box.
Does Termdeck replace Claude Code, Codex, or Grok?
No. It is the UI and control layer for the official CLIs: Claude Code, Codex, and Grok. You keep your existing Anthropic, OpenAI, and xAI access, local tools, configuration, and model choices.
Open a browser.
Get back to your work.
Connect your first machine and pick up a session.
Claude Code, Codex, and Grok are all welcome.