The short version
Your Apple Watch can approve Claude Code and Codex permission prompts, answer agent questions by voice, interrupt a runaway task, and even start a new coding task by dictation — all without pulling out your phone. BotBus on the watch syncs pairing from your iPhone and then connects independently over cellular or Wi-Fi, so it works while you're running, in a meeting, or on the train. The watch screen is small and not the right place to read long diffs, so this article covers both what it does well and where you should wait for a bigger screen.
Why put an agent on your wrist
AI coding agents pause a lot. Claude Code stops for permission before running shell commands. Codex waits for you to review results before continuing. At your desk, that's fine. Away from it, a single approval request can stall a task chain for half an hour.
Approving from your phone already solves most of those stalls — we covered three approaches in a previous post. But your phone isn't always handy. You might be running, driving, or doing the dishes. Unlocking your phone, finding the app, and tapping takes several steps. On your watch it takes one: raise your wrist, see the request, tap. For a binary allow-or-deny decision, the watch is actually the most natural device.
What you can do on the watch
See all your machines and tasks
Open BotBus on your watch and the first screen lists every computer you've paired, along with its online status. Tap into a machine to see its recent projects and running tasks. Tasks waiting for your approval are marked so they stand out. You don't need to hunt for them — when something needs you, a push notification pops up on your wrist.
Approve permission requests
This is the most common watch interaction. When Claude Code or Codex needs to run a command or edit a file, an approval card appears on your watch showing what the agent wants to do. Two large buttons — Allow and Deny — fill the screen, making accidental taps unlikely. If the agent offers scope choices, like "allow this once" or "allow this tool for the rest of the session," those appear above the buttons so you can pick a scope before approving.
Answer agent questions
Sometimes an agent asks you to choose, for example "TypeScript or JavaScript?" Multiple-choice questions show up as buttons on the watch, and you answer with a tap. For free-text responses, BotBus opens the watchOS system dictation — speak your answer and it goes straight to the agent.
Start a new task by voice
You can create a coding task right from the watch. The flow has three steps: pick which agent on which computer, pick a recent project, then dictate what you want done. Something like "replace the loading spinner with a skeleton screen on the homepage." Confirm, and it's sent. The agent starts working on your computer, and progress and approvals keep coming to your watch.
Interrupt a task
If a task goes off the rails and you want it to stop now, you can interrupt it from the task detail screen without going back to your computer. This requires that machine to be online and the connector to support the interrupt action.
Send a follow-up by voice
After a task finishes or when you want to add context mid-run, you can dictate a follow-up from the watch. The transcribed text is appended to the input field. Confirm to send, or swipe left to clear and try again. The agent picks up where it left off, just as if you had typed the follow-up at your keyboard.
What the watch can't do
The watch is good for one-step decisions. It's not the place for tasks that require careful reading:
- Videos, web previews, and file downloads. Videos, dev server previews, and most file types show "Open on iPhone" on the watch. Images can be viewed on the watch, but a 49 mm screen isn't where you want to review a screenshot in detail.
- Adding or removing computers. Pairing a new computer requires scanning a QR code on its screen, and the watch has no camera. Unpairing also needs to be done on the iPhone.
- Reading long code. Agent replies render with Markdown formatting, including code blocks, but a watch screen is not the right place to review a 200-line diff. Approval commands are usually short enough, but if you need to read through a large change before deciding, go back to a bigger screen.
How the watch connects
The watch doesn't pair with your computer directly. When you scan a pairing QR code on your iPhone, the credentials are pushed to the watch through watchOS's WatchConnectivity and stored in the watch's own Keychain. This happens locally between your phone and watch, not over the network. Once synced, the watch has the decryption key and connects to the relay server on its own over cellular or Wi-Fi. It doesn't need the iPhone nearby.
That means you can go for a run with just your watch and AirPods, and as long as the watch has a cellular signal, approval notifications will reach you.
Safety details
- Content is end-to-end encrypted. Like on the phone, tasks, conversations, commands, and notification text are encrypted and decrypted on your devices. The relay server only handles ciphertext. It can see what it needs to route data: device IDs, online status, update times, and data sizes. Full details are in how we encrypt.
- Small screens make careful reading harder. Check the full command before you approve. For hard-to-undo actions like
rm -rf,git push --force, or database migrations, deny and wait until you're at a larger screen if you have any doubt about the details. - A timeout is never an approval. If you don't respond, the pending request times out after about 30 minutes, and the decision falls back to Claude Code's own permission prompt on your computer. No timeout or disconnection ever counts as a yes.
- Your computer has to be online. Browsing already-synced tasks works offline, but approving, following up, starting tasks, and interrupting all require that computer to be online.
When do notifications arrive
Watch notifications are system-level, so they appear even when you're using another app. Approval notifications are marked as time-sensitive and can break through Focus modes. The notification text is a fixed prompt like "You have a pending request" — the actual command details are only visible once you open BotBus, where they're delivered through the end-to-end encrypted channel. This way, even though the notification passes through Apple's push servers, your code and commands stay private.
Can I use it today
On the computer side there's a Mac app (a menu bar app, macOS 26 or later) and a Linux version (a command-line program for x86_64 and aarch64; see the install guide), both available from our homepage. The iPhone and Apple Watch app is available on the App Store. The watch requires watchOS 26 or later. Windows and Android versions are planned.
FAQ
Does the watch need to be the cellular model?
No. Any Apple Watch that can connect to Wi-Fi works. The cellular model's advantage is that it receives notifications and can approve even when you leave your iPhone behind. A GPS-only watch works when your iPhone is nearby or when connected to a known Wi-Fi network.
Is approving on the watch different from approving on the phone?
Functionally identical: allow, deny, pick a scope. The difference is in the interaction — the watch is a wrist tap, the phone gives more room for context. If both devices show the same pending request, whichever you tap first wins.
I started a task from my watch. Can I follow up from my phone?
Yes. Tasks aren't tied to the device that created them. A task started from the watch shows up on the phone and the watch, and you can approve or follow up from either one.
I have several computers. Will the watch show all of them?
Yes. Every computer paired on your iPhone syncs to the watch, listed by machine with online status, projects, and tasks at a glance.
How accurate is the voice dictation?
BotBus uses the built-in watchOS dictation engine, so accuracy is the same as in any other app. For code-specific terms like variable names or commands, it helps to review the transcribed text on the watch before sending.