Keep your words moving.

You need an Apple Watch running watchOS 10 or later, the companion Mac receiver, and Codex on your Mac. The support target includes Series 6 and subsequent existing Series models, plus compatible SE and Ultra models. Live dictation requires a compatible Mac, macOS 26 or later, and the corresponding Apple on-device speech recognition model; not every Mac that runs macOS 26 necessarily supports that model. Prepare the speech model in the Mac receiver before first use. If a download is needed, connect to the internet and wait for it to finish. For live dictation, both devices must be reachable on a local network, and the Mac must stay awake with the receiver running.

Use the same Apple Account with available iCloud service on both devices. Open the Mac receiver, then select that Mac on your Watch. Compare the six-digit numbers shown on both devices and approve the pairing on the Mac. Pairing is coordinated through that account’s CloudKit private database. Live dictation also needs a local-network connection and an available recognition model on the Mac. You do not need to share your Apple Account password or pairing credentials with the developer.

The upcoming 1.0 release supports English (United States) and Chinese (mainland China). Recognition follows the ToFlux app language on your Mac: Chinese uses the Chinese model; English or the default interface uses the English model. To change it, choose a language for ToFlux in macOS System Settings → General → Language & Region → Applications, then restart ToFlux. The Watch interface follows its own language setting and does not change the Mac recognition language. Automatic mixed-language recognition is not provided.

After connection and microphone permission are set up, opening the Watch app starts listening. Tap Pause to stop listening and the microphone to resume. Clear, Insert Only, and Send & Run also pause listening.

Insert Only (shown as Insert on smaller Watch screens) puts the text in the Codex input field so you can review or edit it. Send & Run requests text insertion followed by Return. Check the actual state in Codex to see whether a task started or finished. If delivery is awaiting confirmation, check your Mac before trying again to avoid duplicate submissions.

Make sure your Mac is awake, the receiver is running, the recognition model is available, and the devices can reach each other. Even networks with the same Wi-Fi name can block device-to-device access through guest-network or isolation settings. Pause, restore the connection, then tap the microphone to try again.

On your Mac, check ToFlux under System Settings → Privacy & Security → Accessibility. Open the correct Codex conversation, close any dialog covering the input field, and test with Insert Only.

Live dictation currently needs the local network to carry audio to your Mac. When configured and available, the iCloud channel can carry confirmed text and receipts. It cannot replace live audio transport or speech recognition.

No. You can clear the Watch draft when no message is awaiting confirmation. This does not delete content already passed to CloudKit or Codex, or all connection information. Check your Mac first if a result is still unconfirmed. There is currently no single delete-everything control. For version-specific cleanup guidance, contact the developer using the public support email.

Download ToFlux-Mac-1.0-33.zip from the Get ToFlux section of this website. Unzip it, drag ToFlux.app into Applications, then open it. The package is Apple Developer ID signed and notarized by Apple. Confirm the normal macOS first-open prompt and prepare the speech model in ToFlux; keep an internet connection if a download is needed. Then pair your Watch and allow ToFlux in System Settings → Privacy & Security → Accessibility. If macOS reports a damaged app or a signature problem, stop and contact support. Do not disable system security protections.