Skip to content
SDK Docs
Discord Order Now

On the phone

Voice commands and dictation

Let people control your app by voice and dictate text into it.

Your phone companion can let people talk to your app: short commands for what’s on screen, and dictation for free-form text. Both run while your app captures the glasses microphone, so call startCapture() first (see Microphone audio). They don’t need a MIRA listening session.

Voice commands

Register the commands your visible screen offers with setVoiceCommands(labels). When the wearer says one, your companion receives {type: "voiceCommand", command} through onFromApp:

Voice commands
const stop = miraCompanion.onFromApp((message) => {
  if (message.type !== "voiceCommand") return;
  if (message.command === "New note") startNote();
  if (message.command === "Save") saveNote();
});

await miraCompanion.setVoiceCommands(["New note", "Save"]);
await miraCompanion.startCapture();
  • Register up to 16 labels, and update them whenever the visible screen changes. Don’t register commands for screens that aren’t showing.
  • Use named actions. Left, Right, Up, Down, Select, Approve and Cancel aren’t available to apps. Taps, swipes and the ring still navigate as usual.
  • Pass a second argument naming the screen or record the commands belong to, so a command recognized just before the wearer moves on can’t act on the wrong one.
  • For labels you make up as you go, like a saved note’s title, pass a third argument that describes what each one does:
Labels with descriptions
await miraCompanion.setVoiceCommands(
  ["Open Groceries"],
  "notes-page-1",
  { "Open Groceries": "Open the saved note titled Groceries." }
);

Commands reach your app only while it’s in front, and they pause while MIRA shows its own overlays. MIRA’s own navigation commands keep working, and a pending approval comes first. MIRA clears your commands when the app closes.

When a command picks something on screen, move your highlight to it before you act, the same way a tap would, and don’t run the action a second time when the focus updates.

Dictation

For free-form text, start capture, then call startDictation(). Text arrives through onFromApp as {type: "dictation", text, confirmedText, final, status}, where confirmedText is the start of the text that won’t change anymore.

Dictate a title
const stop = miraCompanion.onFromApp((message) => {
  if (message.type === "dictation" && message.captureID === "title") {
    titleField.value = message.text;
  }
});

await miraCompanion.startCapture();
await miraCompanion.startDictation({ captureID: "title" });

// When the wearer is done:
const { text } = await miraCompanion.stopDictation();
OptionDoes
captureIDComes back on every dictation event, so you can ignore late text after switching fields.
includeRecentAudio: trueStarts with the last two seconds of audio, which can include the command that opened the field. Without it, dictation starts with fresh audio.
endpointDetection: falseKeeps transcribing across pauses. By default, MIRA detects where speech ends.
  • startDictation() fails unless your app’s capture is running. A MIRA listening session doesn’t count.
  • stopDictation() resolves with {text} and keeps capture running, so your voice commands keep working. They also work during dictation.
  • stopCapture() ends both capture and dictation. Closing the app or losing capture also ends dictation.