Voice Invocations lets Meta AI launch or foreground an app when the wearer says “Hey Meta, start {app name}.” It is a Wearables-level stream, not a DeviceSession capability, and does not require a running device session.
Configure the app
Register the bundle identifier, request Voice Invocation approval, and configure the spoken app name in Wearables Developer Center. Do not add invocation phrases to Info.plist. Voice Invocations does not require camera or microphone permission.
Create and start the stream
VoiceInvocationsStream is @MainActor. Retain the stream and both listener tokens for as long as the app should receive actions. Register listeners before starting.
let newStream = try VoiceInvocationsStream(wearables: Wearables.shared)
stream = newStream
invocationToken = newStream.invocationsPublisher.listen { invocation in
guard let launch = invocation as? LaunchApp else { return }
navigateToMainScreen()
Task {
_ = await launch.responseHandle.sendSuccess(actionOutput: nil)
}
}
errorToken = newStream.errorPublisher.listen { error in
showError(error.localizedDescription)
}
try newStream.start(deviceIdentifier: deviceIdentifier)
Store stream, invocationToken, and errorToken on the @MainActor owner. Every delivered invocation must be acknowledged exactly once with sendSuccess or sendFailure. The Boolean result indicates whether the response was delivered.
The initializer can throw invalidWearablesInterface. Starting can throw deviceNotFound or channelNotConnected; later communication failures arrive through errorPublisher. Starting again switches the stream to the new device.
Stop and test
stop() is safe when inactive, and the same stream can start again later. Stop the stream and release its listener tokens when the owning app component is done.
To test without hardware, check glasses.services.voiceInvocation.hasConnectedClients before calling sendLaunchAppAction(). Use sendIncompleteAction() to confirm malformed actions produce an error rather than an invocation.
Availability
Voice Invocations is experimental. Apps can use it for development and beta testing, but cannot publish it to production release channels yet.
See the Voice Invocations guide.