Talk to Claude Code and Codex agents from an Apple Watch and an iPhone. Each agent is a character on the watch.
import AVFoundation
/// Push-to-talk capture: 16 kHz mono AAC, the smallest file Whisper still reads well.
final class Recorder {
private var recorder: AVAudioRecorder?
/// Simulator captures (-fakeVoice YES): no microphone at all (the Simulator stalls seconds reopening it), just an
/// empty clip as long as the hold; the stand-in hub doesn't listen to it.
static var fake = false
private var fakeClip: (url: URL, start: Date)?
static func requestPermission() async -> Bool {
await withCheckedContinuation { cont in
AVAudioApplication.requestRecordPermission { cont.resume(returning: $0) }
}
}
func start() throws {
if Self.fake {
let url = FileManager.default.temporaryDirectory.appendingPathComponent("utterance-\(UUID().uuidString).m4a")
try Data().write(to: url); fakeClip = (url, Date()); return
}
let session = AVAudioSession.sharedInstance()
try session.setCategory(.playAndRecord, mode: .default, options: [])
try session.setActive(true)
let url = FileManager.default.temporaryDirectory.appendingPathComponent("utterance-\(UUID().uuidString).m4a")
let settings: [String: Any] = [
AVFormatIDKey: kAudioFormatMPEG4AAC,
AVSampleRateKey: 16_000,
AVNumberOfChannelsKey: 1,
AVEncoderAudioQualityKey: AVAudioQuality.medium.rawValue,
]
let r = try AVAudioRecorder(url: url, settings: settings)
r.isMeteringEnabled = true
guard r.record() else { throw NSError(domain: "Chums", code: 1) }
recorder = r
}
/// Voice level 0…1 from the input meter, for the character to pulse with.
var level: Double? {
guard let r = recorder, r.isRecording else { return nil }
r.updateMeters()
let db = Double(r.averagePower(forChannel: 0)) // −160 … 0 dBFS
return min(1, max(0, (db + 50) / 40))
}
/// Returns the file and how long it ran; nil if nothing was recording.
func stop() -> (url: URL, seconds: TimeInterval)? {
if let f = fakeClip { fakeClip = nil; return (f.url, Date().timeIntervalSince(f.start)) }
guard let r = recorder else { return nil }
let seconds = r.currentTime
r.stop()
recorder = nil
try? AVAudioSession.sharedInstance().setActive(false, options: .notifyOthersOnDeactivation) // the mic is off between holds
return (r.url, seconds)
}
}
/// Speaks replies through the watch speaker (or AirPods when connected).
final class Speaker: NSObject, AVSpeechSynthesizerDelegate, AVAudioPlayerDelegate {
private let synth = AVSpeechSynthesizer()
private var player: AVAudioPlayer?
var onFinish: (() -> Void)?
override init() { super.init(); synth.delegate = self }
func speechSynthesizer(_ s: AVSpeechSynthesizer, didFinish u: AVSpeechUtterance) { DispatchQueue.main.async { self.onFinish?() } }
func audioPlayerDidFinishPlaying(_ p: AVAudioPlayer, successfully: Bool) { DispatchQueue.main.async { self.onFinish?() } }
/// Plays a reply's voice rendered on the Mac; false if it can't, so the caller falls back to `speak`.
func play(_ data: Data) -> Bool {
let session = AVAudioSession.sharedInstance()
try? session.setCategory(.playback, mode: .spokenAudio, options: [])
try? session.setActive(true)
synth.stopSpeaking(at: .immediate); player?.stop()
guard let p = try? AVAudioPlayer(data: data) else { return false }
p.delegate = self; player = p
return p.play()
}
func speak(_ text: String, voice id: String? = nil) {
let session = AVAudioSession.sharedInstance()
try? session.setCategory(.playback, mode: .spokenAudio, options: [])
try? session.setActive(true)
let u = AVSpeechUtterance(string: text)
u.voice = Self.voice(id ?? UserDefaults.standard.string(forKey: "watchVoice"))
u.rate = 0.52
synth.stopSpeaking(at: .immediate); player?.stop()
synth.speak(u)
}
/// The watch voice chosen in Settings (Watch voice), else the system's US English one.
static func voice(_ id: String?) -> AVSpeechSynthesisVoice? {
id.flatMap { AVSpeechSynthesisVoice(identifier: $0) } ?? AVSpeechSynthesisVoice(language: "en-US")
}
/// The English voices installed on the watch, best first (premium, enhanced, then default), novelty voices left out.
static var englishVoices: [AVSpeechSynthesisVoice] {
AVSpeechSynthesisVoice.speechVoices()
.filter { $0.language.hasPrefix("en") && !$0.voiceTraits.contains(.isNoveltyVoice) }
.sorted { ($0.quality.rawValue, $1.name) > ($1.quality.rawValue, $0.name) }
}
static func qualityName(_ v: AVSpeechSynthesisVoice) -> String {
switch v.quality { case .premium: return "Premium"; case .enhanced: return "Enhanced"; default: return "Default" }
}
func stop() { synth.stopSpeaking(at: .immediate); player?.stop(); player = nil }
}