@athenaintel/react 0.48.0 → 0.50.0

package/dist/agent-runtime/src/use-thread-handoff-flow.d.ts

@@ -43,8 +43,14 @@ export interface StartThreadHandoffInput<D extends ThreadHandoffDestination = Se
     destination: D;
     officeLaunchMode?: OfficeLaunchMode;
     expectedWorkspaceId?: string;
-    /** The composer's unsent text; the destination prefills it and never sends it. */
-    draft?: string;
+    /**
+     * What the source's composer holds, read at the click: the destination
+     * prefills it and never sends it. Pass it as read, blank included: a blank
+     * composer clears the draft of an open ticket the server reuses, and an
+     * Office pane never gets one ({@link handoffDraftToCarry}). `null` (or
+     * omitted) means the surface could not read a composer.
+     */
+    draft?: string | null;
     /** `voice` offers "Continue by voice" on the destination; omitted means text. */
     mode?: HandoffMode;
 }

package/dist/agent-runtime/src/voice-earcon.cjs

@@ -0,0 +1,139 @@
+"use strict";
+Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
+const VOICE_MOVE_EARCON_SAMPLE_RATE = 24e3;
+const VOICE_MOVE_EARCON_MS = 300;
+const VOICE_MOVE_EARCON_PEAK = 0.1;
+const NOTES = [
+  { hz: 587.33, startMs: 0, durationMs: 170 },
+  { hz: 880, startMs: 100, durationMs: 200 }
+];
+const ATTACK_MS = 10;
+const OVERTONE_GAIN = 0.18;
+let samplesCache = null;
+let dataUriCache = null;
+const voiceMoveEarconSamples = () => {
+  if (samplesCache) return samplesCache;
+  const rate = VOICE_MOVE_EARCON_SAMPLE_RATE;
+  const samples = new Float32Array(Math.round(VOICE_MOVE_EARCON_MS / 1e3 * rate));
+  for (const note of NOTES) {
+    const start = Math.round(note.startMs / 1e3 * rate);
+    const length = Math.round(note.durationMs / 1e3 * rate);
+    const attack = Math.round(ATTACK_MS / 1e3 * rate);
+    for (let i = 0; i < length && start + i < samples.length; i += 1) {
+      const t = i / rate;
+      const decay = (1 - i / length) ** 2;
+      const envelope = i < attack ? i / attack * decay : decay;
+      const phase = 2 * Math.PI * note.hz * t;
+      samples[start + i] += envelope * (Math.sin(phase) + OVERTONE_GAIN * Math.sin(2 * phase));
+    }
+  }
+  let peak = 0;
+  for (const sample of samples) peak = Math.max(peak, Math.abs(sample));
+  const scale = peak > 0 ? VOICE_MOVE_EARCON_PEAK / peak : 0;
+  for (let i = 0; i < samples.length; i += 1) samples[i] *= scale;
+  samplesCache = samples;
+  return samples;
+};
+const voiceMoveEarconWav = () => {
+  const samples = voiceMoveEarconSamples();
+  const dataBytes = samples.length * 2;
+  const bytes = new Uint8Array(44 + dataBytes);
+  const view = new DataView(bytes.buffer);
+  const ascii = (offset, text) => {
+    for (let i = 0; i < text.length; i += 1) view.setUint8(offset + i, text.charCodeAt(i));
+  };
+  ascii(0, "RIFF");
+  view.setUint32(4, 36 + dataBytes, true);
+  ascii(8, "WAVE");
+  ascii(12, "fmt ");
+  view.setUint32(16, 16, true);
+  view.setUint16(20, 1, true);
+  view.setUint16(22, 1, true);
+  view.setUint32(24, VOICE_MOVE_EARCON_SAMPLE_RATE, true);
+  view.setUint32(28, VOICE_MOVE_EARCON_SAMPLE_RATE * 2, true);
+  view.setUint16(32, 2, true);
+  view.setUint16(34, 16, true);
+  ascii(36, "data");
+  view.setUint32(40, dataBytes, true);
+  for (let i = 0; i < samples.length; i += 1) {
+    const clamped = Math.max(-1, Math.min(1, samples[i] ?? 0));
+    view.setInt16(44 + i * 2, Math.round(clamped * 32767), true);
+  }
+  return bytes;
+};
+const BASE64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+const base64 = (bytes) => {
+  let out = "";
+  for (let i = 0; i < bytes.length; i += 3) {
+    const a = bytes[i] ?? 0;
+    const b = bytes[i + 1];
+    const c = bytes[i + 2];
+    const triple = a << 16 | (b ?? 0) << 8 | (c ?? 0);
+    out += BASE64[triple >> 18 & 63];
+    out += BASE64[triple >> 12 & 63];
+    out += b === void 0 ? "=" : BASE64[triple >> 6 & 63];
+    out += c === void 0 ? "=" : BASE64[triple & 63];
+  }
+  return out;
+};
+const voiceMoveEarconDataUri = () => {
+  dataUriCache ?? (dataUriCache = `data:audio/wav;base64,${base64(voiceMoveEarconWav())}`);
+  return dataUriCache;
+};
+const earconGlobals = globalThis;
+const PLAYBACK_BOUND_MS = VOICE_MOVE_EARCON_MS + 400;
+const playVoiceMoveEarcon = () => new Promise((resolve) => {
+  var _a;
+  let finished = false;
+  let release = () => void 0;
+  const finish = () => {
+    if (finished) return;
+    finished = true;
+    clearTimeout(bound);
+    try {
+      release();
+    } catch {
+    }
+    resolve();
+  };
+  const bound = setTimeout(finish, PLAYBACK_BOUND_MS);
+  try {
+    const AudioContextCtor = earconGlobals.AudioContext ?? earconGlobals.webkitAudioContext;
+    if (AudioContextCtor) {
+      const context = new AudioContextCtor();
+      release = () => void context.close().catch(() => void 0);
+      const samples = voiceMoveEarconSamples();
+      const buffer = context.createBuffer(1, samples.length, VOICE_MOVE_EARCON_SAMPLE_RATE);
+      buffer.getChannelData(0).set(samples);
+      const source = context.createBufferSource();
+      source.buffer = buffer;
+      source.connect(context.destination);
+      source.onended = finish;
+      if (context.state === "suspended") void ((_a = context.resume) == null ? void 0 : _a.call(context).catch(() => void 0));
+      source.start();
+      return;
+    }
+    const AudioCtor = earconGlobals.Audio;
+    if (AudioCtor) {
+      const element = new AudioCtor(voiceMoveEarconDataUri());
+      element.volume = 1;
+      element.onended = finish;
+      release = () => {
+        element.onended = null;
+      };
+      void Promise.resolve(element.play()).catch(finish);
+      return;
+    }
+    finish();
+  } catch {
+    finish();
+  }
+});
+exports.VOICE_MOVE_EARCON_MS = VOICE_MOVE_EARCON_MS;
+exports.VOICE_MOVE_EARCON_PEAK = VOICE_MOVE_EARCON_PEAK;
+exports.VOICE_MOVE_EARCON_SAMPLE_RATE = VOICE_MOVE_EARCON_SAMPLE_RATE;
+exports.playVoiceMoveEarcon = playVoiceMoveEarcon;
+exports.voiceMoveEarconDataUri = voiceMoveEarconDataUri;
+exports.voiceMoveEarconSamples = voiceMoveEarconSamples;
+exports.voiceMoveEarconWav = voiceMoveEarconWav;
+//# sourceMappingURL=voice-earcon.cjs.map

package/dist/agent-runtime/src/voice-earcon.cjs.map

@@ -0,0 +1 @@
+{"version":3,"file":"voice-earcon.cjs","sources":["../../../../agent-runtime/src/voice-earcon.ts"],"sourcesContent":["/**\n * The earcon a call makes when it moves between a person's own devices: two\n * soft rising notes, played on the device that now holds the call and on the\n * device it left. One sound for every app, defined here once: the samples are\n * synthesized (no binary asset to bundle per host), web hosts play them\n * through WebAudio ({@link playVoiceMoveEarcon}), and native hosts play the\n * same sound as a WAV ({@link voiceMoveEarconDataUri}) through their own audio\n * session. Platform-free at import: nothing here touches a browser global\n * until a web host asks it to play.\n */\n\n/** Sample rate of the earcon (Hz); the realtime voice path's own rate. */\nexport const VOICE_MOVE_EARCON_SAMPLE_RATE = 24_000;\n\n/** How long the earcon lasts, tail included (ms). Short enough to never talk over a turn. */\nexport const VOICE_MOVE_EARCON_MS = 300;\n\n/**\n * The earcon's loudest sample (linear, full scale = 1): about -20 dBFS, well\n * under the agent's speech, so it marks the move without startling anyone.\n */\nexport const VOICE_MOVE_EARCON_PEAK = 0.1;\n\n/** The two notes: D5, then A5 over its tail (a fifth up, read as \"arrived\"). */\nconst NOTES = [\n  { hz: 587.33, startMs: 0, durationMs: 170 },\n  { hz: 880, startMs: 100, durationMs: 200 },\n] as const;\n/** A soft onset (no click) and a quiet octave for warmth. */\nconst ATTACK_MS = 10;\nconst OVERTONE_GAIN = 0.18;\n\nlet samplesCache: Float32Array | null = null;\nlet dataUriCache: string | null = null;\n\n/** The earcon as mono float samples at {@link VOICE_MOVE_EARCON_SAMPLE_RATE}. Do not mutate. */\nexport const voiceMoveEarconSamples = (): Float32Array => {\n  if (samplesCache) return samplesCache;\n  const rate = VOICE_MOVE_EARCON_SAMPLE_RATE;\n  const samples = new Float32Array(Math.round((VOICE_MOVE_EARCON_MS / 1000) * rate));\n  for (const note of NOTES) {\n    const start = Math.round((note.startMs / 1000) * rate);\n    const length = Math.round((note.durationMs / 1000) * rate);\n    const attack = Math.round((ATTACK_MS / 1000) * rate);\n    for (let i = 0; i < length && start + i < samples.length; i += 1) {\n      const t = i / rate;\n      // Linear attack, then a squared decay that lands on exactly zero.\n      const decay = (1 - i / length) ** 2;\n      const envelope = i < attack ? (i / attack) * decay : decay;\n      const phase = 2 * Math.PI * note.hz * t;\n      samples[start + i] += envelope * (Math.sin(phase) + OVERTONE_GAIN * Math.sin(2 * phase));\n    }\n  }\n  let peak = 0;\n  for (const sample of samples) peak = Math.max(peak, Math.abs(sample));\n  const scale = peak > 0 ? VOICE_MOVE_EARCON_PEAK / peak : 0;\n  for (let i = 0; i < samples.length; i += 1) samples[i] *= scale;\n  samplesCache = samples;\n  return samples;\n};\n\n/** The earcon as a 16-bit PCM mono WAV file. */\nexport const voiceMoveEarconWav = (): Uint8Array => {\n  const samples = voiceMoveEarconSamples();\n  const dataBytes = samples.length * 2;\n  const bytes = new Uint8Array(44 + dataBytes);\n  const view = new DataView(bytes.buffer);\n  const ascii = (offset: number, text: string): void => {\n    for (let i = 0; i < text.length; i += 1) view.setUint8(offset + i, text.charCodeAt(i));\n  };\n  ascii(0, 'RIFF');\n  view.setUint32(4, 36 + dataBytes, true);\n  ascii(8, 'WAVE');\n  ascii(12, 'fmt ');\n  view.setUint32(16, 16, true);\n  view.setUint16(20, 1, true); // PCM\n  view.setUint16(22, 1, true); // mono\n  view.setUint32(24, VOICE_MOVE_EARCON_SAMPLE_RATE, true);\n  view.setUint32(28, VOICE_MOVE_EARCON_SAMPLE_RATE * 2, true);\n  view.setUint16(32, 2, true);\n  view.setUint16(34, 16, true);\n  ascii(36, 'data');\n  view.setUint32(40, dataBytes, true);\n  for (let i = 0; i < samples.length; i += 1) {\n    const clamped = Math.max(-1, Math.min(1, samples[i] ?? 0));\n    view.setInt16(44 + i * 2, Math.round(clamped * 0x7fff), true);\n  }\n  return bytes;\n};\n\nconst BASE64 = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/';\n\n/** Standard base64, with no dependency on `btoa` or `Buffer` (React Native has neither for sure). */\nconst base64 = (bytes: Uint8Array): string => {\n  let out = '';\n  for (let i = 0; i < bytes.length; i += 3) {\n    const a = bytes[i] ?? 0;\n    const b = bytes[i + 1];\n    const c = bytes[i + 2];\n    const triple = (a << 16) | ((b ?? 0) << 8) | (c ?? 0);\n    out += BASE64[(triple >> 18) & 63];\n    out += BASE64[(triple >> 12) & 63];\n    out += b === undefined ? '=' : BASE64[(triple >> 6) & 63];\n    out += c === undefined ? '=' : BASE64[triple & 63];\n  }\n  return out;\n};\n\n/**\n * The earcon WAV as a `data:audio/wav;base64,…` URI, for a native player that\n * takes a source URI (expo-audio writes it to a temporary file on iOS and\n * reads it directly on Android and the web).\n */\nexport const voiceMoveEarconDataUri = (): string => {\n  dataUriCache ??= `data:audio/wav;base64,${base64(voiceMoveEarconWav())}`;\n  return dataUriCache;\n};\n\ninterface EarconAudioNode {\n  connect(destination: EarconAudioNode): unknown;\n}\ninterface EarconAudioBuffer {\n  getChannelData(channel: number): Float32Array;\n}\ninterface EarconBufferSource extends EarconAudioNode {\n  buffer: EarconAudioBuffer | null;\n  onended: (() => void) | null;\n  start(when?: number): void;\n}\ninterface EarconAudioContext {\n  readonly state: string;\n  readonly destination: EarconAudioNode;\n  createBuffer(channels: number, length: number, sampleRate: number): EarconAudioBuffer;\n  createBufferSource(): EarconBufferSource;\n  resume?(): Promise<void>;\n  close(): Promise<void>;\n}\ninterface EarconAudioElement {\n  volume: number;\n  onended: (() => void) | null;\n  play(): Promise<void> | void;\n}\n\nconst earconGlobals = globalThis as {\n  AudioContext?: new () => EarconAudioContext;\n  webkitAudioContext?: new () => EarconAudioContext;\n  Audio?: new (src?: string) => EarconAudioElement;\n};\n\n/** A cue that never ends (a blocked autoplay) still lets go this long after it started. */\nconst PLAYBACK_BOUND_MS = VOICE_MOVE_EARCON_MS + 400;\n\n/**\n * Play the earcon on a web host (Spaces, the Chrome side panel, the SDK the\n * desktop app renders): its own short-lived `AudioContext`, apart from the\n * call's media, or an `<audio>` element where WebAudio is missing. Resolves\n * once it has played, or within {@link PLAYBACK_BOUND_MS} when the page may\n * not play audio yet (autoplay policy). Never throws; a no-op where neither\n * exists (React Native, tests).\n */\nexport const playVoiceMoveEarcon = (): Promise<void> =>\n  new Promise<void>((resolve) => {\n    let finished = false;\n    let release: () => void = () => undefined;\n    const finish = (): void => {\n      if (finished) return;\n      finished = true;\n      clearTimeout(bound);\n      try {\n        release();\n      } catch {\n        // Already closed.\n      }\n      resolve();\n    };\n    const bound = setTimeout(finish, PLAYBACK_BOUND_MS);\n    try {\n      const AudioContextCtor = earconGlobals.AudioContext ?? earconGlobals.webkitAudioContext;\n      if (AudioContextCtor) {\n        const context = new AudioContextCtor();\n        release = () => void context.close().catch(() => undefined);\n        const samples = voiceMoveEarconSamples();\n        const buffer = context.createBuffer(1, samples.length, VOICE_MOVE_EARCON_SAMPLE_RATE);\n        buffer.getChannelData(0).set(samples);\n        const source = context.createBufferSource();\n        source.buffer = buffer;\n        source.connect(context.destination);\n        source.onended = finish;\n        if (context.state === 'suspended') void context.resume?.().catch(() => undefined);\n        source.start();\n        return;\n      }\n      const AudioCtor = earconGlobals.Audio;\n      if (AudioCtor) {\n        const element = new AudioCtor(voiceMoveEarconDataUri());\n        element.volume = 1;\n        element.onended = finish;\n        release = () => {\n          element.onended = null;\n        };\n        void Promise.resolve(element.play()).catch(finish);\n        return;\n      }\n      finish();\n    } catch {\n      finish();\n    }\n  });\n"],"names":[],"mappings":";;AAYO,MAAM,gCAAgC;AAGtC,MAAM,uBAAuB;AAM7B,MAAM,yBAAyB;AAGtC,MAAM,QAAQ;AAAA,EACZ,EAAE,IAAI,QAAQ,SAAS,GAAG,YAAY,IAAA;AAAA,EACtC,EAAE,IAAI,KAAK,SAAS,KAAK,YAAY,IAAA;AACvC;AAEA,MAAM,YAAY;AAClB,MAAM,gBAAgB;AAEtB,IAAI,eAAoC;AACxC,IAAI,eAA8B;AAG3B,MAAM,yBAAyB,MAAoB;AACxD,MAAI,aAAc,QAAO;AACzB,QAAM,OAAO;AACb,QAAM,UAAU,IAAI,aAAa,KAAK,MAAO,uBAAuB,MAAQ,IAAI,CAAC;AACjF,aAAW,QAAQ,OAAO;AACxB,UAAM,QAAQ,KAAK,MAAO,KAAK,UAAU,MAAQ,IAAI;AACrD,UAAM,SAAS,KAAK,MAAO,KAAK,aAAa,MAAQ,IAAI;AACzD,UAAM,SAAS,KAAK,MAAO,YAAY,MAAQ,IAAI;AACnD,aAAS,IAAI,GAAG,IAAI,UAAU,QAAQ,IAAI,QAAQ,QAAQ,KAAK,GAAG;AAChE,YAAM,IAAI,IAAI;AAEd,YAAM,SAAS,IAAI,IAAI,WAAW;AAClC,YAAM,WAAW,IAAI,SAAU,IAAI,SAAU,QAAQ;AACrD,YAAM,QAAQ,IAAI,KAAK,KAAK,KAAK,KAAK;AACtC,cAAQ,QAAQ,CAAC,KAAK,YAAY,KAAK,IAAI,KAAK,IAAI,gBAAgB,KAAK,IAAI,IAAI,KAAK;AAAA,IACxF;AAAA,EACF;AACA,MAAI,OAAO;AACX,aAAW,UAAU,QAAS,QAAO,KAAK,IAAI,MAAM,KAAK,IAAI,MAAM,CAAC;AACpE,QAAM,QAAQ,OAAO,IAAI,yBAAyB,OAAO;AACzD,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK,EAAG,SAAQ,CAAC,KAAK;AAC1D,iBAAe;AACf,SAAO;AACT;AAGO,MAAM,qBAAqB,MAAkB;AAClD,QAAM,UAAU,uBAAA;AAChB,QAAM,YAAY,QAAQ,SAAS;AACnC,QAAM,QAAQ,IAAI,WAAW,KAAK,SAAS;AAC3C,QAAM,OAAO,IAAI,SAAS,MAAM,MAAM;AACtC,QAAM,QAAQ,CAAC,QAAgB,SAAuB;AACpD,aAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,EAAG,MAAK,SAAS,SAAS,GAAG,KAAK,WAAW,CAAC,CAAC;AAAA,EACvF;AACA,QAAM,GAAG,MAAM;AACf,OAAK,UAAU,GAAG,KAAK,WAAW,IAAI;AACtC,QAAM,GAAG,MAAM;AACf,QAAM,IAAI,MAAM;AAChB,OAAK,UAAU,IAAI,IAAI,IAAI;AAC3B,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,+BAA+B,IAAI;AACtD,OAAK,UAAU,IAAI,gCAAgC,GAAG,IAAI;AAC1D,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,IAAI,IAAI;AAC3B,QAAM,IAAI,MAAM;AAChB,OAAK,UAAU,IAAI,WAAW,IAAI;AAClC,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK,GAAG;AAC1C,UAAM,UAAU,KAAK,IAAI,IAAI,KAAK,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;AACzD,SAAK,SAAS,KAAK,IAAI,GAAG,KAAK,MAAM,UAAU,KAAM,GAAG,IAAI;AAAA,EAC9D;AACA,SAAO;AACT;AAEA,MAAM,SAAS;AAGf,MAAM,SAAS,CAAC,UAA8B;AAC5C,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK,GAAG;AACxC,UAAM,IAAI,MAAM,CAAC,KAAK;AACtB,UAAM,IAAI,MAAM,IAAI,CAAC;AACrB,UAAM,IAAI,MAAM,IAAI,CAAC;AACrB,UAAM,SAAU,KAAK,MAAQ,KAAK,MAAM,KAAM,KAAK;AACnD,WAAO,OAAQ,UAAU,KAAM,EAAE;AACjC,WAAO,OAAQ,UAAU,KAAM,EAAE;AACjC,WAAO,MAAM,SAAY,MAAM,OAAQ,UAAU,IAAK,EAAE;AACxD,WAAO,MAAM,SAAY,MAAM,OAAO,SAAS,EAAE;AAAA,EACnD;AACA,SAAO;AACT;AAOO,MAAM,yBAAyB,MAAc;AAClD,kCAAiB,yBAAyB,OAAO,mBAAA,CAAoB,CAAC;AACtE,SAAO;AACT;AA2BA,MAAM,gBAAgB;AAOtB,MAAM,oBAAoB,uBAAuB;AAU1C,MAAM,sBAAsB,MACjC,IAAI,QAAc,CAAC,YAAY;;AAC7B,MAAI,WAAW;AACf,MAAI,UAAsB,MAAM;AAChC,QAAM,SAAS,MAAY;AACzB,QAAI,SAAU;AACd,eAAW;AACX,iBAAa,KAAK;AAClB,QAAI;AACF,cAAA;AAAA,IACF,QAAQ;AAAA,IAER;AACA,YAAA;AAAA,EACF;AACA,QAAM,QAAQ,WAAW,QAAQ,iBAAiB;AAClD,MAAI;AACF,UAAM,mBAAmB,cAAc,gBAAgB,cAAc;AACrE,QAAI,kBAAkB;AACpB,YAAM,UAAU,IAAI,iBAAA;AACpB,gBAAU,MAAM,KAAK,QAAQ,QAAQ,MAAM,MAAM,MAAS;AAC1D,YAAM,UAAU,uBAAA;AAChB,YAAM,SAAS,QAAQ,aAAa,GAAG,QAAQ,QAAQ,6BAA6B;AACpF,aAAO,eAAe,CAAC,EAAE,IAAI,OAAO;AACpC,YAAM,SAAS,QAAQ,mBAAA;AACvB,aAAO,SAAS;AAChB,aAAO,QAAQ,QAAQ,WAAW;AAClC,aAAO,UAAU;AACjB,UAAI,QAAQ,UAAU,YAAa,QAAK,aAAQ,WAAR,iCAAmB,MAAM,MAAM;AACvE,aAAO,MAAA;AACP;AAAA,IACF;AACA,UAAM,YAAY,cAAc;AAChC,QAAI,WAAW;AACb,YAAM,UAAU,IAAI,UAAU,wBAAwB;AACtD,cAAQ,SAAS;AACjB,cAAQ,UAAU;AAClB,gBAAU,MAAM;AACd,gBAAQ,UAAU;AAAA,MACpB;AACA,WAAK,QAAQ,QAAQ,QAAQ,MAAM,EAAE,MAAM,MAAM;AACjD;AAAA,IACF;AACA,WAAA;AAAA,EACF,QAAQ;AACN,WAAA;AAAA,EACF;AACF,CAAC;;;;;;;;"}

package/dist/agent-runtime/src/voice-earcon.d.ts

@@ -0,0 +1,38 @@
+/**
+ * The earcon a call makes when it moves between a person's own devices: two
+ * soft rising notes, played on the device that now holds the call and on the
+ * device it left. One sound for every app, defined here once: the samples are
+ * synthesized (no binary asset to bundle per host), web hosts play them
+ * through WebAudio ({@link playVoiceMoveEarcon}), and native hosts play the
+ * same sound as a WAV ({@link voiceMoveEarconDataUri}) through their own audio
+ * session. Platform-free at import: nothing here touches a browser global
+ * until a web host asks it to play.
+ */
+/** Sample rate of the earcon (Hz); the realtime voice path's own rate. */
+export declare const VOICE_MOVE_EARCON_SAMPLE_RATE = 24000;
+/** How long the earcon lasts, tail included (ms). Short enough to never talk over a turn. */
+export declare const VOICE_MOVE_EARCON_MS = 300;
+/**
+ * The earcon's loudest sample (linear, full scale = 1): about -20 dBFS, well
+ * under the agent's speech, so it marks the move without startling anyone.
+ */
+export declare const VOICE_MOVE_EARCON_PEAK = 0.1;
+/** The earcon as mono float samples at {@link VOICE_MOVE_EARCON_SAMPLE_RATE}. Do not mutate. */
+export declare const voiceMoveEarconSamples: () => Float32Array;
+/** The earcon as a 16-bit PCM mono WAV file. */
+export declare const voiceMoveEarconWav: () => Uint8Array;
+/**
+ * The earcon WAV as a `data:audio/wav;base64,…` URI, for a native player that
+ * takes a source URI (expo-audio writes it to a temporary file on iOS and
+ * reads it directly on Android and the web).
+ */
+export declare const voiceMoveEarconDataUri: () => string;
+/**
+ * Play the earcon on a web host (Spaces, the Chrome side panel, the SDK the
+ * desktop app renders): its own short-lived `AudioContext`, apart from the
+ * call's media, or an `<audio>` element where WebAudio is missing. Resolves
+ * once it has played, or within {@link PLAYBACK_BOUND_MS} when the page may
+ * not play audio yet (autoplay policy). Never throws; a no-op where neither
+ * exists (React Native, tests).
+ */
+export declare const playVoiceMoveEarcon: () => Promise<void>;

package/dist/agent-runtime/src/voice-earcon.js

@@ -0,0 +1,139 @@
+const VOICE_MOVE_EARCON_SAMPLE_RATE = 24e3;
+const VOICE_MOVE_EARCON_MS = 300;
+const VOICE_MOVE_EARCON_PEAK = 0.1;
+const NOTES = [
+  { hz: 587.33, startMs: 0, durationMs: 170 },
+  { hz: 880, startMs: 100, durationMs: 200 }
+];
+const ATTACK_MS = 10;
+const OVERTONE_GAIN = 0.18;
+let samplesCache = null;
+let dataUriCache = null;
+const voiceMoveEarconSamples = () => {
+  if (samplesCache) return samplesCache;
+  const rate = VOICE_MOVE_EARCON_SAMPLE_RATE;
+  const samples = new Float32Array(Math.round(VOICE_MOVE_EARCON_MS / 1e3 * rate));
+  for (const note of NOTES) {
+    const start = Math.round(note.startMs / 1e3 * rate);
+    const length = Math.round(note.durationMs / 1e3 * rate);
+    const attack = Math.round(ATTACK_MS / 1e3 * rate);
+    for (let i = 0; i < length && start + i < samples.length; i += 1) {
+      const t = i / rate;
+      const decay = (1 - i / length) ** 2;
+      const envelope = i < attack ? i / attack * decay : decay;
+      const phase = 2 * Math.PI * note.hz * t;
+      samples[start + i] += envelope * (Math.sin(phase) + OVERTONE_GAIN * Math.sin(2 * phase));
+    }
+  }
+  let peak = 0;
+  for (const sample of samples) peak = Math.max(peak, Math.abs(sample));
+  const scale = peak > 0 ? VOICE_MOVE_EARCON_PEAK / peak : 0;
+  for (let i = 0; i < samples.length; i += 1) samples[i] *= scale;
+  samplesCache = samples;
+  return samples;
+};
+const voiceMoveEarconWav = () => {
+  const samples = voiceMoveEarconSamples();
+  const dataBytes = samples.length * 2;
+  const bytes = new Uint8Array(44 + dataBytes);
+  const view = new DataView(bytes.buffer);
+  const ascii = (offset, text) => {
+    for (let i = 0; i < text.length; i += 1) view.setUint8(offset + i, text.charCodeAt(i));
+  };
+  ascii(0, "RIFF");
+  view.setUint32(4, 36 + dataBytes, true);
+  ascii(8, "WAVE");
+  ascii(12, "fmt ");
+  view.setUint32(16, 16, true);
+  view.setUint16(20, 1, true);
+  view.setUint16(22, 1, true);
+  view.setUint32(24, VOICE_MOVE_EARCON_SAMPLE_RATE, true);
+  view.setUint32(28, VOICE_MOVE_EARCON_SAMPLE_RATE * 2, true);
+  view.setUint16(32, 2, true);
+  view.setUint16(34, 16, true);
+  ascii(36, "data");
+  view.setUint32(40, dataBytes, true);
+  for (let i = 0; i < samples.length; i += 1) {
+    const clamped = Math.max(-1, Math.min(1, samples[i] ?? 0));
+    view.setInt16(44 + i * 2, Math.round(clamped * 32767), true);
+  }
+  return bytes;
+};
+const BASE64 = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+const base64 = (bytes) => {
+  let out = "";
+  for (let i = 0; i < bytes.length; i += 3) {
+    const a = bytes[i] ?? 0;
+    const b = bytes[i + 1];
+    const c = bytes[i + 2];
+    const triple = a << 16 | (b ?? 0) << 8 | (c ?? 0);
+    out += BASE64[triple >> 18 & 63];
+    out += BASE64[triple >> 12 & 63];
+    out += b === void 0 ? "=" : BASE64[triple >> 6 & 63];
+    out += c === void 0 ? "=" : BASE64[triple & 63];
+  }
+  return out;
+};
+const voiceMoveEarconDataUri = () => {
+  dataUriCache ?? (dataUriCache = `data:audio/wav;base64,${base64(voiceMoveEarconWav())}`);
+  return dataUriCache;
+};
+const earconGlobals = globalThis;
+const PLAYBACK_BOUND_MS = VOICE_MOVE_EARCON_MS + 400;
+const playVoiceMoveEarcon = () => new Promise((resolve) => {
+  var _a;
+  let finished = false;
+  let release = () => void 0;
+  const finish = () => {
+    if (finished) return;
+    finished = true;
+    clearTimeout(bound);
+    try {
+      release();
+    } catch {
+    }
+    resolve();
+  };
+  const bound = setTimeout(finish, PLAYBACK_BOUND_MS);
+  try {
+    const AudioContextCtor = earconGlobals.AudioContext ?? earconGlobals.webkitAudioContext;
+    if (AudioContextCtor) {
+      const context = new AudioContextCtor();
+      release = () => void context.close().catch(() => void 0);
+      const samples = voiceMoveEarconSamples();
+      const buffer = context.createBuffer(1, samples.length, VOICE_MOVE_EARCON_SAMPLE_RATE);
+      buffer.getChannelData(0).set(samples);
+      const source = context.createBufferSource();
+      source.buffer = buffer;
+      source.connect(context.destination);
+      source.onended = finish;
+      if (context.state === "suspended") void ((_a = context.resume) == null ? void 0 : _a.call(context).catch(() => void 0));
+      source.start();
+      return;
+    }
+    const AudioCtor = earconGlobals.Audio;
+    if (AudioCtor) {
+      const element = new AudioCtor(voiceMoveEarconDataUri());
+      element.volume = 1;
+      element.onended = finish;
+      release = () => {
+        element.onended = null;
+      };
+      void Promise.resolve(element.play()).catch(finish);
+      return;
+    }
+    finish();
+  } catch {
+    finish();
+  }
+});
+export {
+  VOICE_MOVE_EARCON_MS,
+  VOICE_MOVE_EARCON_PEAK,
+  VOICE_MOVE_EARCON_SAMPLE_RATE,
+  playVoiceMoveEarcon,
+  voiceMoveEarconDataUri,
+  voiceMoveEarconSamples,
+  voiceMoveEarconWav
+};
+//# sourceMappingURL=voice-earcon.js.map

package/dist/agent-runtime/src/voice-earcon.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"voice-earcon.js","sources":["../../../../agent-runtime/src/voice-earcon.ts"],"sourcesContent":["/**\n * The earcon a call makes when it moves between a person's own devices: two\n * soft rising notes, played on the device that now holds the call and on the\n * device it left. One sound for every app, defined here once: the samples are\n * synthesized (no binary asset to bundle per host), web hosts play them\n * through WebAudio ({@link playVoiceMoveEarcon}), and native hosts play the\n * same sound as a WAV ({@link voiceMoveEarconDataUri}) through their own audio\n * session. Platform-free at import: nothing here touches a browser global\n * until a web host asks it to play.\n */\n\n/** Sample rate of the earcon (Hz); the realtime voice path's own rate. */\nexport const VOICE_MOVE_EARCON_SAMPLE_RATE = 24_000;\n\n/** How long the earcon lasts, tail included (ms). Short enough to never talk over a turn. */\nexport const VOICE_MOVE_EARCON_MS = 300;\n\n/**\n * The earcon's loudest sample (linear, full scale = 1): about -20 dBFS, well\n * under the agent's speech, so it marks the move without startling anyone.\n */\nexport const VOICE_MOVE_EARCON_PEAK = 0.1;\n\n/** The two notes: D5, then A5 over its tail (a fifth up, read as \"arrived\"). */\nconst NOTES = [\n  { hz: 587.33, startMs: 0, durationMs: 170 },\n  { hz: 880, startMs: 100, durationMs: 200 },\n] as const;\n/** A soft onset (no click) and a quiet octave for warmth. */\nconst ATTACK_MS = 10;\nconst OVERTONE_GAIN = 0.18;\n\nlet samplesCache: Float32Array | null = null;\nlet dataUriCache: string | null = null;\n\n/** The earcon as mono float samples at {@link VOICE_MOVE_EARCON_SAMPLE_RATE}. Do not mutate. */\nexport const voiceMoveEarconSamples = (): Float32Array => {\n  if (samplesCache) return samplesCache;\n  const rate = VOICE_MOVE_EARCON_SAMPLE_RATE;\n  const samples = new Float32Array(Math.round((VOICE_MOVE_EARCON_MS / 1000) * rate));\n  for (const note of NOTES) {\n    const start = Math.round((note.startMs / 1000) * rate);\n    const length = Math.round((note.durationMs / 1000) * rate);\n    const attack = Math.round((ATTACK_MS / 1000) * rate);\n    for (let i = 0; i < length && start + i < samples.length; i += 1) {\n      const t = i / rate;\n      // Linear attack, then a squared decay that lands on exactly zero.\n      const decay = (1 - i / length) ** 2;\n      const envelope = i < attack ? (i / attack) * decay : decay;\n      const phase = 2 * Math.PI * note.hz * t;\n      samples[start + i] += envelope * (Math.sin(phase) + OVERTONE_GAIN * Math.sin(2 * phase));\n    }\n  }\n  let peak = 0;\n  for (const sample of samples) peak = Math.max(peak, Math.abs(sample));\n  const scale = peak > 0 ? VOICE_MOVE_EARCON_PEAK / peak : 0;\n  for (let i = 0; i < samples.length; i += 1) samples[i] *= scale;\n  samplesCache = samples;\n  return samples;\n};\n\n/** The earcon as a 16-bit PCM mono WAV file. */\nexport const voiceMoveEarconWav = (): Uint8Array => {\n  const samples = voiceMoveEarconSamples();\n  const dataBytes = samples.length * 2;\n  const bytes = new Uint8Array(44 + dataBytes);\n  const view = new DataView(bytes.buffer);\n  const ascii = (offset: number, text: string): void => {\n    for (let i = 0; i < text.length; i += 1) view.setUint8(offset + i, text.charCodeAt(i));\n  };\n  ascii(0, 'RIFF');\n  view.setUint32(4, 36 + dataBytes, true);\n  ascii(8, 'WAVE');\n  ascii(12, 'fmt ');\n  view.setUint32(16, 16, true);\n  view.setUint16(20, 1, true); // PCM\n  view.setUint16(22, 1, true); // mono\n  view.setUint32(24, VOICE_MOVE_EARCON_SAMPLE_RATE, true);\n  view.setUint32(28, VOICE_MOVE_EARCON_SAMPLE_RATE * 2, true);\n  view.setUint16(32, 2, true);\n  view.setUint16(34, 16, true);\n  ascii(36, 'data');\n  view.setUint32(40, dataBytes, true);\n  for (let i = 0; i < samples.length; i += 1) {\n    const clamped = Math.max(-1, Math.min(1, samples[i] ?? 0));\n    view.setInt16(44 + i * 2, Math.round(clamped * 0x7fff), true);\n  }\n  return bytes;\n};\n\nconst BASE64 = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/';\n\n/** Standard base64, with no dependency on `btoa` or `Buffer` (React Native has neither for sure). */\nconst base64 = (bytes: Uint8Array): string => {\n  let out = '';\n  for (let i = 0; i < bytes.length; i += 3) {\n    const a = bytes[i] ?? 0;\n    const b = bytes[i + 1];\n    const c = bytes[i + 2];\n    const triple = (a << 16) | ((b ?? 0) << 8) | (c ?? 0);\n    out += BASE64[(triple >> 18) & 63];\n    out += BASE64[(triple >> 12) & 63];\n    out += b === undefined ? '=' : BASE64[(triple >> 6) & 63];\n    out += c === undefined ? '=' : BASE64[triple & 63];\n  }\n  return out;\n};\n\n/**\n * The earcon WAV as a `data:audio/wav;base64,…` URI, for a native player that\n * takes a source URI (expo-audio writes it to a temporary file on iOS and\n * reads it directly on Android and the web).\n */\nexport const voiceMoveEarconDataUri = (): string => {\n  dataUriCache ??= `data:audio/wav;base64,${base64(voiceMoveEarconWav())}`;\n  return dataUriCache;\n};\n\ninterface EarconAudioNode {\n  connect(destination: EarconAudioNode): unknown;\n}\ninterface EarconAudioBuffer {\n  getChannelData(channel: number): Float32Array;\n}\ninterface EarconBufferSource extends EarconAudioNode {\n  buffer: EarconAudioBuffer | null;\n  onended: (() => void) | null;\n  start(when?: number): void;\n}\ninterface EarconAudioContext {\n  readonly state: string;\n  readonly destination: EarconAudioNode;\n  createBuffer(channels: number, length: number, sampleRate: number): EarconAudioBuffer;\n  createBufferSource(): EarconBufferSource;\n  resume?(): Promise<void>;\n  close(): Promise<void>;\n}\ninterface EarconAudioElement {\n  volume: number;\n  onended: (() => void) | null;\n  play(): Promise<void> | void;\n}\n\nconst earconGlobals = globalThis as {\n  AudioContext?: new () => EarconAudioContext;\n  webkitAudioContext?: new () => EarconAudioContext;\n  Audio?: new (src?: string) => EarconAudioElement;\n};\n\n/** A cue that never ends (a blocked autoplay) still lets go this long after it started. */\nconst PLAYBACK_BOUND_MS = VOICE_MOVE_EARCON_MS + 400;\n\n/**\n * Play the earcon on a web host (Spaces, the Chrome side panel, the SDK the\n * desktop app renders): its own short-lived `AudioContext`, apart from the\n * call's media, or an `<audio>` element where WebAudio is missing. Resolves\n * once it has played, or within {@link PLAYBACK_BOUND_MS} when the page may\n * not play audio yet (autoplay policy). Never throws; a no-op where neither\n * exists (React Native, tests).\n */\nexport const playVoiceMoveEarcon = (): Promise<void> =>\n  new Promise<void>((resolve) => {\n    let finished = false;\n    let release: () => void = () => undefined;\n    const finish = (): void => {\n      if (finished) return;\n      finished = true;\n      clearTimeout(bound);\n      try {\n        release();\n      } catch {\n        // Already closed.\n      }\n      resolve();\n    };\n    const bound = setTimeout(finish, PLAYBACK_BOUND_MS);\n    try {\n      const AudioContextCtor = earconGlobals.AudioContext ?? earconGlobals.webkitAudioContext;\n      if (AudioContextCtor) {\n        const context = new AudioContextCtor();\n        release = () => void context.close().catch(() => undefined);\n        const samples = voiceMoveEarconSamples();\n        const buffer = context.createBuffer(1, samples.length, VOICE_MOVE_EARCON_SAMPLE_RATE);\n        buffer.getChannelData(0).set(samples);\n        const source = context.createBufferSource();\n        source.buffer = buffer;\n        source.connect(context.destination);\n        source.onended = finish;\n        if (context.state === 'suspended') void context.resume?.().catch(() => undefined);\n        source.start();\n        return;\n      }\n      const AudioCtor = earconGlobals.Audio;\n      if (AudioCtor) {\n        const element = new AudioCtor(voiceMoveEarconDataUri());\n        element.volume = 1;\n        element.onended = finish;\n        release = () => {\n          element.onended = null;\n        };\n        void Promise.resolve(element.play()).catch(finish);\n        return;\n      }\n      finish();\n    } catch {\n      finish();\n    }\n  });\n"],"names":[],"mappings":"AAYO,MAAM,gCAAgC;AAGtC,MAAM,uBAAuB;AAM7B,MAAM,yBAAyB;AAGtC,MAAM,QAAQ;AAAA,EACZ,EAAE,IAAI,QAAQ,SAAS,GAAG,YAAY,IAAA;AAAA,EACtC,EAAE,IAAI,KAAK,SAAS,KAAK,YAAY,IAAA;AACvC;AAEA,MAAM,YAAY;AAClB,MAAM,gBAAgB;AAEtB,IAAI,eAAoC;AACxC,IAAI,eAA8B;AAG3B,MAAM,yBAAyB,MAAoB;AACxD,MAAI,aAAc,QAAO;AACzB,QAAM,OAAO;AACb,QAAM,UAAU,IAAI,aAAa,KAAK,MAAO,uBAAuB,MAAQ,IAAI,CAAC;AACjF,aAAW,QAAQ,OAAO;AACxB,UAAM,QAAQ,KAAK,MAAO,KAAK,UAAU,MAAQ,IAAI;AACrD,UAAM,SAAS,KAAK,MAAO,KAAK,aAAa,MAAQ,IAAI;AACzD,UAAM,SAAS,KAAK,MAAO,YAAY,MAAQ,IAAI;AACnD,aAAS,IAAI,GAAG,IAAI,UAAU,QAAQ,IAAI,QAAQ,QAAQ,KAAK,GAAG;AAChE,YAAM,IAAI,IAAI;AAEd,YAAM,SAAS,IAAI,IAAI,WAAW;AAClC,YAAM,WAAW,IAAI,SAAU,IAAI,SAAU,QAAQ;AACrD,YAAM,QAAQ,IAAI,KAAK,KAAK,KAAK,KAAK;AACtC,cAAQ,QAAQ,CAAC,KAAK,YAAY,KAAK,IAAI,KAAK,IAAI,gBAAgB,KAAK,IAAI,IAAI,KAAK;AAAA,IACxF;AAAA,EACF;AACA,MAAI,OAAO;AACX,aAAW,UAAU,QAAS,QAAO,KAAK,IAAI,MAAM,KAAK,IAAI,MAAM,CAAC;AACpE,QAAM,QAAQ,OAAO,IAAI,yBAAyB,OAAO;AACzD,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK,EAAG,SAAQ,CAAC,KAAK;AAC1D,iBAAe;AACf,SAAO;AACT;AAGO,MAAM,qBAAqB,MAAkB;AAClD,QAAM,UAAU,uBAAA;AAChB,QAAM,YAAY,QAAQ,SAAS;AACnC,QAAM,QAAQ,IAAI,WAAW,KAAK,SAAS;AAC3C,QAAM,OAAO,IAAI,SAAS,MAAM,MAAM;AACtC,QAAM,QAAQ,CAAC,QAAgB,SAAuB;AACpD,aAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,EAAG,MAAK,SAAS,SAAS,GAAG,KAAK,WAAW,CAAC,CAAC;AAAA,EACvF;AACA,QAAM,GAAG,MAAM;AACf,OAAK,UAAU,GAAG,KAAK,WAAW,IAAI;AACtC,QAAM,GAAG,MAAM;AACf,QAAM,IAAI,MAAM;AAChB,OAAK,UAAU,IAAI,IAAI,IAAI;AAC3B,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,+BAA+B,IAAI;AACtD,OAAK,UAAU,IAAI,gCAAgC,GAAG,IAAI;AAC1D,OAAK,UAAU,IAAI,GAAG,IAAI;AAC1B,OAAK,UAAU,IAAI,IAAI,IAAI;AAC3B,QAAM,IAAI,MAAM;AAChB,OAAK,UAAU,IAAI,WAAW,IAAI;AAClC,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK,GAAG;AAC1C,UAAM,UAAU,KAAK,IAAI,IAAI,KAAK,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;AACzD,SAAK,SAAS,KAAK,IAAI,GAAG,KAAK,MAAM,UAAU,KAAM,GAAG,IAAI;AAAA,EAC9D;AACA,SAAO;AACT;AAEA,MAAM,SAAS;AAGf,MAAM,SAAS,CAAC,UAA8B;AAC5C,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK,GAAG;AACxC,UAAM,IAAI,MAAM,CAAC,KAAK;AACtB,UAAM,IAAI,MAAM,IAAI,CAAC;AACrB,UAAM,IAAI,MAAM,IAAI,CAAC;AACrB,UAAM,SAAU,KAAK,MAAQ,KAAK,MAAM,KAAM,KAAK;AACnD,WAAO,OAAQ,UAAU,KAAM,EAAE;AACjC,WAAO,OAAQ,UAAU,KAAM,EAAE;AACjC,WAAO,MAAM,SAAY,MAAM,OAAQ,UAAU,IAAK,EAAE;AACxD,WAAO,MAAM,SAAY,MAAM,OAAO,SAAS,EAAE;AAAA,EACnD;AACA,SAAO;AACT;AAOO,MAAM,yBAAyB,MAAc;AAClD,kCAAiB,yBAAyB,OAAO,mBAAA,CAAoB,CAAC;AACtE,SAAO;AACT;AA2BA,MAAM,gBAAgB;AAOtB,MAAM,oBAAoB,uBAAuB;AAU1C,MAAM,sBAAsB,MACjC,IAAI,QAAc,CAAC,YAAY;AArJ1B;AAsJH,MAAI,WAAW;AACf,MAAI,UAAsB,MAAM;AAChC,QAAM,SAAS,MAAY;AACzB,QAAI,SAAU;AACd,eAAW;AACX,iBAAa,KAAK;AAClB,QAAI;AACF,cAAA;AAAA,IACF,QAAQ;AAAA,IAER;AACA,YAAA;AAAA,EACF;AACA,QAAM,QAAQ,WAAW,QAAQ,iBAAiB;AAClD,MAAI;AACF,UAAM,mBAAmB,cAAc,gBAAgB,cAAc;AACrE,QAAI,kBAAkB;AACpB,YAAM,UAAU,IAAI,iBAAA;AACpB,gBAAU,MAAM,KAAK,QAAQ,QAAQ,MAAM,MAAM,MAAS;AAC1D,YAAM,UAAU,uBAAA;AAChB,YAAM,SAAS,QAAQ,aAAa,GAAG,QAAQ,QAAQ,6BAA6B;AACpF,aAAO,eAAe,CAAC,EAAE,IAAI,OAAO;AACpC,YAAM,SAAS,QAAQ,mBAAA;AACvB,aAAO,SAAS;AAChB,aAAO,QAAQ,QAAQ,WAAW;AAClC,aAAO,UAAU;AACjB,UAAI,QAAQ,UAAU,YAAa,QAAK,aAAQ,WAAR,iCAAmB,MAAM,MAAM;AACvE,aAAO,MAAA;AACP;AAAA,IACF;AACA,UAAM,YAAY,cAAc;AAChC,QAAI,WAAW;AACb,YAAM,UAAU,IAAI,UAAU,wBAAwB;AACtD,cAAQ,SAAS;AACjB,cAAQ,UAAU;AAClB,gBAAU,MAAM;AACd,gBAAQ,UAAU;AAAA,MACpB;AACA,WAAK,QAAQ,QAAQ,QAAQ,MAAM,EAAE,MAAM,MAAM;AACjD;AAAA,IACF;AACA,WAAA;AAAA,EACF,QAAQ;AACN,WAAA;AAAA,EACF;AACF,CAAC;"}

package/dist/agent-runtime/src/voice-media.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"voice-media.cjs","sources":["../../../../agent-runtime/src/voice-media.ts"],"sourcesContent":["/** Capture processing shared by browser and native chat voice rooms. */\nexport const VOICE_MICROPHONE_CAPTURE = {\n  echoCancellation: true,\n  noiseSuppression: true,\n  autoGainControl: true,\n} as const;\n\n/** `user-<user id>-<client>`: one device of the call's owner (agora's `voice_client_identity`). */\nconst VOICE_DEVICE_IDENTITY = /^(user-.+)-[^-]+$/;\n\n/**\n * Whether a remote participant of a chat voice room is another of the person's\n * own devices. The room holds the voice agent and the owner's devices, and\n * while a call moves (`voice/transfer-*`) the old and the new device are in it\n * together. A device plays the agent only: the other device's microphone is\n * the person's own voice, and on a browser's one `<audio>` sink it would also\n * take the agent's place. `false` while this device's identity is unknown.\n */\nexport function isOwnOtherVoiceDevice({\n  identity,\n  localIdentity,\n}: {\n  identity: string;\n  localIdentity: string | undefined;\n}): boolean {\n  if (localIdentity === undefined || identity === localIdentity) return false;\n  const person = VOICE_DEVICE_IDENTITY.exec(localIdentity)?.[1];\n  return person !== undefined && VOICE_DEVICE_IDENTITY.exec(identity)?.[1] === person;\n}\n\nexport interface VoiceMediaHandlers {\n  micMuted: boolean;\n  signal?: AbortSignal;\n  onDisconnected: (reason: string) => void;\n}\n\nexport interface VoiceMediaSession {\n  setMicrophoneEnabled(enabled: boolean): Promise<void>;\n  disconnect(): Promise<void>;\n}\n\n/**\n * The platform owns its room, audio sink and capture configuration. The shared\n * lifecycle owns cancellation, late capture, mute and loss of that room. No DOM\n * or LiveKit implementation is imported by this entrypoint.\n */\nexport interface VoiceMediaDriver {\n  connect(): Promise<void>;\n  setMicrophoneEnabled(options: {\n    enabled: boolean;\n    initial: boolean;\n  }): Promise<unknown>;\n  /** Must also release the platform's remote audio sinks. Safe to call again. */\n  disconnect(): Promise<unknown>;\n  onDisconnected(\n    listener: (event: { reason: string; joinFailure: boolean }) => void\n  ): void;\n}\n\nexport async function connectVoiceMedia({\n  createDriver,\n  handlers,\n}: {\n  createDriver: () => VoiceMediaDriver;\n  handlers: VoiceMediaHandlers;\n}): Promise<VoiceMediaSession> {\n  const { signal } = handlers;\n  // Hermes does not need a DOMException implementation to cancel a voice join.\n  const endedError = (): Error => {\n    const error = new Error('The voice session ended');\n    error.name = 'AbortError';\n    return error;\n  };\n  if (signal?.aborted) throw endedError();\n  const driver = createDriver();\n  let connected = false;\n  let joined = false;\n  let closed = false;\n  let disconnecting: Promise<void> | null = null;\n  let rejectEnded!: (error: Error) => void;\n  const ended = new Promise<never>((_, reject) => {\n    rejectEnded = reject;\n  });\n  const close = (): Promise<void> => {\n    closed = true;\n    signal?.removeEventListener('abort', onAbort);\n    if (!disconnecting) {\n      try {\n        // Invoke synchronously so a browser's playback sinks stop at hangup.\n        disconnecting = Promise.resolve(driver.disconnect()).then(\n          () => undefined\n        );\n      } catch (error) {\n        disconnecting = Promise.reject(error);\n      }\n    }\n    return disconnecting;\n  };\n  const disconnectAgain = async (): Promise<void> => {\n    try {\n      await driver.disconnect();\n    } catch {\n      /* Preserve the original failure. */\n    }\n  };\n  const onAbort = (): void => {\n    rejectEnded(endedError());\n    void close().catch(() => undefined);\n  };\n  signal?.addEventListener('abort', onAbort, { once: true });\n  // A platform factory may synchronously cancel the attempt that created it.\n  if (signal?.aborted) onAbort();\n  driver.onDisconnected(({ reason, joinFailure }) => {\n    if (closed) return;\n    // Preserve connect()'s real cause when its room emits JOIN_FAILURE first.\n    if (!connected && joinFailure) return;\n    rejectEnded(new Error(reason));\n    void close().catch(() => undefined);\n    if (joined) handlers.onDisconnected(reason);\n  });\n  const checkOpen = (): void => {\n    if (closed) throw endedError();\n  };\n  const joining = (async () => {\n    try {\n      checkOpen();\n      await driver.connect();\n      connected = true;\n      checkOpen();\n      await driver.setMicrophoneEnabled({\n        enabled: !handlers.micMuted,\n        initial: true,\n      });\n      checkOpen();\n    } catch (error) {\n      const alreadyClosed = closed;\n      await close().catch(() => undefined);\n      // An uncancellable platform capture can finish after its first close.\n      if (alreadyClosed) await disconnectAgain();\n      throw error;\n    }\n  })();\n  await Promise.race([joining, ended]);\n  checkOpen();\n  joined = true;\n  return {\n    setMicrophoneEnabled: (enabled) => {\n      const changing = (async () => {\n        checkOpen();\n        try {\n          await driver.setMicrophoneEnabled({ enabled, initial: false });\n          checkOpen();\n        } finally {\n          if (closed) await disconnectAgain();\n        }\n      })();\n      return Promise.race([changing, ended]);\n    },\n    disconnect: () => {\n      rejectEnded(endedError());\n      return close();\n    },\n  };\n}\n"],"names":[],"mappings":";;AACO,MAAM,2BAA2B;AAAA,EACtC,kBAAkB;AAAA,EAClB,kBAAkB;AAAA,EAClB,iBAAiB;AACnB;AAGA,MAAM,wBAAwB;AAUvB,SAAS,sBAAsB;AAAA,EACpC;AAAA,EACA;AACF,GAGY;;AACV,MAAI,kBAAkB,UAAa,aAAa,cAAe,QAAO;AACtE,QAAM,UAAS,2BAAsB,KAAK,aAAa,MAAxC,mBAA4C;AAC3D,SAAO,WAAW,YAAa,2BAAsB,KAAK,QAAQ,MAAnC,mBAAuC,QAAO;AAC/E;AA+BA,eAAsB,kBAAkB;AAAA,EACtC;AAAA,EACA;AACF,GAG+B;AAC7B,QAAM,EAAE,WAAW;AAEnB,QAAM,aAAa,MAAa;AAC9B,UAAM,QAAQ,IAAI,MAAM,yBAAyB;AACjD,UAAM,OAAO;AACb,WAAO;AAAA,EACT;AACA,MAAI,iCAAQ,QAAS,OAAM,WAAA;AAC3B,QAAM,SAAS,aAAA;AACf,MAAI,YAAY;AAChB,MAAI,SAAS;AACb,MAAI,SAAS;AACb,MAAI,gBAAsC;AAC1C,MAAI;AACJ,QAAM,QAAQ,IAAI,QAAe,CAAC,GAAG,WAAW;AAC9C,kBAAc;AAAA,EAChB,CAAC;AACD,QAAM,QAAQ,MAAqB;AACjC,aAAS;AACT,qCAAQ,oBAAoB,SAAS;AACrC,QAAI,CAAC,eAAe;AAClB,UAAI;AAEF,wBAAgB,QAAQ,QAAQ,OAAO,WAAA,CAAY,EAAE;AAAA,UACnD,MAAM;AAAA,QAAA;AAAA,MAEV,SAAS,OAAO;AACd,wBAAgB,QAAQ,OAAO,KAAK;AAAA,MACtC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACA,QAAM,kBAAkB,YAA2B;AACjD,QAAI;AACF,YAAM,OAAO,WAAA;AAAA,IACf,QAAQ;AAAA,IAER;AAAA,EACF;AACA,QAAM,UAAU,MAAY;AAC1B,gBAAY,YAAY;AACxB,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAAA,EACpC;AACA,mCAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM;AAEnD,MAAI,iCAAQ,QAAS,SAAA;AACrB,SAAO,eAAe,CAAC,EAAE,QAAQ,kBAAkB;AACjD,QAAI,OAAQ;AAEZ,QAAI,CAAC,aAAa,YAAa;AAC/B,gBAAY,IAAI,MAAM,MAAM,CAAC;AAC7B,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAClC,QAAI,OAAQ,UAAS,eAAe,MAAM;AAAA,EAC5C,CAAC;AACD,QAAM,YAAY,MAAY;AAC5B,QAAI,cAAc,WAAA;AAAA,EACpB;AACA,QAAM,WAAW,YAAY;AAC3B,QAAI;AACF,gBAAA;AACA,YAAM,OAAO,QAAA;AACb,kBAAY;AACZ,gBAAA;AACA,YAAM,OAAO,qBAAqB;AAAA,QAChC,SAAS,CAAC,SAAS;AAAA,QACnB,SAAS;AAAA,MAAA,CACV;AACD,gBAAA;AAAA,IACF,SAAS,OAAO;AACd,YAAM,gBAAgB;AACtB,YAAM,MAAA,EAAQ,MAAM,MAAM,MAAS;AAEnC,UAAI,qBAAqB,gBAAA;AACzB,YAAM;AAAA,IACR;AAAA,EACF,GAAA;AACA,QAAM,QAAQ,KAAK,CAAC,SAAS,KAAK,CAAC;AACnC,YAAA;AACA,WAAS;AACT,SAAO;AAAA,IACL,sBAAsB,CAAC,YAAY;AACjC,YAAM,YAAY,YAAY;AAC5B,kBAAA;AACA,YAAI;AACF,gBAAM,OAAO,qBAAqB,EAAE,SAAS,SAAS,OAAO;AAC7D,oBAAA;AAAA,QACF,UAAA;AACE,cAAI,cAAc,gBAAA;AAAA,QACpB;AAAA,MACF,GAAA;AACA,aAAO,QAAQ,KAAK,CAAC,UAAU,KAAK,CAAC;AAAA,IACvC;AAAA,IACA,YAAY,MAAM;AAChB,kBAAY,YAAY;AACxB,aAAO,MAAA;AAAA,IACT;AAAA,EAAA;AAEJ;;;;"}
+{"version":3,"file":"voice-media.cjs","sources":["../../../../agent-runtime/src/voice-media.ts"],"sourcesContent":["/** Capture processing shared by browser and native chat voice rooms. */\nexport const VOICE_MICROPHONE_CAPTURE = {\n  echoCancellation: true,\n  noiseSuppression: true,\n  autoGainControl: true,\n} as const;\n\n/** `user-<user id>-<client>`: one device of the call's owner (agora's `voice_client_identity`). */\nconst VOICE_DEVICE_IDENTITY = /^(user-.+)-[^-]+$/;\n\n/**\n * Whether a remote participant of a chat voice room is another of the person's\n * own devices. The room holds the voice agent and the owner's devices, and\n * while a call moves (`voice/transfer-*`) the old and the new device are in it\n * together. A device plays the agent only: the other device's microphone is\n * the person's own voice, and on a browser's one `<audio>` sink it would also\n * take the agent's place. `false` while this device's identity is unknown.\n */\nexport function isOwnOtherVoiceDevice({\n  identity,\n  localIdentity,\n}: {\n  identity: string;\n  localIdentity: string | undefined;\n}): boolean {\n  if (localIdentity === undefined || identity === localIdentity) return false;\n  const person = VOICE_DEVICE_IDENTITY.exec(localIdentity)?.[1];\n  return person !== undefined && VOICE_DEVICE_IDENTITY.exec(identity)?.[1] === person;\n}\n\n/**\n * What a platform's room reports about the call's media while it lasts, for\n * the surface's telemetry: the room reconnecting, the agent leaving or coming\n * back, connection quality, and periodic audio-quality samples of what the\n * person hears (`audio-quality` when an interval was degraded, one\n * `audio-summary` when the room closes). Ids-free; values are scalars.\n */\nexport type VoiceMediaEventType =\n  | 'reconnecting'\n  | 'signal-reconnecting'\n  | 'reconnected'\n  | 'connection-quality'\n  | 'agent-left'\n  | 'agent-joined'\n  | 'agent-track-unsubscribed'\n  | 'track-subscription-failed'\n  | 'device-error'\n  | 'audio-quality'\n  | 'audio-summary';\n\nexport interface VoiceMediaEvent {\n  type: VoiceMediaEventType;\n  data?: Record<string, string | number | boolean | null>;\n}\n\nexport interface VoiceMediaHandlers {\n  micMuted: boolean;\n  signal?: AbortSignal;\n  onDisconnected: (reason: string) => void;\n  /** Media telemetry from the platform's room (optional; must not throw). */\n  onEvent?: (event: VoiceMediaEvent) => void;\n}\n\nexport interface VoiceMediaSession {\n  setMicrophoneEnabled(enabled: boolean): Promise<void>;\n  disconnect(): Promise<void>;\n}\n\n/**\n * The platform owns its room, audio sink and capture configuration. The shared\n * lifecycle owns cancellation, late capture, mute and loss of that room. No DOM\n * or LiveKit implementation is imported by this entrypoint.\n */\nexport interface VoiceMediaDriver {\n  connect(): Promise<void>;\n  setMicrophoneEnabled(options: {\n    enabled: boolean;\n    initial: boolean;\n  }): Promise<unknown>;\n  /** Must also release the platform's remote audio sinks. Safe to call again. */\n  disconnect(): Promise<unknown>;\n  onDisconnected(\n    listener: (event: { reason: string; joinFailure: boolean }) => void\n  ): void;\n}\n\nexport async function connectVoiceMedia({\n  createDriver,\n  handlers,\n}: {\n  createDriver: () => VoiceMediaDriver;\n  handlers: VoiceMediaHandlers;\n}): Promise<VoiceMediaSession> {\n  const { signal } = handlers;\n  // Hermes does not need a DOMException implementation to cancel a voice join.\n  const endedError = (): Error => {\n    const error = new Error('The voice session ended');\n    error.name = 'AbortError';\n    return error;\n  };\n  if (signal?.aborted) throw endedError();\n  const driver = createDriver();\n  let connected = false;\n  let joined = false;\n  let closed = false;\n  let disconnecting: Promise<void> | null = null;\n  let rejectEnded!: (error: Error) => void;\n  const ended = new Promise<never>((_, reject) => {\n    rejectEnded = reject;\n  });\n  const close = (): Promise<void> => {\n    closed = true;\n    signal?.removeEventListener('abort', onAbort);\n    if (!disconnecting) {\n      try {\n        // Invoke synchronously so a browser's playback sinks stop at hangup.\n        disconnecting = Promise.resolve(driver.disconnect()).then(\n          () => undefined\n        );\n      } catch (error) {\n        disconnecting = Promise.reject(error);\n      }\n    }\n    return disconnecting;\n  };\n  const disconnectAgain = async (): Promise<void> => {\n    try {\n      await driver.disconnect();\n    } catch {\n      /* Preserve the original failure. */\n    }\n  };\n  const onAbort = (): void => {\n    rejectEnded(endedError());\n    void close().catch(() => undefined);\n  };\n  signal?.addEventListener('abort', onAbort, { once: true });\n  // A platform factory may synchronously cancel the attempt that created it.\n  if (signal?.aborted) onAbort();\n  driver.onDisconnected(({ reason, joinFailure }) => {\n    if (closed) return;\n    // Preserve connect()'s real cause when its room emits JOIN_FAILURE first.\n    if (!connected && joinFailure) return;\n    rejectEnded(new Error(reason));\n    void close().catch(() => undefined);\n    if (joined) handlers.onDisconnected(reason);\n  });\n  const checkOpen = (): void => {\n    if (closed) throw endedError();\n  };\n  const joining = (async () => {\n    try {\n      checkOpen();\n      await driver.connect();\n      connected = true;\n      checkOpen();\n      await driver.setMicrophoneEnabled({\n        enabled: !handlers.micMuted,\n        initial: true,\n      });\n      checkOpen();\n    } catch (error) {\n      const alreadyClosed = closed;\n      await close().catch(() => undefined);\n      // An uncancellable platform capture can finish after its first close.\n      if (alreadyClosed) await disconnectAgain();\n      throw error;\n    }\n  })();\n  await Promise.race([joining, ended]);\n  checkOpen();\n  joined = true;\n  return {\n    setMicrophoneEnabled: (enabled) => {\n      const changing = (async () => {\n        checkOpen();\n        try {\n          await driver.setMicrophoneEnabled({ enabled, initial: false });\n          checkOpen();\n        } finally {\n          if (closed) await disconnectAgain();\n        }\n      })();\n      return Promise.race([changing, ended]);\n    },\n    disconnect: () => {\n      rejectEnded(endedError());\n      return close();\n    },\n  };\n}\n"],"names":[],"mappings":";;AACO,MAAM,2BAA2B;AAAA,EACtC,kBAAkB;AAAA,EAClB,kBAAkB;AAAA,EAClB,iBAAiB;AACnB;AAGA,MAAM,wBAAwB;AAUvB,SAAS,sBAAsB;AAAA,EACpC;AAAA,EACA;AACF,GAGY;;AACV,MAAI,kBAAkB,UAAa,aAAa,cAAe,QAAO;AACtE,QAAM,UAAS,2BAAsB,KAAK,aAAa,MAAxC,mBAA4C;AAC3D,SAAO,WAAW,YAAa,2BAAsB,KAAK,QAAQ,MAAnC,mBAAuC,QAAO;AAC/E;AA0DA,eAAsB,kBAAkB;AAAA,EACtC;AAAA,EACA;AACF,GAG+B;AAC7B,QAAM,EAAE,WAAW;AAEnB,QAAM,aAAa,MAAa;AAC9B,UAAM,QAAQ,IAAI,MAAM,yBAAyB;AACjD,UAAM,OAAO;AACb,WAAO;AAAA,EACT;AACA,MAAI,iCAAQ,QAAS,OAAM,WAAA;AAC3B,QAAM,SAAS,aAAA;AACf,MAAI,YAAY;AAChB,MAAI,SAAS;AACb,MAAI,SAAS;AACb,MAAI,gBAAsC;AAC1C,MAAI;AACJ,QAAM,QAAQ,IAAI,QAAe,CAAC,GAAG,WAAW;AAC9C,kBAAc;AAAA,EAChB,CAAC;AACD,QAAM,QAAQ,MAAqB;AACjC,aAAS;AACT,qCAAQ,oBAAoB,SAAS;AACrC,QAAI,CAAC,eAAe;AAClB,UAAI;AAEF,wBAAgB,QAAQ,QAAQ,OAAO,WAAA,CAAY,EAAE;AAAA,UACnD,MAAM;AAAA,QAAA;AAAA,MAEV,SAAS,OAAO;AACd,wBAAgB,QAAQ,OAAO,KAAK;AAAA,MACtC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACA,QAAM,kBAAkB,YAA2B;AACjD,QAAI;AACF,YAAM,OAAO,WAAA;AAAA,IACf,QAAQ;AAAA,IAER;AAAA,EACF;AACA,QAAM,UAAU,MAAY;AAC1B,gBAAY,YAAY;AACxB,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAAA,EACpC;AACA,mCAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM;AAEnD,MAAI,iCAAQ,QAAS,SAAA;AACrB,SAAO,eAAe,CAAC,EAAE,QAAQ,kBAAkB;AACjD,QAAI,OAAQ;AAEZ,QAAI,CAAC,aAAa,YAAa;AAC/B,gBAAY,IAAI,MAAM,MAAM,CAAC;AAC7B,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAClC,QAAI,OAAQ,UAAS,eAAe,MAAM;AAAA,EAC5C,CAAC;AACD,QAAM,YAAY,MAAY;AAC5B,QAAI,cAAc,WAAA;AAAA,EACpB;AACA,QAAM,WAAW,YAAY;AAC3B,QAAI;AACF,gBAAA;AACA,YAAM,OAAO,QAAA;AACb,kBAAY;AACZ,gBAAA;AACA,YAAM,OAAO,qBAAqB;AAAA,QAChC,SAAS,CAAC,SAAS;AAAA,QACnB,SAAS;AAAA,MAAA,CACV;AACD,gBAAA;AAAA,IACF,SAAS,OAAO;AACd,YAAM,gBAAgB;AACtB,YAAM,MAAA,EAAQ,MAAM,MAAM,MAAS;AAEnC,UAAI,qBAAqB,gBAAA;AACzB,YAAM;AAAA,IACR;AAAA,EACF,GAAA;AACA,QAAM,QAAQ,KAAK,CAAC,SAAS,KAAK,CAAC;AACnC,YAAA;AACA,WAAS;AACT,SAAO;AAAA,IACL,sBAAsB,CAAC,YAAY;AACjC,YAAM,YAAY,YAAY;AAC5B,kBAAA;AACA,YAAI;AACF,gBAAM,OAAO,qBAAqB,EAAE,SAAS,SAAS,OAAO;AAC7D,oBAAA;AAAA,QACF,UAAA;AACE,cAAI,cAAc,gBAAA;AAAA,QACpB;AAAA,MACF,GAAA;AACA,aAAO,QAAQ,KAAK,CAAC,UAAU,KAAK,CAAC;AAAA,IACvC;AAAA,IACA,YAAY,MAAM;AAChB,kBAAY,YAAY;AACxB,aAAO,MAAA;AAAA,IACT;AAAA,EAAA;AAEJ;;;;"}

package/dist/agent-runtime/src/voice-media.d.ts

@@ -16,10 +16,24 @@ export declare function isOwnOtherVoiceDevice({ identity, localIdentity, }: {
     identity: string;
     localIdentity: string | undefined;
 }): boolean;
+/**
+ * What a platform's room reports about the call's media while it lasts, for
+ * the surface's telemetry: the room reconnecting, the agent leaving or coming
+ * back, connection quality, and periodic audio-quality samples of what the
+ * person hears (`audio-quality` when an interval was degraded, one
+ * `audio-summary` when the room closes). Ids-free; values are scalars.
+ */
+export type VoiceMediaEventType = 'reconnecting' | 'signal-reconnecting' | 'reconnected' | 'connection-quality' | 'agent-left' | 'agent-joined' | 'agent-track-unsubscribed' | 'track-subscription-failed' | 'device-error' | 'audio-quality' | 'audio-summary';
+export interface VoiceMediaEvent {
+    type: VoiceMediaEventType;
+    data?: Record<string, string | number | boolean | null>;
+}
 export interface VoiceMediaHandlers {
     micMuted: boolean;
     signal?: AbortSignal;
     onDisconnected: (reason: string) => void;
+    /** Media telemetry from the platform's room (optional; must not throw). */
+    onEvent?: (event: VoiceMediaEvent) => void;
 }
 export interface VoiceMediaSession {
     setMicrophoneEnabled(enabled: boolean): Promise<void>;

package/dist/agent-runtime/src/voice-media.js.map

@@ -1 +1 @@
-{"version":3,"file":"voice-media.js","sources":["../../../../agent-runtime/src/voice-media.ts"],"sourcesContent":["/** Capture processing shared by browser and native chat voice rooms. */\nexport const VOICE_MICROPHONE_CAPTURE = {\n  echoCancellation: true,\n  noiseSuppression: true,\n  autoGainControl: true,\n} as const;\n\n/** `user-<user id>-<client>`: one device of the call's owner (agora's `voice_client_identity`). */\nconst VOICE_DEVICE_IDENTITY = /^(user-.+)-[^-]+$/;\n\n/**\n * Whether a remote participant of a chat voice room is another of the person's\n * own devices. The room holds the voice agent and the owner's devices, and\n * while a call moves (`voice/transfer-*`) the old and the new device are in it\n * together. A device plays the agent only: the other device's microphone is\n * the person's own voice, and on a browser's one `<audio>` sink it would also\n * take the agent's place. `false` while this device's identity is unknown.\n */\nexport function isOwnOtherVoiceDevice({\n  identity,\n  localIdentity,\n}: {\n  identity: string;\n  localIdentity: string | undefined;\n}): boolean {\n  if (localIdentity === undefined || identity === localIdentity) return false;\n  const person = VOICE_DEVICE_IDENTITY.exec(localIdentity)?.[1];\n  return person !== undefined && VOICE_DEVICE_IDENTITY.exec(identity)?.[1] === person;\n}\n\nexport interface VoiceMediaHandlers {\n  micMuted: boolean;\n  signal?: AbortSignal;\n  onDisconnected: (reason: string) => void;\n}\n\nexport interface VoiceMediaSession {\n  setMicrophoneEnabled(enabled: boolean): Promise<void>;\n  disconnect(): Promise<void>;\n}\n\n/**\n * The platform owns its room, audio sink and capture configuration. The shared\n * lifecycle owns cancellation, late capture, mute and loss of that room. No DOM\n * or LiveKit implementation is imported by this entrypoint.\n */\nexport interface VoiceMediaDriver {\n  connect(): Promise<void>;\n  setMicrophoneEnabled(options: {\n    enabled: boolean;\n    initial: boolean;\n  }): Promise<unknown>;\n  /** Must also release the platform's remote audio sinks. Safe to call again. */\n  disconnect(): Promise<unknown>;\n  onDisconnected(\n    listener: (event: { reason: string; joinFailure: boolean }) => void\n  ): void;\n}\n\nexport async function connectVoiceMedia({\n  createDriver,\n  handlers,\n}: {\n  createDriver: () => VoiceMediaDriver;\n  handlers: VoiceMediaHandlers;\n}): Promise<VoiceMediaSession> {\n  const { signal } = handlers;\n  // Hermes does not need a DOMException implementation to cancel a voice join.\n  const endedError = (): Error => {\n    const error = new Error('The voice session ended');\n    error.name = 'AbortError';\n    return error;\n  };\n  if (signal?.aborted) throw endedError();\n  const driver = createDriver();\n  let connected = false;\n  let joined = false;\n  let closed = false;\n  let disconnecting: Promise<void> | null = null;\n  let rejectEnded!: (error: Error) => void;\n  const ended = new Promise<never>((_, reject) => {\n    rejectEnded = reject;\n  });\n  const close = (): Promise<void> => {\n    closed = true;\n    signal?.removeEventListener('abort', onAbort);\n    if (!disconnecting) {\n      try {\n        // Invoke synchronously so a browser's playback sinks stop at hangup.\n        disconnecting = Promise.resolve(driver.disconnect()).then(\n          () => undefined\n        );\n      } catch (error) {\n        disconnecting = Promise.reject(error);\n      }\n    }\n    return disconnecting;\n  };\n  const disconnectAgain = async (): Promise<void> => {\n    try {\n      await driver.disconnect();\n    } catch {\n      /* Preserve the original failure. */\n    }\n  };\n  const onAbort = (): void => {\n    rejectEnded(endedError());\n    void close().catch(() => undefined);\n  };\n  signal?.addEventListener('abort', onAbort, { once: true });\n  // A platform factory may synchronously cancel the attempt that created it.\n  if (signal?.aborted) onAbort();\n  driver.onDisconnected(({ reason, joinFailure }) => {\n    if (closed) return;\n    // Preserve connect()'s real cause when its room emits JOIN_FAILURE first.\n    if (!connected && joinFailure) return;\n    rejectEnded(new Error(reason));\n    void close().catch(() => undefined);\n    if (joined) handlers.onDisconnected(reason);\n  });\n  const checkOpen = (): void => {\n    if (closed) throw endedError();\n  };\n  const joining = (async () => {\n    try {\n      checkOpen();\n      await driver.connect();\n      connected = true;\n      checkOpen();\n      await driver.setMicrophoneEnabled({\n        enabled: !handlers.micMuted,\n        initial: true,\n      });\n      checkOpen();\n    } catch (error) {\n      const alreadyClosed = closed;\n      await close().catch(() => undefined);\n      // An uncancellable platform capture can finish after its first close.\n      if (alreadyClosed) await disconnectAgain();\n      throw error;\n    }\n  })();\n  await Promise.race([joining, ended]);\n  checkOpen();\n  joined = true;\n  return {\n    setMicrophoneEnabled: (enabled) => {\n      const changing = (async () => {\n        checkOpen();\n        try {\n          await driver.setMicrophoneEnabled({ enabled, initial: false });\n          checkOpen();\n        } finally {\n          if (closed) await disconnectAgain();\n        }\n      })();\n      return Promise.race([changing, ended]);\n    },\n    disconnect: () => {\n      rejectEnded(endedError());\n      return close();\n    },\n  };\n}\n"],"names":[],"mappings":"AACO,MAAM,2BAA2B;AAAA,EACtC,kBAAkB;AAAA,EAClB,kBAAkB;AAAA,EAClB,iBAAiB;AACnB;AAGA,MAAM,wBAAwB;AAUvB,SAAS,sBAAsB;AAAA,EACpC;AAAA,EACA;AACF,GAGY;AAvBL;AAwBL,MAAI,kBAAkB,UAAa,aAAa,cAAe,QAAO;AACtE,QAAM,UAAS,2BAAsB,KAAK,aAAa,MAAxC,mBAA4C;AAC3D,SAAO,WAAW,YAAa,2BAAsB,KAAK,QAAQ,MAAnC,mBAAuC,QAAO;AAC/E;AA+BA,eAAsB,kBAAkB;AAAA,EACtC;AAAA,EACA;AACF,GAG+B;AAC7B,QAAM,EAAE,WAAW;AAEnB,QAAM,aAAa,MAAa;AAC9B,UAAM,QAAQ,IAAI,MAAM,yBAAyB;AACjD,UAAM,OAAO;AACb,WAAO;AAAA,EACT;AACA,MAAI,iCAAQ,QAAS,OAAM,WAAA;AAC3B,QAAM,SAAS,aAAA;AACf,MAAI,YAAY;AAChB,MAAI,SAAS;AACb,MAAI,SAAS;AACb,MAAI,gBAAsC;AAC1C,MAAI;AACJ,QAAM,QAAQ,IAAI,QAAe,CAAC,GAAG,WAAW;AAC9C,kBAAc;AAAA,EAChB,CAAC;AACD,QAAM,QAAQ,MAAqB;AACjC,aAAS;AACT,qCAAQ,oBAAoB,SAAS;AACrC,QAAI,CAAC,eAAe;AAClB,UAAI;AAEF,wBAAgB,QAAQ,QAAQ,OAAO,WAAA,CAAY,EAAE;AAAA,UACnD,MAAM;AAAA,QAAA;AAAA,MAEV,SAAS,OAAO;AACd,wBAAgB,QAAQ,OAAO,KAAK;AAAA,MACtC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACA,QAAM,kBAAkB,YAA2B;AACjD,QAAI;AACF,YAAM,OAAO,WAAA;AAAA,IACf,QAAQ;AAAA,IAER;AAAA,EACF;AACA,QAAM,UAAU,MAAY;AAC1B,gBAAY,YAAY;AACxB,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAAA,EACpC;AACA,mCAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM;AAEnD,MAAI,iCAAQ,QAAS,SAAA;AACrB,SAAO,eAAe,CAAC,EAAE,QAAQ,kBAAkB;AACjD,QAAI,OAAQ;AAEZ,QAAI,CAAC,aAAa,YAAa;AAC/B,gBAAY,IAAI,MAAM,MAAM,CAAC;AAC7B,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAClC,QAAI,OAAQ,UAAS,eAAe,MAAM;AAAA,EAC5C,CAAC;AACD,QAAM,YAAY,MAAY;AAC5B,QAAI,cAAc,WAAA;AAAA,EACpB;AACA,QAAM,WAAW,YAAY;AAC3B,QAAI;AACF,gBAAA;AACA,YAAM,OAAO,QAAA;AACb,kBAAY;AACZ,gBAAA;AACA,YAAM,OAAO,qBAAqB;AAAA,QAChC,SAAS,CAAC,SAAS;AAAA,QACnB,SAAS;AAAA,MAAA,CACV;AACD,gBAAA;AAAA,IACF,SAAS,OAAO;AACd,YAAM,gBAAgB;AACtB,YAAM,MAAA,EAAQ,MAAM,MAAM,MAAS;AAEnC,UAAI,qBAAqB,gBAAA;AACzB,YAAM;AAAA,IACR;AAAA,EACF,GAAA;AACA,QAAM,QAAQ,KAAK,CAAC,SAAS,KAAK,CAAC;AACnC,YAAA;AACA,WAAS;AACT,SAAO;AAAA,IACL,sBAAsB,CAAC,YAAY;AACjC,YAAM,YAAY,YAAY;AAC5B,kBAAA;AACA,YAAI;AACF,gBAAM,OAAO,qBAAqB,EAAE,SAAS,SAAS,OAAO;AAC7D,oBAAA;AAAA,QACF,UAAA;AACE,cAAI,cAAc,gBAAA;AAAA,QACpB;AAAA,MACF,GAAA;AACA,aAAO,QAAQ,KAAK,CAAC,UAAU,KAAK,CAAC;AAAA,IACvC;AAAA,IACA,YAAY,MAAM;AAChB,kBAAY,YAAY;AACxB,aAAO,MAAA;AAAA,IACT;AAAA,EAAA;AAEJ;"}
+{"version":3,"file":"voice-media.js","sources":["../../../../agent-runtime/src/voice-media.ts"],"sourcesContent":["/** Capture processing shared by browser and native chat voice rooms. */\nexport const VOICE_MICROPHONE_CAPTURE = {\n  echoCancellation: true,\n  noiseSuppression: true,\n  autoGainControl: true,\n} as const;\n\n/** `user-<user id>-<client>`: one device of the call's owner (agora's `voice_client_identity`). */\nconst VOICE_DEVICE_IDENTITY = /^(user-.+)-[^-]+$/;\n\n/**\n * Whether a remote participant of a chat voice room is another of the person's\n * own devices. The room holds the voice agent and the owner's devices, and\n * while a call moves (`voice/transfer-*`) the old and the new device are in it\n * together. A device plays the agent only: the other device's microphone is\n * the person's own voice, and on a browser's one `<audio>` sink it would also\n * take the agent's place. `false` while this device's identity is unknown.\n */\nexport function isOwnOtherVoiceDevice({\n  identity,\n  localIdentity,\n}: {\n  identity: string;\n  localIdentity: string | undefined;\n}): boolean {\n  if (localIdentity === undefined || identity === localIdentity) return false;\n  const person = VOICE_DEVICE_IDENTITY.exec(localIdentity)?.[1];\n  return person !== undefined && VOICE_DEVICE_IDENTITY.exec(identity)?.[1] === person;\n}\n\n/**\n * What a platform's room reports about the call's media while it lasts, for\n * the surface's telemetry: the room reconnecting, the agent leaving or coming\n * back, connection quality, and periodic audio-quality samples of what the\n * person hears (`audio-quality` when an interval was degraded, one\n * `audio-summary` when the room closes). Ids-free; values are scalars.\n */\nexport type VoiceMediaEventType =\n  | 'reconnecting'\n  | 'signal-reconnecting'\n  | 'reconnected'\n  | 'connection-quality'\n  | 'agent-left'\n  | 'agent-joined'\n  | 'agent-track-unsubscribed'\n  | 'track-subscription-failed'\n  | 'device-error'\n  | 'audio-quality'\n  | 'audio-summary';\n\nexport interface VoiceMediaEvent {\n  type: VoiceMediaEventType;\n  data?: Record<string, string | number | boolean | null>;\n}\n\nexport interface VoiceMediaHandlers {\n  micMuted: boolean;\n  signal?: AbortSignal;\n  onDisconnected: (reason: string) => void;\n  /** Media telemetry from the platform's room (optional; must not throw). */\n  onEvent?: (event: VoiceMediaEvent) => void;\n}\n\nexport interface VoiceMediaSession {\n  setMicrophoneEnabled(enabled: boolean): Promise<void>;\n  disconnect(): Promise<void>;\n}\n\n/**\n * The platform owns its room, audio sink and capture configuration. The shared\n * lifecycle owns cancellation, late capture, mute and loss of that room. No DOM\n * or LiveKit implementation is imported by this entrypoint.\n */\nexport interface VoiceMediaDriver {\n  connect(): Promise<void>;\n  setMicrophoneEnabled(options: {\n    enabled: boolean;\n    initial: boolean;\n  }): Promise<unknown>;\n  /** Must also release the platform's remote audio sinks. Safe to call again. */\n  disconnect(): Promise<unknown>;\n  onDisconnected(\n    listener: (event: { reason: string; joinFailure: boolean }) => void\n  ): void;\n}\n\nexport async function connectVoiceMedia({\n  createDriver,\n  handlers,\n}: {\n  createDriver: () => VoiceMediaDriver;\n  handlers: VoiceMediaHandlers;\n}): Promise<VoiceMediaSession> {\n  const { signal } = handlers;\n  // Hermes does not need a DOMException implementation to cancel a voice join.\n  const endedError = (): Error => {\n    const error = new Error('The voice session ended');\n    error.name = 'AbortError';\n    return error;\n  };\n  if (signal?.aborted) throw endedError();\n  const driver = createDriver();\n  let connected = false;\n  let joined = false;\n  let closed = false;\n  let disconnecting: Promise<void> | null = null;\n  let rejectEnded!: (error: Error) => void;\n  const ended = new Promise<never>((_, reject) => {\n    rejectEnded = reject;\n  });\n  const close = (): Promise<void> => {\n    closed = true;\n    signal?.removeEventListener('abort', onAbort);\n    if (!disconnecting) {\n      try {\n        // Invoke synchronously so a browser's playback sinks stop at hangup.\n        disconnecting = Promise.resolve(driver.disconnect()).then(\n          () => undefined\n        );\n      } catch (error) {\n        disconnecting = Promise.reject(error);\n      }\n    }\n    return disconnecting;\n  };\n  const disconnectAgain = async (): Promise<void> => {\n    try {\n      await driver.disconnect();\n    } catch {\n      /* Preserve the original failure. */\n    }\n  };\n  const onAbort = (): void => {\n    rejectEnded(endedError());\n    void close().catch(() => undefined);\n  };\n  signal?.addEventListener('abort', onAbort, { once: true });\n  // A platform factory may synchronously cancel the attempt that created it.\n  if (signal?.aborted) onAbort();\n  driver.onDisconnected(({ reason, joinFailure }) => {\n    if (closed) return;\n    // Preserve connect()'s real cause when its room emits JOIN_FAILURE first.\n    if (!connected && joinFailure) return;\n    rejectEnded(new Error(reason));\n    void close().catch(() => undefined);\n    if (joined) handlers.onDisconnected(reason);\n  });\n  const checkOpen = (): void => {\n    if (closed) throw endedError();\n  };\n  const joining = (async () => {\n    try {\n      checkOpen();\n      await driver.connect();\n      connected = true;\n      checkOpen();\n      await driver.setMicrophoneEnabled({\n        enabled: !handlers.micMuted,\n        initial: true,\n      });\n      checkOpen();\n    } catch (error) {\n      const alreadyClosed = closed;\n      await close().catch(() => undefined);\n      // An uncancellable platform capture can finish after its first close.\n      if (alreadyClosed) await disconnectAgain();\n      throw error;\n    }\n  })();\n  await Promise.race([joining, ended]);\n  checkOpen();\n  joined = true;\n  return {\n    setMicrophoneEnabled: (enabled) => {\n      const changing = (async () => {\n        checkOpen();\n        try {\n          await driver.setMicrophoneEnabled({ enabled, initial: false });\n          checkOpen();\n        } finally {\n          if (closed) await disconnectAgain();\n        }\n      })();\n      return Promise.race([changing, ended]);\n    },\n    disconnect: () => {\n      rejectEnded(endedError());\n      return close();\n    },\n  };\n}\n"],"names":[],"mappings":"AACO,MAAM,2BAA2B;AAAA,EACtC,kBAAkB;AAAA,EAClB,kBAAkB;AAAA,EAClB,iBAAiB;AACnB;AAGA,MAAM,wBAAwB;AAUvB,SAAS,sBAAsB;AAAA,EACpC;AAAA,EACA;AACF,GAGY;AAvBL;AAwBL,MAAI,kBAAkB,UAAa,aAAa,cAAe,QAAO;AACtE,QAAM,UAAS,2BAAsB,KAAK,aAAa,MAAxC,mBAA4C;AAC3D,SAAO,WAAW,YAAa,2BAAsB,KAAK,QAAQ,MAAnC,mBAAuC,QAAO;AAC/E;AA0DA,eAAsB,kBAAkB;AAAA,EACtC;AAAA,EACA;AACF,GAG+B;AAC7B,QAAM,EAAE,WAAW;AAEnB,QAAM,aAAa,MAAa;AAC9B,UAAM,QAAQ,IAAI,MAAM,yBAAyB;AACjD,UAAM,OAAO;AACb,WAAO;AAAA,EACT;AACA,MAAI,iCAAQ,QAAS,OAAM,WAAA;AAC3B,QAAM,SAAS,aAAA;AACf,MAAI,YAAY;AAChB,MAAI,SAAS;AACb,MAAI,SAAS;AACb,MAAI,gBAAsC;AAC1C,MAAI;AACJ,QAAM,QAAQ,IAAI,QAAe,CAAC,GAAG,WAAW;AAC9C,kBAAc;AAAA,EAChB,CAAC;AACD,QAAM,QAAQ,MAAqB;AACjC,aAAS;AACT,qCAAQ,oBAAoB,SAAS;AACrC,QAAI,CAAC,eAAe;AAClB,UAAI;AAEF,wBAAgB,QAAQ,QAAQ,OAAO,WAAA,CAAY,EAAE;AAAA,UACnD,MAAM;AAAA,QAAA;AAAA,MAEV,SAAS,OAAO;AACd,wBAAgB,QAAQ,OAAO,KAAK;AAAA,MACtC;AAAA,IACF;AACA,WAAO;AAAA,EACT;AACA,QAAM,kBAAkB,YAA2B;AACjD,QAAI;AACF,YAAM,OAAO,WAAA;AAAA,IACf,QAAQ;AAAA,IAER;AAAA,EACF;AACA,QAAM,UAAU,MAAY;AAC1B,gBAAY,YAAY;AACxB,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAAA,EACpC;AACA,mCAAQ,iBAAiB,SAAS,SAAS,EAAE,MAAM;AAEnD,MAAI,iCAAQ,QAAS,SAAA;AACrB,SAAO,eAAe,CAAC,EAAE,QAAQ,kBAAkB;AACjD,QAAI,OAAQ;AAEZ,QAAI,CAAC,aAAa,YAAa;AAC/B,gBAAY,IAAI,MAAM,MAAM,CAAC;AAC7B,SAAK,MAAA,EAAQ,MAAM,MAAM,MAAS;AAClC,QAAI,OAAQ,UAAS,eAAe,MAAM;AAAA,EAC5C,CAAC;AACD,QAAM,YAAY,MAAY;AAC5B,QAAI,cAAc,WAAA;AAAA,EACpB;AACA,QAAM,WAAW,YAAY;AAC3B,QAAI;AACF,gBAAA;AACA,YAAM,OAAO,QAAA;AACb,kBAAY;AACZ,gBAAA;AACA,YAAM,OAAO,qBAAqB;AAAA,QAChC,SAAS,CAAC,SAAS;AAAA,QACnB,SAAS;AAAA,MAAA,CACV;AACD,gBAAA;AAAA,IACF,SAAS,OAAO;AACd,YAAM,gBAAgB;AACtB,YAAM,MAAA,EAAQ,MAAM,MAAM,MAAS;AAEnC,UAAI,qBAAqB,gBAAA;AACzB,YAAM;AAAA,IACR;AAAA,EACF,GAAA;AACA,QAAM,QAAQ,KAAK,CAAC,SAAS,KAAK,CAAC;AACnC,YAAA;AACA,WAAS;AACT,SAAO;AAAA,IACL,sBAAsB,CAAC,YAAY;AACjC,YAAM,YAAY,YAAY;AAC5B,kBAAA;AACA,YAAI;AACF,gBAAM,OAAO,qBAAqB,EAAE,SAAS,SAAS,OAAO;AAC7D,oBAAA;AAAA,QACF,UAAA;AACE,cAAI,cAAc,gBAAA;AAAA,QACpB;AAAA,MACF,GAAA;AACA,aAAO,QAAQ,KAAK,CAAC,UAAU,KAAK,CAAC;AAAA,IACvC;AAAA,IACA,YAAY,MAAM;AAChB,kBAAY,YAAY;AACxB,aAAO,MAAA;AAAA,IACT;AAAA,EAAA;AAEJ;"}

package/dist/agent-runtime/src/voice-presentation.cjs

@@ -2,6 +2,7 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: "Module" });
 const core = require("@assistant-ui/core");
 const VOICE_SOURCE_SPOKEN = "voice";
+const VOICE_SOURCE_TYPED = "text";
 const VOICE_SOURCE_DELEGATION = "voice_delegation";
 const VOICE_DELEGATION_ID_PREFIX = "voicetask_";
 const VOICE_LIVE_KEY = "voice_live";
@@ -50,6 +51,7 @@ exports.VOICE_LIVE_KEY = VOICE_LIVE_KEY;
 exports.VOICE_LIVE_TYPED_HINT = VOICE_LIVE_TYPED_HINT;
 exports.VOICE_SOURCE_DELEGATION = VOICE_SOURCE_DELEGATION;
 exports.VOICE_SOURCE_SPOKEN = VOICE_SOURCE_SPOKEN;
+exports.VOICE_SOURCE_TYPED = VOICE_SOURCE_TYPED;
 exports.VOICE_TURN_IDS_KEY = VOICE_TURN_IDS_KEY;
 exports.VOICE_UPDATE_TOOL_NAME = VOICE_UPDATE_TOOL_NAME;
 exports.describeVoiceUpdateToolCall = describeVoiceUpdateToolCall;

package/dist/agent-runtime/src/voice-presentation.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"voice-presentation.cjs","sources":["../../../../agent-runtime/src/voice-presentation.ts"],"sourcesContent":["/**\n * How voice-mode transcript entries present, shared by every chat surface\n * (Olympus, the Chrome side panel, the mobile app) so a spoken turn, a\n * delegation the live voice model handed off, and an interim spoken update\n * read the same everywhere. The wire contract is the deep-agent host's\n * `additional_kwargs.metadata` (exposed as `message.metadata.custom`):\n *\n * - `source: \"voice\"` — a spoken turn; `voice_live: true` while it is still\n *   being heard or spoken (the ephemeral entry the durable row replaces);\n * - `source: \"text\"` — a message typed into the live conversation;\n * - `source: \"voice_delegation\"` (id prefix `voicetask_`) — the unit of work\n *   the live voice model delegated to the thread.\n */\nimport { getExternalStoreMessages, type ThreadMessage } from '@assistant-ui/core';\n\n/** `metadata.custom.source` of a spoken turn. */\nexport const VOICE_SOURCE_SPOKEN = 'voice';\n/** `metadata.custom.source` of a message typed into the live conversation. */\nexport const VOICE_SOURCE_TYPED = 'text';\n/** `metadata.custom.source` of a voice delegation's anchor message. */\nexport const VOICE_SOURCE_DELEGATION = 'voice_delegation';\n/** Id prefix of a voice delegation's anchor message (the working signal on\n * hosts that drop the command's metadata). */\nexport const VOICE_DELEGATION_ID_PREFIX = 'voicetask_';\n/** `metadata.custom` flag on a spoken turn still in flight. */\nexport const VOICE_LIVE_KEY = 'voice_live';\n/** `metadata.custom` list of the wire ids a grouped spoken bubble stands for\n * (see `groupSpokenUserTurns`), so id-keyed lookups such as checkpoint markers\n * can resolve any member to the bubble. */\nexport const VOICE_TURN_IDS_KEY = 'voice_turn_ids';\n\n/** The delegated run's interim-update tool (agora `send_voice_update`). */\nexport const VOICE_UPDATE_TOOL_NAME = 'send_voice_update';\n\n/** Trailing hint on the live-session strip, worded once for every surface. */\nexport const VOICE_LIVE_TYPED_HINT = 'typed messages go to the voice conversation';\n\nexport interface VoiceMessageMarker {\n  /** The thread message id; a host may not have assigned one yet. */\n  messageId: string | undefined;\n  /** The thread message's `metadata.custom`, when any. */\n  custom: Record<string, unknown> | undefined;\n}\n\n/** Whether a thread message is a voice delegation's anchor (the ask the live\n * voice model handed to the thread), by source or by id prefix. */\nexport function isVoiceDelegationMessage({ messageId, custom }: VoiceMessageMarker): boolean {\n  return (\n    custom?.source === VOICE_SOURCE_DELEGATION ||\n    (typeof messageId === 'string' && messageId.startsWith(VOICE_DELEGATION_ID_PREFIX))\n  );\n}\n\nexport interface VoiceProvenance {\n  /** Chip copy: \"Listening\", \"Speaking\", \"Spoken\", or \"Delegated from voice\". */\n  readonly label: string;\n  /** True while the turn is still being heard or spoken. */\n  readonly live: boolean;\n}\n\n// These values are read directly by useAuiState/useSyncExternalStore selectors.\n// Equal presentation must retain its identity, even when a transport replaces\n// the message metadata. Fresh objects here cause React's update-depth crash.\nconst SPOKEN_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Spoken',\n  live: false,\n});\nconst LISTENING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Listening',\n  live: true,\n});\nconst SPEAKING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Speaking',\n  live: true,\n});\nconst DELEGATED_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Delegated from voice',\n  live: false,\n});\n\n/**\n * The provenance chip for a message, or `null` for ordinary chat messages so\n * the chip is free to mount on every row. A spoken turn still in flight reads\n * as what is happening (\"Listening\" for the user, \"Speaking\" for the\n * assistant) rather than as a finished \"Spoken\" turn; a typed-in-voice turn\n * is a normal message and gets no chip. Results are immutable and stable for\n * equal presentation, so all surfaces can use this function as a selector.\n */\nexport function voiceProvenanceLabel({\n  messageId,\n  custom,\n  role,\n}: VoiceMessageMarker & { role: 'user' | 'assistant' | string }): VoiceProvenance | null {\n  const source = custom?.source;\n  if (source === VOICE_SOURCE_SPOKEN) {\n    if (custom?.[VOICE_LIVE_KEY] === true) {\n      return role === 'user' ? LISTENING_PROVENANCE : SPEAKING_PROVENANCE;\n    }\n    return SPOKEN_PROVENANCE;\n  }\n  if (isVoiceDelegationMessage({ messageId, custom })) {\n    return DELEGATED_PROVENANCE;\n  }\n  return null;\n}\n\nconst isRecord = (value: unknown): value is Record<string, unknown> =>\n  typeof value === 'object' && value !== null && !Array.isArray(value);\n\n/**\n * Whether a part of an assistant message holds the words of a spoken\n * assistant turn.\n *\n * The converter joins consecutive assistant entries into one message. The\n * host appends a live turn while it is heard, and a delegated run keeps\n * appending its own steps after it, so a turn spoken during a run becomes a\n * text part inside the run's message, before the run's later tool calls.\n * Grouped views put the text before the last tool call in the details panel\n * as the run's narration; they use this to keep the spoken words visible.\n *\n * The joined message's `metadata.custom` merges every entry in it, so this\n * reads the part's own entry, which the converter binds to the part.\n */\nexport function isSpokenTurnTextPart(part: { readonly type: string }): boolean {\n  if (part.type !== 'text') return false;\n  const entries: readonly unknown[] = getExternalStoreMessages<unknown>(\n    part as ThreadMessage['content'][number]\n  );\n  return entries.some((entry) => {\n    if (!isRecord(entry) || entry.role !== 'assistant' || !isRecord(entry.metadata)) return false;\n    const custom = entry.metadata.custom;\n    return isRecord(custom) && custom.source === VOICE_SOURCE_SPOKEN;\n  });\n}\n\nexport interface VoiceUpdateToolCall {\n  /** `mode: \"speak\"` — read aloud to the user; otherwise a silent progress note. */\n  spoken: boolean;\n  /** The update's words. */\n  text: string;\n  /** Whether the live session took it; `null` while the call is still running. */\n  delivered: boolean | null;\n}\n\n/**\n * What a `send_voice_update` tool call did, from its arguments and the tool's\n * teaching-text result (agora returns prose, not JSON, so the run can read it).\n */\nexport function describeVoiceUpdateToolCall({\n  args,\n  result,\n}: {\n  args: Record<string, unknown> | undefined;\n  result: unknown;\n}): VoiceUpdateToolCall {\n  const text = typeof args?.text === 'string' ? args.text.trim() : '';\n  const spoken = args?.mode === 'speak';\n  if (typeof result !== 'string' || result.trim() === '') {\n    return { spoken, text, delivered: null };\n  }\n  const outcome = result.toLowerCase();\n  const failed =\n    outcome.includes('could not be delivered') ||\n    outcome.includes('was not delivered') ||\n    outcome.includes('rejected the update') ||\n    outcome.includes('has ended') ||\n    outcome.includes('not configured') ||\n    outcome.includes('no voice session');\n  return { spoken, text, delivered: !failed };\n}\n\n/** The card's status line for a voice update in a given tool-call state. */\nexport function voiceUpdateStatusLabel(\n  call: VoiceUpdateToolCall,\n  status: 'running' | 'complete' | 'incomplete' | 'requires-action' | undefined\n): string {\n  if (status === 'running' || (call.delivered === null && status !== 'incomplete')) {\n    return call.spoken ? 'Saying to you…' : 'Noting progress…';\n  }\n  if (call.delivered === false || status === 'incomplete') {\n    return call.spoken ? 'Could not say this to you' : 'Progress note not delivered';\n  }\n  return call.spoken ? 'Said aloud' : 'Progress noted silently';\n}\n"],"names":["getExternalStoreMessages"],"mappings":";;;AAgBO,MAAM,sBAAsB;AAI5B,MAAM,0BAA0B;AAGhC,MAAM,6BAA6B;AAEnC,MAAM,iBAAiB;AAIvB,MAAM,qBAAqB;AAG3B,MAAM,yBAAyB;AAG/B,MAAM,wBAAwB;AAW9B,SAAS,yBAAyB,EAAE,WAAW,UAAuC;AAC3F,UACE,iCAAQ,YAAW,2BAClB,OAAO,cAAc,YAAY,UAAU,WAAW,0BAA0B;AAErF;AAuDA,MAAM,WAAW,CAAC,UAChB,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAgB9D,SAAS,qBAAqB,MAA0C;AAC7E,MAAI,KAAK,SAAS,OAAQ,QAAO;AACjC,QAAM,UAA8BA,KAAAA;AAAAA,IAClC;AAAA,EAAA;AAEF,SAAO,QAAQ,KAAK,CAAC,UAAU;AAC7B,QAAI,CAAC,SAAS,KAAK,KAAK,MAAM,SAAS,eAAe,CAAC,SAAS,MAAM,QAAQ,EAAG,QAAO;AACxF,UAAM,SAAS,MAAM,SAAS;AAC9B,WAAO,SAAS,MAAM,KAAK,OAAO,WAAW;AAAA,EAC/C,CAAC;AACH;AAeO,SAAS,4BAA4B;AAAA,EAC1C;AAAA,EACA;AACF,GAGwB;AACtB,QAAM,OAAO,QAAO,6BAAM,UAAS,WAAW,KAAK,KAAK,SAAS;AACjE,QAAM,UAAS,6BAAM,UAAS;AAC9B,MAAI,OAAO,WAAW,YAAY,OAAO,KAAA,MAAW,IAAI;AACtD,WAAO,EAAE,QAAQ,MAAM,WAAW,KAAA;AAAA,EACpC;AACA,QAAM,UAAU,OAAO,YAAA;AACvB,QAAM,SACJ,QAAQ,SAAS,wBAAwB,KACzC,QAAQ,SAAS,mBAAmB,KACpC,QAAQ,SAAS,qBAAqB,KACtC,QAAQ,SAAS,WAAW,KAC5B,QAAQ,SAAS,gBAAgB,KACjC,QAAQ,SAAS,kBAAkB;AACrC,SAAO,EAAE,QAAQ,MAAM,WAAW,CAAC,OAAA;AACrC;AAGO,SAAS,uBACd,MACA,QACQ;AACR,MAAI,WAAW,aAAc,KAAK,cAAc,QAAQ,WAAW,cAAe;AAChF,WAAO,KAAK,SAAS,mBAAmB;AAAA,EAC1C;AACA,MAAI,KAAK,cAAc,SAAS,WAAW,cAAc;AACvD,WAAO,KAAK,SAAS,8BAA8B;AAAA,EACrD;AACA,SAAO,KAAK,SAAS,eAAe;AACtC;;;;;;;;;;;;"}
+{"version":3,"file":"voice-presentation.cjs","sources":["../../../../agent-runtime/src/voice-presentation.ts"],"sourcesContent":["/**\n * How voice-mode transcript entries present, shared by every chat surface\n * (Olympus, the Chrome side panel, the mobile app) so a spoken turn, a\n * delegation the live voice model handed off, and an interim spoken update\n * read the same everywhere. The wire contract is the deep-agent host's\n * `additional_kwargs.metadata` (exposed as `message.metadata.custom`):\n *\n * - `source: \"voice\"` — a spoken turn; `voice_live: true` while it is still\n *   being heard or spoken (the ephemeral entry the durable row replaces);\n * - `source: \"text\"` — a message typed into the live conversation;\n * - `source: \"voice_delegation\"` (id prefix `voicetask_`) — the unit of work\n *   the live voice model delegated to the thread.\n */\nimport { getExternalStoreMessages, type ThreadMessage } from '@assistant-ui/core';\n\n/** `metadata.custom.source` of a spoken turn. */\nexport const VOICE_SOURCE_SPOKEN = 'voice';\n/** `metadata.custom.source` of a message typed into the live conversation. */\nexport const VOICE_SOURCE_TYPED = 'text';\n/** `metadata.custom.source` of a voice delegation's anchor message. */\nexport const VOICE_SOURCE_DELEGATION = 'voice_delegation';\n/** Id prefix of a voice delegation's anchor message (the working signal on\n * hosts that drop the command's metadata). */\nexport const VOICE_DELEGATION_ID_PREFIX = 'voicetask_';\n/** `metadata.custom` flag on a spoken turn still in flight. */\nexport const VOICE_LIVE_KEY = 'voice_live';\n/** `metadata.custom` list of the wire ids a grouped spoken bubble stands for\n * (see `groupSpokenUserTurns`), so id-keyed lookups such as checkpoint markers\n * can resolve any member to the bubble. */\nexport const VOICE_TURN_IDS_KEY = 'voice_turn_ids';\n\n/** The delegated run's interim-update tool (agora `send_voice_update`). */\nexport const VOICE_UPDATE_TOOL_NAME = 'send_voice_update';\n\n/** Trailing hint on the live-session strip, worded once for every surface. */\nexport const VOICE_LIVE_TYPED_HINT = 'typed messages go to the voice conversation';\n\nexport interface VoiceMessageMarker {\n  /** The thread message id; a host may not have assigned one yet. */\n  messageId: string | undefined;\n  /** The thread message's `metadata.custom`, when any. */\n  custom: Record<string, unknown> | undefined;\n}\n\n/** Whether a thread message is a voice delegation's anchor (the ask the live\n * voice model handed to the thread), by source or by id prefix. */\nexport function isVoiceDelegationMessage({ messageId, custom }: VoiceMessageMarker): boolean {\n  return (\n    custom?.source === VOICE_SOURCE_DELEGATION ||\n    (typeof messageId === 'string' && messageId.startsWith(VOICE_DELEGATION_ID_PREFIX))\n  );\n}\n\nexport interface VoiceProvenance {\n  /** Chip copy: \"Listening\", \"Speaking\", \"Spoken\", or \"Delegated from voice\". */\n  readonly label: string;\n  /** True while the turn is still being heard or spoken. */\n  readonly live: boolean;\n}\n\n// These values are read directly by useAuiState/useSyncExternalStore selectors.\n// Equal presentation must retain its identity, even when a transport replaces\n// the message metadata. Fresh objects here cause React's update-depth crash.\nconst SPOKEN_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Spoken',\n  live: false,\n});\nconst LISTENING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Listening',\n  live: true,\n});\nconst SPEAKING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Speaking',\n  live: true,\n});\nconst DELEGATED_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Delegated from voice',\n  live: false,\n});\n\n/**\n * The provenance chip for a message, or `null` for ordinary chat messages so\n * the chip is free to mount on every row. A spoken turn still in flight reads\n * as what is happening (\"Listening\" for the user, \"Speaking\" for the\n * assistant) rather than as a finished \"Spoken\" turn; a typed-in-voice turn\n * is a normal message and gets no chip. Results are immutable and stable for\n * equal presentation, so all surfaces can use this function as a selector.\n */\nexport function voiceProvenanceLabel({\n  messageId,\n  custom,\n  role,\n}: VoiceMessageMarker & { role: 'user' | 'assistant' | string }): VoiceProvenance | null {\n  const source = custom?.source;\n  if (source === VOICE_SOURCE_SPOKEN) {\n    if (custom?.[VOICE_LIVE_KEY] === true) {\n      return role === 'user' ? LISTENING_PROVENANCE : SPEAKING_PROVENANCE;\n    }\n    return SPOKEN_PROVENANCE;\n  }\n  if (isVoiceDelegationMessage({ messageId, custom })) {\n    return DELEGATED_PROVENANCE;\n  }\n  return null;\n}\n\nconst isRecord = (value: unknown): value is Record<string, unknown> =>\n  typeof value === 'object' && value !== null && !Array.isArray(value);\n\n/**\n * Whether a part of an assistant message holds the words of a spoken\n * assistant turn.\n *\n * The converter joins consecutive assistant entries into one message. The\n * host appends a live turn while it is heard, and a delegated run keeps\n * appending its own steps after it, so a turn spoken during a run becomes a\n * text part inside the run's message, before the run's later tool calls.\n * Grouped views put the text before the last tool call in the details panel\n * as the run's narration; they use this to keep the spoken words visible.\n *\n * The joined message's `metadata.custom` merges every entry in it, so this\n * reads the part's own entry, which the converter binds to the part.\n */\nexport function isSpokenTurnTextPart(part: { readonly type: string }): boolean {\n  if (part.type !== 'text') return false;\n  const entries: readonly unknown[] = getExternalStoreMessages<unknown>(\n    part as ThreadMessage['content'][number]\n  );\n  return entries.some((entry) => {\n    if (!isRecord(entry) || entry.role !== 'assistant' || !isRecord(entry.metadata)) return false;\n    const custom = entry.metadata.custom;\n    return isRecord(custom) && custom.source === VOICE_SOURCE_SPOKEN;\n  });\n}\n\nexport interface VoiceUpdateToolCall {\n  /** `mode: \"speak\"` — read aloud to the user; otherwise a silent progress note. */\n  spoken: boolean;\n  /** The update's words. */\n  text: string;\n  /** Whether the live session took it; `null` while the call is still running. */\n  delivered: boolean | null;\n}\n\n/**\n * What a `send_voice_update` tool call did, from its arguments and the tool's\n * teaching-text result (agora returns prose, not JSON, so the run can read it).\n */\nexport function describeVoiceUpdateToolCall({\n  args,\n  result,\n}: {\n  args: Record<string, unknown> | undefined;\n  result: unknown;\n}): VoiceUpdateToolCall {\n  const text = typeof args?.text === 'string' ? args.text.trim() : '';\n  const spoken = args?.mode === 'speak';\n  if (typeof result !== 'string' || result.trim() === '') {\n    return { spoken, text, delivered: null };\n  }\n  const outcome = result.toLowerCase();\n  const failed =\n    outcome.includes('could not be delivered') ||\n    outcome.includes('was not delivered') ||\n    outcome.includes('rejected the update') ||\n    outcome.includes('has ended') ||\n    outcome.includes('not configured') ||\n    outcome.includes('no voice session');\n  return { spoken, text, delivered: !failed };\n}\n\n/** The card's status line for a voice update in a given tool-call state. */\nexport function voiceUpdateStatusLabel(\n  call: VoiceUpdateToolCall,\n  status: 'running' | 'complete' | 'incomplete' | 'requires-action' | undefined\n): string {\n  if (status === 'running' || (call.delivered === null && status !== 'incomplete')) {\n    return call.spoken ? 'Saying to you…' : 'Noting progress…';\n  }\n  if (call.delivered === false || status === 'incomplete') {\n    return call.spoken ? 'Could not say this to you' : 'Progress note not delivered';\n  }\n  return call.spoken ? 'Said aloud' : 'Progress noted silently';\n}\n"],"names":["getExternalStoreMessages"],"mappings":";;;AAgBO,MAAM,sBAAsB;AAE5B,MAAM,qBAAqB;AAE3B,MAAM,0BAA0B;AAGhC,MAAM,6BAA6B;AAEnC,MAAM,iBAAiB;AAIvB,MAAM,qBAAqB;AAG3B,MAAM,yBAAyB;AAG/B,MAAM,wBAAwB;AAW9B,SAAS,yBAAyB,EAAE,WAAW,UAAuC;AAC3F,UACE,iCAAQ,YAAW,2BAClB,OAAO,cAAc,YAAY,UAAU,WAAW,0BAA0B;AAErF;AAuDA,MAAM,WAAW,CAAC,UAChB,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAgB9D,SAAS,qBAAqB,MAA0C;AAC7E,MAAI,KAAK,SAAS,OAAQ,QAAO;AACjC,QAAM,UAA8BA,KAAAA;AAAAA,IAClC;AAAA,EAAA;AAEF,SAAO,QAAQ,KAAK,CAAC,UAAU;AAC7B,QAAI,CAAC,SAAS,KAAK,KAAK,MAAM,SAAS,eAAe,CAAC,SAAS,MAAM,QAAQ,EAAG,QAAO;AACxF,UAAM,SAAS,MAAM,SAAS;AAC9B,WAAO,SAAS,MAAM,KAAK,OAAO,WAAW;AAAA,EAC/C,CAAC;AACH;AAeO,SAAS,4BAA4B;AAAA,EAC1C;AAAA,EACA;AACF,GAGwB;AACtB,QAAM,OAAO,QAAO,6BAAM,UAAS,WAAW,KAAK,KAAK,SAAS;AACjE,QAAM,UAAS,6BAAM,UAAS;AAC9B,MAAI,OAAO,WAAW,YAAY,OAAO,KAAA,MAAW,IAAI;AACtD,WAAO,EAAE,QAAQ,MAAM,WAAW,KAAA;AAAA,EACpC;AACA,QAAM,UAAU,OAAO,YAAA;AACvB,QAAM,SACJ,QAAQ,SAAS,wBAAwB,KACzC,QAAQ,SAAS,mBAAmB,KACpC,QAAQ,SAAS,qBAAqB,KACtC,QAAQ,SAAS,WAAW,KAC5B,QAAQ,SAAS,gBAAgB,KACjC,QAAQ,SAAS,kBAAkB;AACrC,SAAO,EAAE,QAAQ,MAAM,WAAW,CAAC,OAAA;AACrC;AAGO,SAAS,uBACd,MACA,QACQ;AACR,MAAI,WAAW,aAAc,KAAK,cAAc,QAAQ,WAAW,cAAe;AAChF,WAAO,KAAK,SAAS,mBAAmB;AAAA,EAC1C;AACA,MAAI,KAAK,cAAc,SAAS,WAAW,cAAc;AACvD,WAAO,KAAK,SAAS,8BAA8B;AAAA,EACrD;AACA,SAAO,KAAK,SAAS,eAAe;AACtC;;;;;;;;;;;;;"}

package/dist/agent-runtime/src/voice-presentation.js

@@ -1,5 +1,6 @@
 import { getExternalStoreMessages } from "@assistant-ui/core";
 const VOICE_SOURCE_SPOKEN = "voice";
+const VOICE_SOURCE_TYPED = "text";
 const VOICE_SOURCE_DELEGATION = "voice_delegation";
 const VOICE_DELEGATION_ID_PREFIX = "voicetask_";
 const VOICE_LIVE_KEY = "voice_live";
@@ -49,6 +50,7 @@ export {
   VOICE_LIVE_TYPED_HINT,
   VOICE_SOURCE_DELEGATION,
   VOICE_SOURCE_SPOKEN,
+  VOICE_SOURCE_TYPED,
   VOICE_TURN_IDS_KEY,
   VOICE_UPDATE_TOOL_NAME,
   describeVoiceUpdateToolCall,

package/dist/agent-runtime/src/voice-presentation.js.map

@@ -1 +1 @@
-{"version":3,"file":"voice-presentation.js","sources":["../../../../agent-runtime/src/voice-presentation.ts"],"sourcesContent":["/**\n * How voice-mode transcript entries present, shared by every chat surface\n * (Olympus, the Chrome side panel, the mobile app) so a spoken turn, a\n * delegation the live voice model handed off, and an interim spoken update\n * read the same everywhere. The wire contract is the deep-agent host's\n * `additional_kwargs.metadata` (exposed as `message.metadata.custom`):\n *\n * - `source: \"voice\"` — a spoken turn; `voice_live: true` while it is still\n *   being heard or spoken (the ephemeral entry the durable row replaces);\n * - `source: \"text\"` — a message typed into the live conversation;\n * - `source: \"voice_delegation\"` (id prefix `voicetask_`) — the unit of work\n *   the live voice model delegated to the thread.\n */\nimport { getExternalStoreMessages, type ThreadMessage } from '@assistant-ui/core';\n\n/** `metadata.custom.source` of a spoken turn. */\nexport const VOICE_SOURCE_SPOKEN = 'voice';\n/** `metadata.custom.source` of a message typed into the live conversation. */\nexport const VOICE_SOURCE_TYPED = 'text';\n/** `metadata.custom.source` of a voice delegation's anchor message. */\nexport const VOICE_SOURCE_DELEGATION = 'voice_delegation';\n/** Id prefix of a voice delegation's anchor message (the working signal on\n * hosts that drop the command's metadata). */\nexport const VOICE_DELEGATION_ID_PREFIX = 'voicetask_';\n/** `metadata.custom` flag on a spoken turn still in flight. */\nexport const VOICE_LIVE_KEY = 'voice_live';\n/** `metadata.custom` list of the wire ids a grouped spoken bubble stands for\n * (see `groupSpokenUserTurns`), so id-keyed lookups such as checkpoint markers\n * can resolve any member to the bubble. */\nexport const VOICE_TURN_IDS_KEY = 'voice_turn_ids';\n\n/** The delegated run's interim-update tool (agora `send_voice_update`). */\nexport const VOICE_UPDATE_TOOL_NAME = 'send_voice_update';\n\n/** Trailing hint on the live-session strip, worded once for every surface. */\nexport const VOICE_LIVE_TYPED_HINT = 'typed messages go to the voice conversation';\n\nexport interface VoiceMessageMarker {\n  /** The thread message id; a host may not have assigned one yet. */\n  messageId: string | undefined;\n  /** The thread message's `metadata.custom`, when any. */\n  custom: Record<string, unknown> | undefined;\n}\n\n/** Whether a thread message is a voice delegation's anchor (the ask the live\n * voice model handed to the thread), by source or by id prefix. */\nexport function isVoiceDelegationMessage({ messageId, custom }: VoiceMessageMarker): boolean {\n  return (\n    custom?.source === VOICE_SOURCE_DELEGATION ||\n    (typeof messageId === 'string' && messageId.startsWith(VOICE_DELEGATION_ID_PREFIX))\n  );\n}\n\nexport interface VoiceProvenance {\n  /** Chip copy: \"Listening\", \"Speaking\", \"Spoken\", or \"Delegated from voice\". */\n  readonly label: string;\n  /** True while the turn is still being heard or spoken. */\n  readonly live: boolean;\n}\n\n// These values are read directly by useAuiState/useSyncExternalStore selectors.\n// Equal presentation must retain its identity, even when a transport replaces\n// the message metadata. Fresh objects here cause React's update-depth crash.\nconst SPOKEN_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Spoken',\n  live: false,\n});\nconst LISTENING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Listening',\n  live: true,\n});\nconst SPEAKING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Speaking',\n  live: true,\n});\nconst DELEGATED_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Delegated from voice',\n  live: false,\n});\n\n/**\n * The provenance chip for a message, or `null` for ordinary chat messages so\n * the chip is free to mount on every row. A spoken turn still in flight reads\n * as what is happening (\"Listening\" for the user, \"Speaking\" for the\n * assistant) rather than as a finished \"Spoken\" turn; a typed-in-voice turn\n * is a normal message and gets no chip. Results are immutable and stable for\n * equal presentation, so all surfaces can use this function as a selector.\n */\nexport function voiceProvenanceLabel({\n  messageId,\n  custom,\n  role,\n}: VoiceMessageMarker & { role: 'user' | 'assistant' | string }): VoiceProvenance | null {\n  const source = custom?.source;\n  if (source === VOICE_SOURCE_SPOKEN) {\n    if (custom?.[VOICE_LIVE_KEY] === true) {\n      return role === 'user' ? LISTENING_PROVENANCE : SPEAKING_PROVENANCE;\n    }\n    return SPOKEN_PROVENANCE;\n  }\n  if (isVoiceDelegationMessage({ messageId, custom })) {\n    return DELEGATED_PROVENANCE;\n  }\n  return null;\n}\n\nconst isRecord = (value: unknown): value is Record<string, unknown> =>\n  typeof value === 'object' && value !== null && !Array.isArray(value);\n\n/**\n * Whether a part of an assistant message holds the words of a spoken\n * assistant turn.\n *\n * The converter joins consecutive assistant entries into one message. The\n * host appends a live turn while it is heard, and a delegated run keeps\n * appending its own steps after it, so a turn spoken during a run becomes a\n * text part inside the run's message, before the run's later tool calls.\n * Grouped views put the text before the last tool call in the details panel\n * as the run's narration; they use this to keep the spoken words visible.\n *\n * The joined message's `metadata.custom` merges every entry in it, so this\n * reads the part's own entry, which the converter binds to the part.\n */\nexport function isSpokenTurnTextPart(part: { readonly type: string }): boolean {\n  if (part.type !== 'text') return false;\n  const entries: readonly unknown[] = getExternalStoreMessages<unknown>(\n    part as ThreadMessage['content'][number]\n  );\n  return entries.some((entry) => {\n    if (!isRecord(entry) || entry.role !== 'assistant' || !isRecord(entry.metadata)) return false;\n    const custom = entry.metadata.custom;\n    return isRecord(custom) && custom.source === VOICE_SOURCE_SPOKEN;\n  });\n}\n\nexport interface VoiceUpdateToolCall {\n  /** `mode: \"speak\"` — read aloud to the user; otherwise a silent progress note. */\n  spoken: boolean;\n  /** The update's words. */\n  text: string;\n  /** Whether the live session took it; `null` while the call is still running. */\n  delivered: boolean | null;\n}\n\n/**\n * What a `send_voice_update` tool call did, from its arguments and the tool's\n * teaching-text result (agora returns prose, not JSON, so the run can read it).\n */\nexport function describeVoiceUpdateToolCall({\n  args,\n  result,\n}: {\n  args: Record<string, unknown> | undefined;\n  result: unknown;\n}): VoiceUpdateToolCall {\n  const text = typeof args?.text === 'string' ? args.text.trim() : '';\n  const spoken = args?.mode === 'speak';\n  if (typeof result !== 'string' || result.trim() === '') {\n    return { spoken, text, delivered: null };\n  }\n  const outcome = result.toLowerCase();\n  const failed =\n    outcome.includes('could not be delivered') ||\n    outcome.includes('was not delivered') ||\n    outcome.includes('rejected the update') ||\n    outcome.includes('has ended') ||\n    outcome.includes('not configured') ||\n    outcome.includes('no voice session');\n  return { spoken, text, delivered: !failed };\n}\n\n/** The card's status line for a voice update in a given tool-call state. */\nexport function voiceUpdateStatusLabel(\n  call: VoiceUpdateToolCall,\n  status: 'running' | 'complete' | 'incomplete' | 'requires-action' | undefined\n): string {\n  if (status === 'running' || (call.delivered === null && status !== 'incomplete')) {\n    return call.spoken ? 'Saying to you…' : 'Noting progress…';\n  }\n  if (call.delivered === false || status === 'incomplete') {\n    return call.spoken ? 'Could not say this to you' : 'Progress note not delivered';\n  }\n  return call.spoken ? 'Said aloud' : 'Progress noted silently';\n}\n"],"names":[],"mappings":";AAgBO,MAAM,sBAAsB;AAI5B,MAAM,0BAA0B;AAGhC,MAAM,6BAA6B;AAEnC,MAAM,iBAAiB;AAIvB,MAAM,qBAAqB;AAG3B,MAAM,yBAAyB;AAG/B,MAAM,wBAAwB;AAW9B,SAAS,yBAAyB,EAAE,WAAW,UAAuC;AAC3F,UACE,iCAAQ,YAAW,2BAClB,OAAO,cAAc,YAAY,UAAU,WAAW,0BAA0B;AAErF;AAuDA,MAAM,WAAW,CAAC,UAChB,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAgB9D,SAAS,qBAAqB,MAA0C;AAC7E,MAAI,KAAK,SAAS,OAAQ,QAAO;AACjC,QAAM,UAA8B;AAAA,IAClC;AAAA,EAAA;AAEF,SAAO,QAAQ,KAAK,CAAC,UAAU;AAC7B,QAAI,CAAC,SAAS,KAAK,KAAK,MAAM,SAAS,eAAe,CAAC,SAAS,MAAM,QAAQ,EAAG,QAAO;AACxF,UAAM,SAAS,MAAM,SAAS;AAC9B,WAAO,SAAS,MAAM,KAAK,OAAO,WAAW;AAAA,EAC/C,CAAC;AACH;AAeO,SAAS,4BAA4B;AAAA,EAC1C;AAAA,EACA;AACF,GAGwB;AACtB,QAAM,OAAO,QAAO,6BAAM,UAAS,WAAW,KAAK,KAAK,SAAS;AACjE,QAAM,UAAS,6BAAM,UAAS;AAC9B,MAAI,OAAO,WAAW,YAAY,OAAO,KAAA,MAAW,IAAI;AACtD,WAAO,EAAE,QAAQ,MAAM,WAAW,KAAA;AAAA,EACpC;AACA,QAAM,UAAU,OAAO,YAAA;AACvB,QAAM,SACJ,QAAQ,SAAS,wBAAwB,KACzC,QAAQ,SAAS,mBAAmB,KACpC,QAAQ,SAAS,qBAAqB,KACtC,QAAQ,SAAS,WAAW,KAC5B,QAAQ,SAAS,gBAAgB,KACjC,QAAQ,SAAS,kBAAkB;AACrC,SAAO,EAAE,QAAQ,MAAM,WAAW,CAAC,OAAA;AACrC;AAGO,SAAS,uBACd,MACA,QACQ;AACR,MAAI,WAAW,aAAc,KAAK,cAAc,QAAQ,WAAW,cAAe;AAChF,WAAO,KAAK,SAAS,mBAAmB;AAAA,EAC1C;AACA,MAAI,KAAK,cAAc,SAAS,WAAW,cAAc;AACvD,WAAO,KAAK,SAAS,8BAA8B;AAAA,EACrD;AACA,SAAO,KAAK,SAAS,eAAe;AACtC;"}
+{"version":3,"file":"voice-presentation.js","sources":["../../../../agent-runtime/src/voice-presentation.ts"],"sourcesContent":["/**\n * How voice-mode transcript entries present, shared by every chat surface\n * (Olympus, the Chrome side panel, the mobile app) so a spoken turn, a\n * delegation the live voice model handed off, and an interim spoken update\n * read the same everywhere. The wire contract is the deep-agent host's\n * `additional_kwargs.metadata` (exposed as `message.metadata.custom`):\n *\n * - `source: \"voice\"` — a spoken turn; `voice_live: true` while it is still\n *   being heard or spoken (the ephemeral entry the durable row replaces);\n * - `source: \"text\"` — a message typed into the live conversation;\n * - `source: \"voice_delegation\"` (id prefix `voicetask_`) — the unit of work\n *   the live voice model delegated to the thread.\n */\nimport { getExternalStoreMessages, type ThreadMessage } from '@assistant-ui/core';\n\n/** `metadata.custom.source` of a spoken turn. */\nexport const VOICE_SOURCE_SPOKEN = 'voice';\n/** `metadata.custom.source` of a message typed into the live conversation. */\nexport const VOICE_SOURCE_TYPED = 'text';\n/** `metadata.custom.source` of a voice delegation's anchor message. */\nexport const VOICE_SOURCE_DELEGATION = 'voice_delegation';\n/** Id prefix of a voice delegation's anchor message (the working signal on\n * hosts that drop the command's metadata). */\nexport const VOICE_DELEGATION_ID_PREFIX = 'voicetask_';\n/** `metadata.custom` flag on a spoken turn still in flight. */\nexport const VOICE_LIVE_KEY = 'voice_live';\n/** `metadata.custom` list of the wire ids a grouped spoken bubble stands for\n * (see `groupSpokenUserTurns`), so id-keyed lookups such as checkpoint markers\n * can resolve any member to the bubble. */\nexport const VOICE_TURN_IDS_KEY = 'voice_turn_ids';\n\n/** The delegated run's interim-update tool (agora `send_voice_update`). */\nexport const VOICE_UPDATE_TOOL_NAME = 'send_voice_update';\n\n/** Trailing hint on the live-session strip, worded once for every surface. */\nexport const VOICE_LIVE_TYPED_HINT = 'typed messages go to the voice conversation';\n\nexport interface VoiceMessageMarker {\n  /** The thread message id; a host may not have assigned one yet. */\n  messageId: string | undefined;\n  /** The thread message's `metadata.custom`, when any. */\n  custom: Record<string, unknown> | undefined;\n}\n\n/** Whether a thread message is a voice delegation's anchor (the ask the live\n * voice model handed to the thread), by source or by id prefix. */\nexport function isVoiceDelegationMessage({ messageId, custom }: VoiceMessageMarker): boolean {\n  return (\n    custom?.source === VOICE_SOURCE_DELEGATION ||\n    (typeof messageId === 'string' && messageId.startsWith(VOICE_DELEGATION_ID_PREFIX))\n  );\n}\n\nexport interface VoiceProvenance {\n  /** Chip copy: \"Listening\", \"Speaking\", \"Spoken\", or \"Delegated from voice\". */\n  readonly label: string;\n  /** True while the turn is still being heard or spoken. */\n  readonly live: boolean;\n}\n\n// These values are read directly by useAuiState/useSyncExternalStore selectors.\n// Equal presentation must retain its identity, even when a transport replaces\n// the message metadata. Fresh objects here cause React's update-depth crash.\nconst SPOKEN_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Spoken',\n  live: false,\n});\nconst LISTENING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Listening',\n  live: true,\n});\nconst SPEAKING_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Speaking',\n  live: true,\n});\nconst DELEGATED_PROVENANCE: VoiceProvenance = Object.freeze({\n  label: 'Delegated from voice',\n  live: false,\n});\n\n/**\n * The provenance chip for a message, or `null` for ordinary chat messages so\n * the chip is free to mount on every row. A spoken turn still in flight reads\n * as what is happening (\"Listening\" for the user, \"Speaking\" for the\n * assistant) rather than as a finished \"Spoken\" turn; a typed-in-voice turn\n * is a normal message and gets no chip. Results are immutable and stable for\n * equal presentation, so all surfaces can use this function as a selector.\n */\nexport function voiceProvenanceLabel({\n  messageId,\n  custom,\n  role,\n}: VoiceMessageMarker & { role: 'user' | 'assistant' | string }): VoiceProvenance | null {\n  const source = custom?.source;\n  if (source === VOICE_SOURCE_SPOKEN) {\n    if (custom?.[VOICE_LIVE_KEY] === true) {\n      return role === 'user' ? LISTENING_PROVENANCE : SPEAKING_PROVENANCE;\n    }\n    return SPOKEN_PROVENANCE;\n  }\n  if (isVoiceDelegationMessage({ messageId, custom })) {\n    return DELEGATED_PROVENANCE;\n  }\n  return null;\n}\n\nconst isRecord = (value: unknown): value is Record<string, unknown> =>\n  typeof value === 'object' && value !== null && !Array.isArray(value);\n\n/**\n * Whether a part of an assistant message holds the words of a spoken\n * assistant turn.\n *\n * The converter joins consecutive assistant entries into one message. The\n * host appends a live turn while it is heard, and a delegated run keeps\n * appending its own steps after it, so a turn spoken during a run becomes a\n * text part inside the run's message, before the run's later tool calls.\n * Grouped views put the text before the last tool call in the details panel\n * as the run's narration; they use this to keep the spoken words visible.\n *\n * The joined message's `metadata.custom` merges every entry in it, so this\n * reads the part's own entry, which the converter binds to the part.\n */\nexport function isSpokenTurnTextPart(part: { readonly type: string }): boolean {\n  if (part.type !== 'text') return false;\n  const entries: readonly unknown[] = getExternalStoreMessages<unknown>(\n    part as ThreadMessage['content'][number]\n  );\n  return entries.some((entry) => {\n    if (!isRecord(entry) || entry.role !== 'assistant' || !isRecord(entry.metadata)) return false;\n    const custom = entry.metadata.custom;\n    return isRecord(custom) && custom.source === VOICE_SOURCE_SPOKEN;\n  });\n}\n\nexport interface VoiceUpdateToolCall {\n  /** `mode: \"speak\"` — read aloud to the user; otherwise a silent progress note. */\n  spoken: boolean;\n  /** The update's words. */\n  text: string;\n  /** Whether the live session took it; `null` while the call is still running. */\n  delivered: boolean | null;\n}\n\n/**\n * What a `send_voice_update` tool call did, from its arguments and the tool's\n * teaching-text result (agora returns prose, not JSON, so the run can read it).\n */\nexport function describeVoiceUpdateToolCall({\n  args,\n  result,\n}: {\n  args: Record<string, unknown> | undefined;\n  result: unknown;\n}): VoiceUpdateToolCall {\n  const text = typeof args?.text === 'string' ? args.text.trim() : '';\n  const spoken = args?.mode === 'speak';\n  if (typeof result !== 'string' || result.trim() === '') {\n    return { spoken, text, delivered: null };\n  }\n  const outcome = result.toLowerCase();\n  const failed =\n    outcome.includes('could not be delivered') ||\n    outcome.includes('was not delivered') ||\n    outcome.includes('rejected the update') ||\n    outcome.includes('has ended') ||\n    outcome.includes('not configured') ||\n    outcome.includes('no voice session');\n  return { spoken, text, delivered: !failed };\n}\n\n/** The card's status line for a voice update in a given tool-call state. */\nexport function voiceUpdateStatusLabel(\n  call: VoiceUpdateToolCall,\n  status: 'running' | 'complete' | 'incomplete' | 'requires-action' | undefined\n): string {\n  if (status === 'running' || (call.delivered === null && status !== 'incomplete')) {\n    return call.spoken ? 'Saying to you…' : 'Noting progress…';\n  }\n  if (call.delivered === false || status === 'incomplete') {\n    return call.spoken ? 'Could not say this to you' : 'Progress note not delivered';\n  }\n  return call.spoken ? 'Said aloud' : 'Progress noted silently';\n}\n"],"names":[],"mappings":";AAgBO,MAAM,sBAAsB;AAE5B,MAAM,qBAAqB;AAE3B,MAAM,0BAA0B;AAGhC,MAAM,6BAA6B;AAEnC,MAAM,iBAAiB;AAIvB,MAAM,qBAAqB;AAG3B,MAAM,yBAAyB;AAG/B,MAAM,wBAAwB;AAW9B,SAAS,yBAAyB,EAAE,WAAW,UAAuC;AAC3F,UACE,iCAAQ,YAAW,2BAClB,OAAO,cAAc,YAAY,UAAU,WAAW,0BAA0B;AAErF;AAuDA,MAAM,WAAW,CAAC,UAChB,OAAO,UAAU,YAAY,UAAU,QAAQ,CAAC,MAAM,QAAQ,KAAK;AAgB9D,SAAS,qBAAqB,MAA0C;AAC7E,MAAI,KAAK,SAAS,OAAQ,QAAO;AACjC,QAAM,UAA8B;AAAA,IAClC;AAAA,EAAA;AAEF,SAAO,QAAQ,KAAK,CAAC,UAAU;AAC7B,QAAI,CAAC,SAAS,KAAK,KAAK,MAAM,SAAS,eAAe,CAAC,SAAS,MAAM,QAAQ,EAAG,QAAO;AACxF,UAAM,SAAS,MAAM,SAAS;AAC9B,WAAO,SAAS,MAAM,KAAK,OAAO,WAAW;AAAA,EAC/C,CAAC;AACH;AAeO,SAAS,4BAA4B;AAAA,EAC1C;AAAA,EACA;AACF,GAGwB;AACtB,QAAM,OAAO,QAAO,6BAAM,UAAS,WAAW,KAAK,KAAK,SAAS;AACjE,QAAM,UAAS,6BAAM,UAAS;AAC9B,MAAI,OAAO,WAAW,YAAY,OAAO,KAAA,MAAW,IAAI;AACtD,WAAO,EAAE,QAAQ,MAAM,WAAW,KAAA;AAAA,EACpC;AACA,QAAM,UAAU,OAAO,YAAA;AACvB,QAAM,SACJ,QAAQ,SAAS,wBAAwB,KACzC,QAAQ,SAAS,mBAAmB,KACpC,QAAQ,SAAS,qBAAqB,KACtC,QAAQ,SAAS,WAAW,KAC5B,QAAQ,SAAS,gBAAgB,KACjC,QAAQ,SAAS,kBAAkB;AACrC,SAAO,EAAE,QAAQ,MAAM,WAAW,CAAC,OAAA;AACrC;AAGO,SAAS,uBACd,MACA,QACQ;AACR,MAAI,WAAW,aAAc,KAAK,cAAc,QAAQ,WAAW,cAAe;AAChF,WAAO,KAAK,SAAS,mBAAmB;AAAA,EAC1C;AACA,MAAI,KAAK,cAAc,SAAS,WAAW,cAAc;AACvD,WAAO,KAAK,SAAS,8BAA8B;AAAA,EACrD;AACA,SAAO,KAAK,SAAS,eAAe;AACtC;"}