@layercode/js-sdk 2.8.4 → 2.8.6

package/dist/layercode-js-sdk.esm.js

@@ -5283,7 +5283,7 @@ class WavRecorder {
           .join(',');
       const cb = async () => {
         let id = ++lastId;
-        const devices = await this.listDevices();
+        const devices = await this.listDevices({ requestPermission: false });
         if (id === lastId) {
           if (serializeDevices(lastDevices) !== serializeDevices(devices)) {
             lastDevices = devices;
@@ -5336,14 +5336,22 @@ class WavRecorder {
   }
 
   /**
-   * List all eligible devices for recording, will request permission to use microphone
+   * List all eligible devices for recording.
+   *
+   * By default this will *not* request mic permission; labels may be empty until
+   * the user has granted permission via begin()/getUserMedia.
+   * Pass { requestPermission: true } to explicitly trigger a permission prompt.
+   *
+   * @param {{ requestPermission?: boolean }} [options]
    * @returns {Promise<Array<MediaDeviceInfo & {default: boolean}>>}
    */
-  async listDevices() {
+  async listDevices({ requestPermission = false } = {}) {
     if (!navigator.mediaDevices || !('enumerateDevices' in navigator.mediaDevices)) {
       throw new Error('Could not request user devices');
     }
-    await this.requestPermission();
+    if (requestPermission) {
+      await this.requestPermission();
+    }
 
     const devices = await navigator.mediaDevices.enumerateDevices();
     const audioDevices = devices.filter((device) => device.kind === 'audioinput');
@@ -5393,7 +5401,7 @@ class WavRecorder {
         config.audio.deviceId = { exact: deviceId };
       }
       this.stream = await navigator.mediaDevices.getUserMedia(config);
-      // Mark permission as granted so listDevices() won't call requestPermission() again
+      // Mark permission as granted so requestPermission() can skip expensive redundant getUserMedia calls
       this._hasPermission = true;
     } catch (err) {
       throw err;
@@ -5746,6 +5754,29 @@ function arrayBufferToBase64(arrayBuffer) {
 /* eslint-env browser */
 // import { env as ortEnv } from 'onnxruntime-web';
 // @ts-ignore - VAD package does not provide TypeScript types
+/**
+ * Layercode Web SDK notes / gotchas
+ *
+ * Modes:
+ * - audioInput=false => text-only mode. Do NOT touch getUserMedia or request mic permissions.
+ * - audioInput=true  => capture/send mic audio (may trigger permission prompt).
+ * - audioOutput=false => do not play audio, but MUST still send
+ *   `trigger.response.audio.replay_finished` once per assistant turn so the server can advance.
+ *
+ * Transcription:
+ * - trigger='push_to_talk' => no VAD; only send audio between triggerUserTurnStarted/Finished.
+ * - trigger='automatic'   => VAD drives userSpeaking + optional audio gating.
+ *
+ * Performance:
+ * - Mobile browsers e.g. iOS Safari media APIs can be slow; avoid multiple sequential getUserMedia calls.
+ *   This file starts the recorder first, then sets up device listeners.
+ *
+ * Compatibility:
+ * - This is consumed by the Layercode React SDK; treat public API/event shape changes as breaking.
+ *
+ * Assets:
+ * - VAD/ORT assets are loaded from assets.layercode.com (Cloudflare).
+ */
 const NOOP = () => { };
 const DEFAULT_WS_URL = 'wss://api.layercode.com/v1/agents/web/websocket';
 // SDK version - updated when publishing
@@ -5763,12 +5794,12 @@ const toLayercodeAudioInputDevice = (device) => {
     }
     return cloned;
 };
-const listAudioInputDevices = async () => {
+const listAudioInputDevices = async (options = {}) => {
     if (!hasMediaDevicesSupport()) {
         throw new Error('Media devices are not available in this environment');
     }
     const recorder = new WavRecorder({ sampleRate: DEFAULT_RECORDER_SAMPLE_RATE });
-    const devices = (await recorder.listDevices());
+    const devices = (await recorder.listDevices({ requestPermission: Boolean(options.requestPermission) }));
     return devices.map(toLayercodeAudioInputDevice);
 };
 const watchAudioInputDevices = (callback) => {
@@ -6202,27 +6233,34 @@ class LayercodeClient {
      * @param {ArrayBuffer} data - The audio data buffer
      */
     _handleDataAvailable(data) {
-        var _a, _b, _c;
+        var _a, _b, _c, _d;
         try {
-            const base64 = arrayBufferToBase64(data.mono);
-            // Don't send audio if muted
+            // Don't send or buffer audio if muted. Also clear any stale buffer so we
+            // don't accidentally flush old audio after unmute.
             if (this.isMuted) {
+                this.audioBuffer = [];
                 return;
             }
             // Determine if we should gate audio based on VAD configuration
-            const shouldGateAudio = ((_a = this.vadConfig) === null || _a === void 0 ? void 0 : _a.gate_audio) !== false; // Default to true if not specified
-            const bufferFrames = (_c = (_b = this.vadConfig) === null || _b === void 0 ? void 0 : _b.buffer_frames) !== null && _c !== void 0 ? _c : 10; // Default to 10 if not specified
+            const shouldGateAudio = ((_a = this.vadConfig) === null || _a === void 0 ? void 0 : _a.gate_audio) !== false; // default true
+            const bufferFrames = (_c = (_b = this.vadConfig) === null || _b === void 0 ? void 0 : _b.buffer_frames) !== null && _c !== void 0 ? _c : 10; // default 10
+            // If VAD is disabled or failed to init, gating would deadlock (userIsSpeaking never flips true).
+            // Only gate if we actually have a running VAD instance.
+            const vadEnabledByConfig = ((_d = this.vadConfig) === null || _d === void 0 ? void 0 : _d.enabled) !== false; // default true
+            const vadAvailable = vadEnabledByConfig && !!this.vad && !this.pushToTalkEnabled;
             let sendAudio;
             if (this.pushToTalkEnabled) {
                 sendAudio = this.pushToTalkActive;
             }
             else if (shouldGateAudio) {
-                sendAudio = this.userIsSpeaking;
+                // Key fix: if VAD isn't available, don't gate — send audio.
+                sendAudio = vadAvailable ? this.userIsSpeaking : true;
             }
             else {
                 // If gate_audio is false, always send audio
                 sendAudio = true;
             }
+            const base64 = arrayBufferToBase64(data.mono);
             if (sendAudio) {
                 // If we have buffered audio and we're gating, send it first
                 if (shouldGateAudio && this.audioBuffer.length > 0) {
@@ -6395,13 +6433,17 @@ class LayercodeClient {
                 this.recorderStarted = true;
                 this._sendReadyIfNeeded();
             }
-            const reportedDeviceId = (_a = this.activeDeviceId) !== null && _a !== void 0 ? _a : (this.useSystemDefaultDevice ? 'default' : ((_b = this.deviceId) !== null && _b !== void 0 ? _b : 'default'));
+            const reportedDeviceId = (_a = this.activeDeviceId) !== null && _a !== void 0 ? _a : (this.useSystemDefaultDevice ? 'default' : (_b = this.deviceId) !== null && _b !== void 0 ? _b : 'default');
             if (reportedDeviceId !== this.lastReportedDeviceId) {
                 this.lastReportedDeviceId = reportedDeviceId;
                 if (this.options.onDeviceSwitched) {
                     this.options.onDeviceSwitched(reportedDeviceId);
                 }
             }
+            // Ensure automatic mode has a VAD instance once the recorder stream is live
+            if (!this.vad && !this.pushToTalkEnabled) {
+                await this._initializeVAD();
+            }
             console.debug('Recorder started successfully with device:', reportedDeviceId);
         }
         catch (error) {
@@ -6700,8 +6742,8 @@ class LayercodeClient {
      * List all available audio input devices
      * @returns {Promise<Array<MediaDeviceInfo & {default: boolean}>>}
      */
-    async listDevices() {
-        return this.wavRecorder.listDevices();
+    async listDevices(options) {
+        return this.wavRecorder.listDevices(options);
     }
     /**
      * Switches the input device for the microphone and restarts recording
@@ -6728,7 +6770,7 @@ class LayercodeClient {
                 const newStream = this.wavRecorder.getStream();
                 await this._reinitializeVAD(newStream);
             }
-            const reportedDeviceId = (_c = (_b = this.lastReportedDeviceId) !== null && _b !== void 0 ? _b : this.activeDeviceId) !== null && _c !== void 0 ? _c : (this.useSystemDefaultDevice ? 'default' : (normalizedDeviceId !== null && normalizedDeviceId !== void 0 ? normalizedDeviceId : 'default'));
+            const reportedDeviceId = (_c = (_b = this.lastReportedDeviceId) !== null && _b !== void 0 ? _b : this.activeDeviceId) !== null && _c !== void 0 ? _c : (this.useSystemDefaultDevice ? 'default' : normalizedDeviceId !== null && normalizedDeviceId !== void 0 ? normalizedDeviceId : 'default');
             console.debug(`Successfully switched to input device: ${reportedDeviceId}`);
         }
         catch (error) {
@@ -6782,7 +6824,7 @@ class LayercodeClient {
                 this.recorderStarted = true;
                 this._sendReadyIfNeeded();
             }
-            const reportedDeviceId = (_a = this.activeDeviceId) !== null && _a !== void 0 ? _a : (this.useSystemDefaultDevice ? 'default' : ((_b = this.deviceId) !== null && _b !== void 0 ? _b : 'default'));
+            const reportedDeviceId = (_a = this.activeDeviceId) !== null && _a !== void 0 ? _a : (this.useSystemDefaultDevice ? 'default' : (_b = this.deviceId) !== null && _b !== void 0 ? _b : 'default');
             if (reportedDeviceId !== previousReportedDeviceId) {
                 this.lastReportedDeviceId = reportedDeviceId;
                 if (this.options.onDeviceSwitched) {
@@ -7027,11 +7069,7 @@ class LayercodeClient {
     async _shouldWarnAudioDevicesRequireUserGesture(error) {
         const e = error;
         const name = typeof (e === null || e === void 0 ? void 0 : e.name) === 'string' ? e.name : '';
-        const msg = typeof (e === null || e === void 0 ? void 0 : e.message) === 'string'
-            ? e.message
-            : typeof error === 'string'
-                ? error
-                : '';
+        const msg = typeof (e === null || e === void 0 ? void 0 : e.message) === 'string' ? e.message : typeof error === 'string' ? error : '';
         const isPermissionLike = name === 'NotAllowedError' || name === 'SecurityError' || name === 'PermissionDeniedError';
         if (!isPermissionLike)
             return false;