@vapi-ai/web 3.0.1 → 3.1.1

package/dist/vapi.js

@@ -58,6 +58,8 @@ function serializeError(error) {
     }
     return { message: String(error) };
 }
+// The app message Vapi sends when it couldn't start a server-mode recording.
+const VIDEO_RECORDING_START_FAILED_MESSAGE_TYPE = 'video-recording.start-failed';
 async function startAudioPlayer(player, track) {
     player.muted = false;
     player.autoplay = true;
@@ -88,13 +90,13 @@ function destroyAudioPlayer(participantId) {
     const player = document.querySelector(`audio[data-participant-id="${participantId}"]`);
     player?.remove();
 }
-function subscribeToTracks(e, call, isVideoRecordingEnabled, isVideoEnabled) {
+function subscribeToTracks(e, call, isVideoRecordingEnabled) {
     if (e.participant.local)
         return;
     call.updateParticipant(e.participant.session_id, {
         setSubscribedTracks: {
             audio: true,
-            video: isVideoRecordingEnabled || isVideoEnabled,
+            video: isVideoRecordingEnabled,
         },
     });
 }
@@ -136,6 +138,7 @@ class Vapi extends VapiEventEmitter {
     dailyCallObject = {};
     microphoneTrack = null;
     cancelMicrophonePreflight = null;
+    clientSideNoiseCancellationEnabled = true;
     async microphoneAcquire() {
         const source = this.dailyCallObject.audioSource;
         if (source && typeof source === 'object') {
@@ -278,6 +281,7 @@ class Vapi extends VapiEventEmitter {
             }
         });
         this.started = true;
+        this.clientSideNoiseCancellationEnabled = options?.clientSideNoiseCancellationEnabled ?? true;
         let failureStage = 'microphone-preflight';
         try {
             if (this.call) {
@@ -295,7 +299,9 @@ class Vapi extends VapiEventEmitter {
                 timestamp: new Date().toISOString()
             });
             const webCallStartTime = Date.now();
-            const webCall = (await client_1.client.call.callControllerCreateWebCall({
+            // An unset option stays undefined, so JSON.stringify drops the key and
+            // older servers never see it.
+            const createWebCallBody = {
                 assistant: typeof assistant === 'string' ? undefined : assistant,
                 assistantId: typeof assistant === 'string' ? assistant : undefined,
                 assistantOverrides,
@@ -305,7 +311,9 @@ class Vapi extends VapiEventEmitter {
                 workflowId: typeof workflow === 'string' ? workflow : undefined,
                 workflowOverrides,
                 roomDeleteOnUserLeaveEnabled: options?.roomDeleteOnUserLeaveEnabled,
-            })).data;
+                videoRecordingStartMode: options?.videoRecordingStartMode,
+            };
+            const webCall = (await client_1.client.call.callControllerCreateWebCall(createWebCallBody)).data;
             const webCallDuration = Date.now() - webCallStartTime;
             this.emit('call-start-progress', {
                 stage: 'web-call-creation',
@@ -319,12 +327,16 @@ class Vapi extends VapiEventEmitter {
                 }
             });
             const isVideoRecordingEnabled = webCall?.artifactPlan?.videoRecordingEnabled ?? false;
-            const isVideoEnabled = webCall?.assistant?.voice?.provider === 'tavus';
             // When video recording is enabled, the server sends a Daily meeting
-            // token that auto-starts the cloud recording on join — more reliable
-            // than calling startRecording() from a weak client network. Older
-            // servers don't send one; then the client starts the recording itself.
-            const callToken = webCall?.transport?.callToken;
+            // token. In client mode it auto-starts the cloud recording on join. In
+            // server mode it starts nothing: Vapi starts the recording from its
+            // servers once the user has joined, which a weak client network can't
+            // stop. Without a token the SDK emits recording-error and starts
+            // nothing.
+            const transport = webCall?.transport;
+            const callToken = transport?.callToken;
+            const isRecordingStartedByServer = transport?.videoRecordingStartMode === 'server';
+            const recordingTokenMissing = isVideoRecordingEnabled && !callToken;
             // Stage 2: Create Daily call object
             this.emit('call-start-progress', {
                 stage: 'daily-call-object-creation',
@@ -334,7 +346,6 @@ class Vapi extends VapiEventEmitter {
                     audioSource: this.dailyCallObject.audioSource ?? true,
                     videoSource: this.dailyCallObject.videoSource ?? isVideoRecordingEnabled,
                     isVideoRecordingEnabled,
-                    isVideoEnabled
                 }
             });
             const dailyCallStartTime = Date.now();
@@ -435,7 +446,7 @@ class Vapi extends VapiEventEmitter {
             this.call.on('participant-joined', (e) => {
                 if (!e || !this.call)
                     return;
-                subscribeToTracks(e, this.call, isVideoRecordingEnabled, isVideoEnabled);
+                subscribeToTracks(e, this.call, isVideoRecordingEnabled);
             });
             this.call.on('participant-updated', (e) => {
                 if (!e) {
@@ -449,6 +460,10 @@ class Vapi extends VapiEventEmitter {
                 }
                 this.detachAudioPlayer(e.participant.session_id);
             });
+            // Registered before join(), as in reconnect(): Vapi starts a server-mode
+            // recording when it sees the user join, so its start-failed message can
+            // arrive while join() is still in flight.
+            this.call.on('app-message', (e) => this.onAppMessage(e));
             // Registered before join(): a token-auto-started recording can begin
             // while join() is still in flight, so a listener added after join
             // could miss the event.
@@ -543,51 +558,29 @@ class Vapi extends VapiEventEmitter {
             }
             // Stage 5: Video recording setup (if enabled)
             if (isVideoRecordingEnabled && callToken) {
-                // The meeting token auto-started the recording at join, so there is
-                // nothing to request from the client.
+                // The meeting token auto-started the recording at join, or Vapi
+                // starts it from its servers, so there is nothing to request from the
+                // client.
                 this.emit('call-start-progress', {
                     stage: 'video-recording-setup',
                     status: 'completed',
                     timestamp: new Date().toISOString(),
-                    metadata: { action: 'auto-started-via-meeting-token' }
+                    metadata: {
+                        action: isRecordingStartedByServer
+                            ? 'started-by-server'
+                            : 'auto-started-via-meeting-token'
+                    }
                 });
             }
             else if (isVideoRecordingEnabled) {
+                // No token: nothing starts the recording. The recording-error is
+                // reported once start() has finished.
                 this.emit('call-start-progress', {
                     stage: 'video-recording-setup',
-                    status: 'started',
-                    timestamp: new Date().toISOString()
+                    status: 'failed',
+                    timestamp: new Date().toISOString(),
+                    metadata: { action: 'no-meeting-token' }
                 });
-                const recordingStartTime = Date.now();
-                try {
-                    recordingRequestedTime = new Date().getTime();
-                    this.startRecording();
-                    const recordingSetupDuration = Date.now() - recordingStartTime;
-                    this.emit('call-start-progress', {
-                        stage: 'video-recording-setup',
-                        status: 'completed',
-                        duration: recordingSetupDuration,
-                        timestamp: new Date().toISOString()
-                    });
-                }
-                catch (error) {
-                    const recordingSetupDuration = Date.now() - recordingStartTime;
-                    const serializedError = serializeError(error);
-                    this.emit('call-start-progress', {
-                        stage: 'video-recording-setup',
-                        status: 'failed',
-                        duration: recordingSetupDuration,
-                        timestamp: new Date().toISOString(),
-                        metadata: { error: serializedError.message }
-                    });
-                    this.emit('error', {
-                        type: 'video-recording-setup-error',
-                        stage: 'video-recording-setup',
-                        error: serializedError,
-                        timestamp: new Date().toISOString()
-                    });
-                    // Don't throw here, video recording is optional
-                }
             }
             else {
                 this.emit('call-start-progress', {
@@ -640,7 +633,6 @@ class Vapi extends VapiEventEmitter {
                 if (e)
                     this.handleLocalAudioLevel(e);
             });
-            this.call.on('app-message', (e) => this.onAppMessage(e));
             this.call.on('nonfatal-error', (e) => {
                 // https://docs.daily.co/reference/daily-js/events/meeting-events#type-audio-processor-error
                 if (e?.type === 'local-audio-level-observer-error') {
@@ -663,59 +655,66 @@ class Vapi extends VapiEventEmitter {
                     });
                 }
             });
-            // Stage 7: Audio processing setup
-            this.emit('call-start-progress', {
-                stage: 'audio-processing-setup',
-                status: 'started',
-                timestamp: new Date().toISOString()
-            });
-            const audioProcessingStartTime = Date.now();
-            try {
-                const audioProcessingUpdate = this.call.updateInputSettings({
-                    audio: {
-                        processor: {
-                            type: 'noise-cancellation',
-                        },
-                    },
-                });
-                audioProcessingUpdate.catch((error) => {
-                    this.emitAudioProcessingError('audio-processing-setup', error);
-                });
-                // The observer waits for this update rather than starting alongside the
-                // remote one above, so that Krisp has already swapped the microphone
-                // track. See maybeStartLocalAudioLevelObserver().
-                //
-                // A separate chain, not another link on the one above: EventEmitter
-                // rethrows out of emit('error') when a consumer registered no 'error'
-                // listener, and whether the observer starts must not hinge on that.
-                audioProcessingUpdate
-                    .catch(() => { })
-                    .then(() => this.maybeStartLocalAudioLevelObserver());
-                const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+            if (this.clientSideNoiseCancellationEnabled) {
+                // Stage 7: Audio processing setup
                 this.emit('call-start-progress', {
                     stage: 'audio-processing-setup',
-                    status: 'completed',
-                    duration: audioProcessingDuration,
+                    status: 'started',
                     timestamp: new Date().toISOString()
                 });
+                const audioProcessingStartTime = Date.now();
+                try {
+                    const audioProcessingUpdate = this.call.updateInputSettings({
+                        audio: {
+                            processor: {
+                                type: 'noise-cancellation',
+                            },
+                        },
+                    });
+                    audioProcessingUpdate.catch((error) => {
+                        this.emitAudioProcessingError('audio-processing-setup', error);
+                    });
+                    // The observer waits for this update rather than starting alongside the
+                    // remote one above, so that Krisp has already swapped the microphone
+                    // track. See maybeStartLocalAudioLevelObserver().
+                    //
+                    // A separate chain, not another link on the one above: EventEmitter
+                    // rethrows out of emit('error') when a consumer registered no 'error'
+                    // listener, and whether the observer starts must not hinge on that.
+                    audioProcessingUpdate
+                        .catch(() => { })
+                        .then(() => this.maybeStartLocalAudioLevelObserver());
+                    const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+                    this.emit('call-start-progress', {
+                        stage: 'audio-processing-setup',
+                        status: 'completed',
+                        duration: audioProcessingDuration,
+                        timestamp: new Date().toISOString()
+                    });
+                }
+                catch (error) {
+                    const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+                    const serializedError = serializeError(error);
+                    this.emit('call-start-progress', {
+                        stage: 'audio-processing-setup',
+                        status: 'failed',
+                        duration: audioProcessingDuration,
+                        timestamp: new Date().toISOString(),
+                        metadata: { error: serializedError.message }
+                    });
+                    this.emit('error', {
+                        type: 'audio-processing-setup-error',
+                        stage: 'audio-processing-setup',
+                        error: serializedError,
+                        timestamp: new Date().toISOString()
+                    });
+                    // Don't throw here, this is non-critical
+                }
             }
-            catch (error) {
-                const audioProcessingDuration = Date.now() - audioProcessingStartTime;
-                const serializedError = serializeError(error);
-                this.emit('call-start-progress', {
-                    stage: 'audio-processing-setup',
-                    status: 'failed',
-                    duration: audioProcessingDuration,
-                    timestamp: new Date().toISOString(),
-                    metadata: { error: serializedError.message }
-                });
-                this.emit('error', {
-                    type: 'audio-processing-setup-error',
-                    stage: 'audio-processing-setup',
-                    error: serializedError,
-                    timestamp: new Date().toISOString()
-                });
-                // Don't throw here, this is non-critical
+            else {
+                // Daily's default processor is 'none', so the microphone track is never
+                // swapped and the observer has nothing to wait for.
+                void this.maybeStartLocalAudioLevelObserver();
             }
             const totalDuration = Date.now() - startTime;
             this.emit('call-start-success', {
@@ -723,6 +722,9 @@ class Vapi extends VapiEventEmitter {
                 callId: webCall?.id || 'unknown',
                 timestamp: new Date().toISOString()
             });
+            if (recordingTokenMissing) {
+                this.recordingTokenMissingReportSchedule();
+            }
             return webCall;
         }
         catch (e) {
@@ -772,6 +774,12 @@ class Vapi extends VapiEventEmitter {
             else {
                 try {
                     const parsedMessage = JSON.parse(e.data);
+                    // Not a Vapi client message: re-emitted as recording-error, and kept
+                    // out of 'message' because its consumers switch on `type`.
+                    if (parsedMessage?.type === VIDEO_RECORDING_START_FAILED_MESSAGE_TYPE) {
+                        this.videoRecordingStartFailedReport(parsedMessage);
+                        return;
+                    }
                     this.emit('message', parsedMessage);
                     if (parsedMessage && 'type' in parsedMessage && 'status' in parsedMessage && parsedMessage.type === 'status-update' && parsedMessage.status === 'ended') {
                         this.hasEmittedCallEndedStatus = true;
@@ -939,6 +947,37 @@ class Vapi extends VapiEventEmitter {
             console.error(`[vapi] ${event} event listener threw`, error);
         }
     }
+    /**
+     * Reported in a later task, like Daily's own recording-error: start() or
+     * reconnect() has resolved and the caller's await continuation has run, so a
+     * listener registered right after it still hears this, and one that calls
+     * stop() can't interrupt call setup. Skipped if that call was stopped or
+     * replaced.
+     */
+    recordingTokenMissingReportSchedule() {
+        const call = this.call;
+        if (!call)
+            return;
+        setTimeout(() => {
+            if (!this.started || this.call !== call)
+                return;
+            this.emitToConsumer('recording-error', {
+                action: 'recording-error',
+                callClientId: call.callClientId,
+                errorMsg: 'Vapi returned no recording token, so the recording was not started. Call startRecording() to try starting it from the client.',
+            });
+        }, 0);
+    }
+    /** The fields are read defensively: the app message is untrusted JSON. */
+    videoRecordingStartFailedReport(message) {
+        const outcome = typeof message.outcome === 'string' ? message.outcome : 'unknown';
+        const error = typeof message.error === 'string' ? message.error : 'unknown';
+        this.emitToConsumer('recording-error', {
+            action: 'recording-error',
+            callClientId: this.call?.callClientId ?? '',
+            errorMsg: `Vapi couldn't start the recording (${outcome}): ${error}`,
+        });
+    }
     /** Drops the reference before removing, so nothing hands out a detached element. */
     detachAudioPlayer(participantId) {
         if (this.audioPlayer?.dataset.participantId === participantId) {
@@ -1133,10 +1172,10 @@ class Vapi extends VapiEventEmitter {
     /**
      * Reconnects to an active call.
      *
-     *
      * @param webCall
+     * @param options
      */
-    async reconnect(webCall) {
+    async reconnect(webCall, options) {
         const startTime = Date.now();
         if (this.started) {
             throw new Error('Cannot reconnect while a call is already in progress. Call stop() first.');
@@ -1144,6 +1183,9 @@ class Vapi extends VapiEventEmitter {
         if (!webCall.webCallUrl) {
             throw new Error('webCallUrl is required for reconnection.');
         }
+        if (options?.clientSideNoiseCancellationEnabled !== undefined) {
+            this.clientSideNoiseCancellationEnabled = options.clientSideNoiseCancellationEnabled;
+        }
         this.emit('call-start-progress', {
             stage: 'reconnect-initialization',
             status: 'started',
@@ -1171,11 +1213,14 @@ class Vapi extends VapiEventEmitter {
                 });
             }
             const isVideoRecordingEnabled = webCall?.artifactPlan?.videoRecordingEnabled ?? false;
-            const isVideoEnabled = webCall?.assistant?.voice?.provider === 'tavus';
-            // Same auto-start token as in start(). Rejoining with it while the
+            // Same token as in start(). In client mode, rejoining with it while the
             // recording is still running is safe; if the recording stopped when the
-            // user left, the token starts a new one.
+            // user left, the token starts a new one. In server mode it starts
+            // nothing. Without a token the SDK emits recording-error and starts
+            // nothing.
             const callToken = webCall?.transport?.callToken;
+            const isRecordingStartedByServer = webCall?.transport?.videoRecordingStartMode === 'server';
+            const recordingTokenMissing = isVideoRecordingEnabled && !callToken;
             // Stage 1: Create Daily call object
             this.emit('call-start-progress', {
                 stage: 'daily-call-object-creation',
@@ -1185,7 +1230,6 @@ class Vapi extends VapiEventEmitter {
                     audioSource: this.dailyCallObject.audioSource ?? true,
                     videoSource: this.dailyCallObject.videoSource ?? isVideoRecordingEnabled,
                     isVideoRecordingEnabled,
-                    isVideoEnabled
                 }
             });
             const dailyCallStartTime = Date.now();
@@ -1267,7 +1311,7 @@ class Vapi extends VapiEventEmitter {
             this.call.on('participant-joined', (e) => {
                 if (!e || !this.call)
                     return;
-                subscribeToTracks(e, this.call, isVideoRecordingEnabled, isVideoEnabled);
+                subscribeToTracks(e, this.call, isVideoRecordingEnabled);
             });
             this.call.on('participant-updated', (e) => {
                 if (!e) {
@@ -1384,45 +1428,29 @@ class Vapi extends VapiEventEmitter {
             });
             // Stage 4: Video recording setup (if enabled)
             if (isVideoRecordingEnabled && callToken) {
-                // The meeting token auto-started the recording at join, so there is
-                // nothing to request from the client.
+                // The meeting token auto-started the recording at join, or Vapi
+                // starts it from its servers, so there is nothing to request from the
+                // client.
                 this.emit('call-start-progress', {
                     stage: 'video-recording-setup',
                     status: 'completed',
                     timestamp: new Date().toISOString(),
-                    metadata: { action: 'auto-started-via-meeting-token' }
+                    metadata: {
+                        action: isRecordingStartedByServer
+                            ? 'started-by-server'
+                            : 'auto-started-via-meeting-token'
+                    }
                 });
             }
             else if (isVideoRecordingEnabled) {
+                // No token: nothing starts the recording. The recording-error is
+                // reported once reconnect() has finished.
                 this.emit('call-start-progress', {
                     stage: 'video-recording-setup',
-                    status: 'started',
-                    timestamp: new Date().toISOString()
+                    status: 'failed',
+                    timestamp: new Date().toISOString(),
+                    metadata: { action: 'no-meeting-token' }
                 });
-                const recordingStartTime = Date.now();
-                try {
-                    recordingRequestedTime = new Date().getTime();
-                    this.startRecording();
-                    const recordingSetupDuration = Date.now() - recordingStartTime;
-                    this.emit('call-start-progress', {
-                        stage: 'video-recording-setup',
-                        status: 'completed',
-                        duration: recordingSetupDuration,
-                        timestamp: new Date().toISOString()
-                    });
-                }
-                catch (error) {
-                    const recordingSetupDuration = Date.now() - recordingStartTime;
-                    const serializedError = serializeError(error);
-                    this.emit('call-start-progress', {
-                        stage: 'video-recording-setup',
-                        status: 'failed',
-                        duration: recordingSetupDuration,
-                        timestamp: new Date().toISOString(),
-                        metadata: { error: serializedError.message }
-                    });
-                    // Don't throw here, video recording is optional
-                }
             }
             else {
                 this.emit('call-start-progress', {
@@ -1461,53 +1489,60 @@ class Vapi extends VapiEventEmitter {
                 });
                 // Don't throw here, this is non-critical
             }
-            // Stage 6: Audio processing setup
-            this.emit('call-start-progress', {
-                stage: 'audio-processing-setup',
-                status: 'started',
-                timestamp: new Date().toISOString()
-            });
-            const audioProcessingStartTime = Date.now();
-            try {
-                const audioProcessingUpdate = this.call.updateInputSettings({
-                    audio: {
-                        processor: {
-                            type: 'noise-cancellation',
-                        },
-                    },
-                });
-                audioProcessingUpdate.catch((error) => {
-                    this.emitAudioProcessingError('audio-processing-setup', error);
-                });
-                // The observer waits for this update rather than starting alongside the
-                // remote one above, so that Krisp has already swapped the microphone
-                // track. See maybeStartLocalAudioLevelObserver().
-                //
-                // A separate chain, not another link on the one above: EventEmitter
-                // rethrows out of emit('error') when a consumer registered no 'error'
-                // listener, and whether the observer starts must not hinge on that.
-                audioProcessingUpdate
-                    .catch(() => { })
-                    .then(() => this.maybeStartLocalAudioLevelObserver());
-                const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+            if (this.clientSideNoiseCancellationEnabled) {
+                // Stage 6: Audio processing setup
                 this.emit('call-start-progress', {
                     stage: 'audio-processing-setup',
-                    status: 'completed',
-                    duration: audioProcessingDuration,
+                    status: 'started',
                     timestamp: new Date().toISOString()
                 });
+                const audioProcessingStartTime = Date.now();
+                try {
+                    const audioProcessingUpdate = this.call.updateInputSettings({
+                        audio: {
+                            processor: {
+                                type: 'noise-cancellation',
+                            },
+                        },
+                    });
+                    audioProcessingUpdate.catch((error) => {
+                        this.emitAudioProcessingError('audio-processing-setup', error);
+                    });
+                    // The observer waits for this update rather than starting alongside the
+                    // remote one above, so that Krisp has already swapped the microphone
+                    // track. See maybeStartLocalAudioLevelObserver().
+                    //
+                    // A separate chain, not another link on the one above: EventEmitter
+                    // rethrows out of emit('error') when a consumer registered no 'error'
+                    // listener, and whether the observer starts must not hinge on that.
+                    audioProcessingUpdate
+                        .catch(() => { })
+                        .then(() => this.maybeStartLocalAudioLevelObserver());
+                    const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+                    this.emit('call-start-progress', {
+                        stage: 'audio-processing-setup',
+                        status: 'completed',
+                        duration: audioProcessingDuration,
+                        timestamp: new Date().toISOString()
+                    });
+                }
+                catch (error) {
+                    const audioProcessingDuration = Date.now() - audioProcessingStartTime;
+                    const serializedError = serializeError(error);
+                    this.emit('call-start-progress', {
+                        stage: 'audio-processing-setup',
+                        status: 'failed',
+                        duration: audioProcessingDuration,
+                        timestamp: new Date().toISOString(),
+                        metadata: { error: serializedError.message }
+                    });
+                    // Don't throw here, this is non-critical
+                }
             }
-            catch (error) {
-                const audioProcessingDuration = Date.now() - audioProcessingStartTime;
-                const serializedError = serializeError(error);
-                this.emit('call-start-progress', {
-                    stage: 'audio-processing-setup',
-                    status: 'failed',
-                    duration: audioProcessingDuration,
-                    timestamp: new Date().toISOString(),
-                    metadata: { error: serializedError.message }
-                });
-                // Don't throw here, this is non-critical
+            else {
+                // Daily's default processor is 'none', so the microphone track is never
+                // swapped and the observer has nothing to wait for.
+                void this.maybeStartLocalAudioLevelObserver();
             }
             const totalDuration = Date.now() - startTime;
             this.emit('call-start-success', {
@@ -1516,6 +1551,9 @@ class Vapi extends VapiEventEmitter {
                 timestamp: new Date().toISOString()
             });
             this.emit('call-start');
+            if (recordingTokenMissing) {
+                this.recordingTokenMissingReportSchedule();
+            }
         }
         catch (e) {
             const totalDuration = Date.now() - startTime;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vapi-ai/web",
-  "version": "3.0.1",
+  "version": "3.1.1",
   "description": "",
   "main": "dist/vapi.js",
   "types": "dist/vapi.d.ts",