@vapi-ai/web 3.0.0 → 3.1.0

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,
         },
     });
 }
@@ -134,6 +136,39 @@ class Vapi extends VapiEventEmitter {
     speakingTimeout = null;
     dailyCallConfig = {};
     dailyCallObject = {};
+    microphoneTrack = null;
+    cancelMicrophonePreflight = null;
+    async microphoneAcquire() {
+        const source = this.dailyCallObject.audioSource;
+        if (source && typeof source === 'object') {
+            if (source.readyState === 'ended') {
+                throw new DOMException('The supplied microphone track has ended.', 'NotReadableError');
+            }
+            return source;
+        }
+        return new Promise((resolve, reject) => {
+            let cancelled = false;
+            this.cancelMicrophonePreflight = () => {
+                cancelled = true;
+                reject(new DOMException('Microphone preflight cancelled.', 'AbortError'));
+            };
+            navigator.mediaDevices.getUserMedia({
+                audio: typeof source === 'string' ? { deviceId: { exact: source } } : true,
+            }).then((stream) => {
+                const track = stream.getAudioTracks()[0];
+                if (cancelled || !track || track.readyState === 'ended') {
+                    stream.getTracks().forEach((item) => item.stop());
+                    reject(new DOMException('No live microphone track available.', 'NotFoundError'));
+                    return;
+                }
+                stream.getTracks().filter((item) => item !== track).forEach((item) => item.stop());
+                this.microphoneTrack = track;
+                resolve(track);
+            }, reject);
+        }).finally(() => {
+            this.cancelMicrophonePreflight = null;
+        });
+    }
     hasEmittedCallEndedStatus = false;
     constructor(apiToken, apiBaseUrl, dailyCallConfig, dailyCallObject) {
         super();
@@ -143,6 +178,8 @@ class Vapi extends VapiEventEmitter {
         this.dailyCallObject = (0, daily_guards_1.createSafeDailyFactoryOptions)(dailyCallObject);
     }
     async cleanup() {
+        this.microphoneTrack?.stop();
+        this.microphoneTrack = null;
         this.started = false;
         this.hasEmittedCallEndedStatus = false;
         if (this.call) {
@@ -243,7 +280,16 @@ class Vapi extends VapiEventEmitter {
             }
         });
         this.started = true;
+        let failureStage = 'microphone-preflight';
         try {
+            if (this.call) {
+                await this.cleanup();
+                this.started = true;
+            }
+            const audioSource = options?.microphonePreflightEnabled === false
+                ? this.dailyCallObject.audioSource ?? true
+                : await this.microphoneAcquire();
+            failureStage = 'unknown';
             // Stage 1: Create web call
             this.emit('call-start-progress', {
                 stage: 'web-call-creation',
@@ -251,7 +297,9 @@ 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,
@@ -261,7 +309,9 @@ 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',
@@ -274,22 +324,17 @@ class Vapi extends VapiEventEmitter {
                     voiceProvider: webCall?.assistant?.voice?.provider || 'unknown'
                 }
             });
-            if (this.call) {
-                this.emit('call-start-progress', {
-                    stage: 'daily-call-object-creation',
-                    status: 'started',
-                    timestamp: new Date().toISOString(),
-                    metadata: { action: 'cleanup-existing' }
-                });
-                await this.cleanup();
-            }
             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',
@@ -299,13 +344,12 @@ class Vapi extends VapiEventEmitter {
                     audioSource: this.dailyCallObject.audioSource ?? true,
                     videoSource: this.dailyCallObject.videoSource ?? isVideoRecordingEnabled,
                     isVideoRecordingEnabled,
-                    isVideoEnabled
                 }
             });
             const dailyCallStartTime = Date.now();
             try {
                 this.call = daily_js_1.default.createCallObject({
-                    audioSource: this.dailyCallObject.audioSource ?? true,
+                    audioSource,
                     videoSource: this.dailyCallObject.videoSource ?? isVideoRecordingEnabled,
                     dailyConfig: this.dailyCallConfig,
                 });
@@ -400,7 +444,7 @@ 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) {
@@ -414,6 +458,10 @@ 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.
@@ -508,51 +556,29 @@ 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', {
@@ -605,7 +631,6 @@ 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') {
@@ -688,13 +713,17 @@ class Vapi extends VapiEventEmitter {
                 callId: webCall?.id || 'unknown',
                 timestamp: new Date().toISOString()
             });
+            if (recordingTokenMissing) {
+                this.recordingTokenMissingReportSchedule();
+            }
             return webCall;
         }
         catch (e) {
             const totalDuration = Date.now() - startTime;
             const serializedError = serializeError(e);
+            await this.cleanup();
             this.emit('call-start-failed', {
-                stage: 'unknown',
+                stage: failureStage,
                 totalDuration,
                 error: serializedError.message,
                 errorStack: serializedError.stack || 'No stack trace available',
@@ -708,8 +737,10 @@ class Vapi extends VapiEventEmitter {
             });
             // Also emit the generic error event for backward compatibility
             this.emit('error', {
-                type: 'start-method-error',
-                stage: 'unknown',
+                type: failureStage === 'microphone-preflight'
+                    ? `microphone-preflight-${e && typeof e === 'object' && 'name' in e ? e.name : 'error'}`
+                    : 'start-method-error',
+                stage: failureStage,
                 error: serializedError,
                 totalDuration,
                 timestamp: new Date().toISOString(),
@@ -720,7 +751,6 @@ class Vapi extends VapiEventEmitter {
                     isMobile: this.isMobileDevice()
                 }
             });
-            await this.cleanup();
             return null;
         }
     }
@@ -735,6 +765,12 @@ 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;
@@ -778,6 +814,12 @@ class Vapi extends VapiEventEmitter {
      * If `roomDeleteOnUserLeaveEnabled` is set to `true`, the Vapi call will also be destroyed, preventing any reconnections.
      */
     async stop() {
+        if (this.cancelMicrophonePreflight) {
+            this.cancelMicrophonePreflight();
+            return;
+        }
+        this.microphoneTrack?.stop();
+        this.microphoneTrack = null;
         this.started = false;
         if (this.call) {
             await this.call.destroy();
@@ -896,6 +938,37 @@ 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) {
@@ -1128,11 +1201,14 @@ 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',
@@ -1142,7 +1218,6 @@ class Vapi extends VapiEventEmitter {
                     audioSource: this.dailyCallObject.audioSource ?? true,
                     videoSource: this.dailyCallObject.videoSource ?? isVideoRecordingEnabled,
                     isVideoRecordingEnabled,
-                    isVideoEnabled
                 }
             });
             const dailyCallStartTime = Date.now();
@@ -1224,7 +1299,7 @@ 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) {
@@ -1341,45 +1416,29 @@ 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', {
@@ -1473,6 +1532,9 @@ 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": "3.0.0",
+  "version": "3.1.0",
   "description": "",
   "main": "dist/vapi.js",
   "types": "dist/vapi.d.ts",
@@ -31,7 +31,7 @@
     "node": ">=22.14.0"
   },
   "dependencies": {
-    "@daily-co/daily-js": "^0.93.0",
+    "@daily-co/daily-js": "^0.94.0",
     "events": "^3.3.0"
   },
   "devDependencies": {