Release

.github/workflows/publish-packages.yml

@@ -111,8 +111,8 @@ jobs:
                   echo "📍 Submodule commit: $CURRENT_COMMIT"
 
                   # Check if this version exists on npm
-                  if npm view @answerai/aai-embed@$EMBED_VERSION version 2>/dev/null; then
-                    echo "✅ @answerai/aai-embed@$EMBED_VERSION is published on npm"
+                  if npm view aai-embed@$EMBED_VERSION version 2>/dev/null; then
+                    echo "✅ aai-embed@$EMBED_VERSION is published on npm"
 
                     # Optional: Verify git tag matches commit for extra safety
                     EXPECTED_TAG="v${EMBED_VERSION}"
@@ -132,13 +132,13 @@ jobs:
                     echo "is_published=true" >> $GITHUB_OUTPUT
                     echo "embed_version=$EMBED_VERSION" >> $GITHUB_OUTPUT
                   else
-                    echo "❌ @answerai/aai-embed@$EMBED_VERSION NOT found on npm"
+                    echo "❌ aai-embed@$EMBED_VERSION NOT found on npm"
                     echo ""
                     echo "🛑 Cannot publish aai-embed-react because aai-embed@$EMBED_VERSION hasn't been published yet."
                     echo ""
                     echo "Next steps:"
                     echo "1. Ensure aai-embed workflow ran successfully in the chat-embed repository"
-                    echo "2. Verify the package was published: npm view @answerai/aai-embed@$EMBED_VERSION"
+                    echo "2. Verify the package was published: npm view aai-embed@$EMBED_VERSION"
                     echo "3. If not published, trigger the workflow manually in the chat-embed repo"
 
                     echo "is_published=false" >> $GITHUB_OUTPUT
@@ -162,12 +162,19 @@ jobs:
             packages: write
             id-token: write
         steps:
+            - name: Generate GitHub App Token
+              id: generate-token
+              uses: tibdex/github-app-token@v1
+              with:
+                  app_id: ${{ secrets.STAGING_RESET_APP_ID }}
+                  private_key: ${{ secrets.STAGING_RESET_APP_PRIVATE_KEY }}
+
             - name: Checkout repository
               uses: actions/checkout@v4
               with:
                   fetch-depth: 0
                   submodules: recursive
-                  token: ${{ secrets.GITHUB_TOKEN }}
+                  token: ${{ steps.generate-token.outputs.token }}
 
             - name: Setup Node.js
               uses: actions/setup-node@v4
@@ -253,8 +260,7 @@ jobs:
             - name: Update package metadata
               working-directory: ./packages/embed-react
               run: |
-                  # Update package name to scoped name
-                  node -e "const pkg = require('./package.json'); pkg.name = '@answerai/aai-embed-react'; require('fs').writeFileSync('package.json', JSON.stringify(pkg, null, 2));"
+                  # Keep package name as-is (aai-embed-react)
 
                   # Add publishConfig
                   node -e "const pkg = require('./package.json'); pkg.publishConfig = { access: 'public' }; require('fs').writeFileSync('package.json', JSON.stringify(pkg, null, 2));"
@@ -309,6 +315,6 @@ jobs:
 
             - name: Summary
               run: |
-                  echo "✅ Published @answerai/aai-embed-react@${{ steps.version.outputs.new_version }} to npm"
+                  echo "✅ Published aai-embed-react@${{ steps.version.outputs.new_version }} to npm"
                   echo "✅ Peer dependency: aai-embed@^${{ steps.embed_version.outputs.embed_version }}"
                   echo "✅ Created tag aai-embed-react-v${{ steps.version.outputs.new_version }}"

packages-answers/ui/src/VideoCreator/VideoCreator.Client.tsx

@@ -178,7 +178,18 @@ const getProviderForModel = (model: string): Provider => {
     return 'openai'
 }
 
-const durationOptions = [4, 6, 8, 10, 12, 15]
+/**
+ * Provider-specific duration constraints based on provider API documentation.
+ * Frontend uses these discrete sets so users cannot pick unsupported durations.
+ */
+const getProviderDurationOptions = (provider: Provider): number[] => {
+    if (provider === 'openai') {
+        // OpenAI Sora models support only 4, 8, and 12 second videos.
+        return [4, 8, 12]
+    }
+    // Google Veo models support discrete duration values.
+    return [4, 6, 8]
+}
 
 const formatDateTime = (value?: string) => {
     if (!value) return 'Unknown date'
@@ -236,6 +247,7 @@ const VideoCreator = () => {
     const [dialogEmotion, setDialogEmotion] = useState('calm')
 
     const provider = useMemo(() => getProviderForModel(model), [model])
+    const durationOptions = useMemo(() => getProviderDurationOptions(provider), [provider])
     const jobPollers = useRef<Map<string, number>>(new Map())
 
     useEffect(() => {
@@ -333,6 +345,8 @@ const VideoCreator = () => {
     }
 
     useEffect(() => {
+        const validDurations = getProviderDurationOptions(provider)
+
         if (provider === 'google') {
             // Veo: 1080p only supports 16:9, 720p supports both 16:9 and 9:16
             setAspectRatio((current) => {
@@ -344,10 +358,16 @@ const VideoCreator = () => {
                 return current
             })
             setSize((current) => (current === '1280x720' || current === '1920x1080' ? current : '1280x720'))
-            setSeconds((current) => (durationOptions.includes(current) ? current : 8))
+            setSeconds((current) => (validDurations.includes(current) ? current : 8))
         } else {
+            // OpenAI: validate against strict constraints [4, 8, 12]
             setSize((current) => (current === '1280x720' || current === '1920x1080' ? current : '1280x720'))
-            setSeconds((current) => (durationOptions.includes(current) ? current : 8))
+            setSeconds((current) => {
+                if (validDurations.includes(current)) return current
+                // Find closest valid duration for OpenAI (4, 8, or 12)
+                const closest = validDurations.reduce((prev, curr) => (Math.abs(curr - current) < Math.abs(prev - current) ? curr : prev))
+                return closest
+            })
         }
         // eslint-disable-next-line react-hooks/exhaustive-deps
     }, [provider])
@@ -1087,9 +1107,12 @@ const VideoCreator = () => {
                             <Slider
                                 value={seconds}
                                 onChange={(_, newValue) => setSeconds(newValue as number)}
-                                min={4}
-                                max={15}
-                                step={1}
+                                min={Math.min(...durationOptions)}
+                                max={Math.max(...durationOptions)}
+                                step={null}
+                                marks={durationOptions.map((duration) => ({
+                                    value: duration
+                                }))}
                                 disabled={loading}
                                 size='small'
                                 sx={{ flexGrow: 1 }}

packages/server/src/aai-utils/billing/config.ts

@@ -42,9 +42,13 @@ export const BILLING_CONFIG = {
 
     // Sync configuration
     SYNC: {
-        LOOKBACK_DAYS: parseInt(process.env.BILLING_SYNC_LOOKBACK_DAYS || '90'),
-        PAGE_BATCH_SIZE: 15, // Number of pages to fetch in parallel
-        RATE_LIMIT_DELAY_MS: 1000 // Delay between page batches
+        LOOKBACK_DAYS: parseInt(process.env.BILLING_SYNC_LOOKBACK_DAYS || '7'), // Reduced from 90 to 7 days
+        PAGE_BATCH_SIZE: parseInt(process.env.BILLING_SYNC_PAGE_BATCH_SIZE || '3'), // Reduced from 15 to 3 pages
+        RATE_LIMIT_DELAY_MS: parseInt(process.env.BILLING_SYNC_RATE_LIMIT_MS || '2000'), // Increased from 1000 to 2000ms
+        TRACE_BATCH_SIZE: parseInt(process.env.BILLING_SYNC_TRACE_BATCH_SIZE || '5'), // Reduced from 15 to 5 traces
+        MAX_RETRIES: 3,
+        RETRY_DELAY_MS: 1000,
+        EXPONENTIAL_BACKOFF: true
     },
 
     // Resource configuration

packages/server/src/aai-utils/billing/langfuse/LangfuseProvider.ts

@@ -78,6 +78,7 @@ export class LangfuseProvider {
         page?: number
         userId?: string
         filter?: any[]
+        fields?: string
     }): Promise<any> {
         const queryParams: any = { ...params }
 
@@ -218,12 +219,14 @@ export class LangfuseProvider {
             })
 
             // Fetch and process first page
+            // Use minimal fields for initial filtering - exclude observations & scores for performance
             const initialResponse = await this.fetchTraces({
                 fromTimestamp: fromTimestamp.toISOString(),
                 toTimestamp: toTimestamp.toISOString(),
                 limit: 100,
                 page: 1,
-                filter: LangfuseProvider.UNPROCESSED_FILTER
+                filter: LangfuseProvider.UNPROCESSED_FILTER,
+                fields: 'core,metrics,io' // Exclude observations & scores - reduces payload by 80-90%
             })
             // console.log('initialResponse', initialResponse)
             const totalPages = initialResponse.meta.totalPages
@@ -312,7 +315,8 @@ export class LangfuseProvider {
                     toTimestamp: toTimestamp.toISOString(),
                     limit: 100,
                     page,
-                    filter: LangfuseProvider.UNPROCESSED_FILTER
+                    filter: LangfuseProvider.UNPROCESSED_FILTER,
+                    fields: 'core,metrics,io' // Exclude observations & scores - reduces payload by 80-90%
                 })
             )
         }
@@ -547,13 +551,15 @@ export class LangfuseProvider {
             startDate.setDate(startDate.getDate() - 30)
 
             // Fetch traces from Langfuse API with pagination
+            // Use minimal fields for usage events display - exclude observations to reduce load
             const langfuseResponse = await this.fetchTraces({
                 fromTimestamp: startDate.toISOString(),
                 toTimestamp: endDate.toISOString(),
                 limit,
                 page,
                 userId,
-                filter: LangfuseProvider.UNPROCESSED_FILTER
+                filter: LangfuseProvider.UNPROCESSED_FILTER,
+                fields: 'core,metrics,io' // Exclude observations & scores - faster response
                 // Note: We can't directly filter by customerId in the API call
                 // We'll filter the results after fetching
             })

packages/server/src/services/video-generator/index.ts

@@ -14,6 +14,9 @@ import { Blob } from 'node:buffer'
 const STORAGE_FOLDER = 'generated-videos'
 const OPENAI_MODELS = new Set(['sora-2', 'sora-2-pro'])
 const GOOGLE_MODELS = new Set(['veo-3.0-generate-001', 'veo-3.0-fast-generate-001'])
+// Duration constraints based on provider API specifications
+const OPENAI_VALID_DURATIONS = [4, 8, 12] // OpenAI Sora: discrete values only
+const GOOGLE_VALID_DURATIONS = [4, 6, 8] // Google Veo: supported discrete values
 const POLL_INTERVAL_MS = 5000
 const MAX_POLL_ATTEMPTS = 120 // ~10 minutes
 const JOB_RETENTION_MS = 24 * 60 * 60 * 1000 // keep completed jobs for 24 hours
@@ -148,6 +151,33 @@ const mapOpenAIStatus = (status?: string): JobStatus => {
     }
 }
 
+/**
+ * Validates and normalizes video duration based on provider constraints.
+ * Auto-corrects invalid values to the nearest supported duration instead of failing.
+ *
+ * @param seconds - Requested duration in seconds
+ * @param model - Video generation model name
+ * @returns Duration guaranteed to be accepted by the downstream provider
+ */
+const validateAndNormalizeDuration = (seconds: number | undefined, model: string): number => {
+    if (!seconds) return 8 // Works for both providers
+
+    const isOpenAI = OPENAI_MODELS.has(model)
+    const validDurations = isOpenAI ? OPENAI_VALID_DURATIONS : GOOGLE_VALID_DURATIONS
+
+    if (validDurations.includes(seconds)) {
+        return seconds
+    }
+
+    const normalized = validDurations.reduce((previous, current) =>
+        Math.abs(current - seconds) < Math.abs(previous - seconds) ? current : previous
+    )
+
+    console.warn(`[Video Generator] Adjusted duration from ${seconds}s to ${normalized}s for model ${model} (provider constraint)`)
+
+    return normalized
+}
+
 const getStorageType = (): string => {
     return process.env.STORAGE_TYPE ? process.env.STORAGE_TYPE : 'local'
 }
@@ -677,6 +707,9 @@ const startOpenAIJob = async (request: VideoGenerationRequest): Promise<VideoJob
         throw new InternalFlowiseError(StatusCodes.BAD_REQUEST, 'Prompt is required for OpenAI video generation')
     }
 
+    // Validate and normalize duration once for the entire request lifecycle
+    const validatedSeconds = validateAndNormalizeDuration(request.seconds, request.model)
+
     let creationResponse: OpenAIVideoStatus
 
     if (request.remixVideoProviderId) {
@@ -685,7 +718,7 @@ const startOpenAIJob = async (request: VideoGenerationRequest): Promise<VideoJob
         }
 
         if (request.size) body.size = request.size
-        if (request.seconds) body.seconds = request.seconds
+        if (validatedSeconds) body.seconds = validatedSeconds
 
         const remixResponse = await fetch(`https://api.openai.com/v1/videos/${request.remixVideoProviderId}/remix`, {
             method: 'POST',
@@ -711,8 +744,8 @@ const startOpenAIJob = async (request: VideoGenerationRequest): Promise<VideoJob
             formData.append('size', request.size)
         }
 
-        if (request.seconds) {
-            formData.append('seconds', request.seconds.toString())
+        if (validatedSeconds) {
+            formData.append('seconds', validatedSeconds.toString())
         }
 
         if (request.referenceImage?.base64) {
@@ -762,7 +795,8 @@ const startOpenAIJob = async (request: VideoGenerationRequest): Promise<VideoJob
     }
 
     registerJob(job)
-    setImmediate(() => processOpenAIJob(jobId, { ...request, prompt }, creationResponse))
+    // Pass normalized seconds for consistent telemetry/logging downstream
+    setImmediate(() => processOpenAIJob(jobId, { ...request, prompt, seconds: validatedSeconds }, creationResponse))
 
     return job
 }
@@ -1102,14 +1136,15 @@ const startGoogleJob = async (request: VideoGenerationRequest): Promise<VideoJob
     }
 
     const client = new GoogleGenAI({ apiKey })
+    const validatedSeconds = validateAndNormalizeDuration(request.seconds, request.model)
 
     const aspectRatio = request.aspectRatio || deriveAspectRatio(request.size)
     const params: GenerateVideosParameters = {
         model: request.model,
         prompt: request.prompt,
         config: {
             aspectRatio,
-            durationSeconds: request.seconds || 8,
+            durationSeconds: validatedSeconds,
             negativePrompt: request.negativePrompt
         }
     }
@@ -1145,7 +1180,8 @@ const startGoogleJob = async (request: VideoGenerationRequest): Promise<VideoJob
     }
 
     registerJob(job)
-    setImmediate(() => processGoogleJob(jobId, request, operation))
+    // Pass normalized seconds for consistent telemetry/logging downstream
+    setImmediate(() => processGoogleJob(jobId, { ...request, seconds: validatedSeconds }, operation))
 
     return job
 }