Changelog

ReelsBuilder v3 Beta — Changelog

This page tracks fixes shipped and known issues during the v3 beta period. It's updated frequently and focuses on user-visible changes.

View this changelog: Knowledge Base | Direct Link

If you hit an issue, please include: page URL, steps to reproduce, expected vs actual, screenshot, and timestamp.

Patch IDs: Each shipped update should include a Patch ID like

v3.0.YYYYMMDD.<shortsha>

2026-04-07 — V5 Launch

Patch ID: v5.0.20260407

Shipped — ReelsBuilder V5

V5 is a complete platform rebuild. Every part of the product — AI, generation quality, UI, and infrastructure — has been rewritten or significantly upgraded.

🧠 AI Chat Studio

• Conversational video creation — Describe what you want in plain language, ReelsBuilder builds it.
• Smart Wizard — 7-step onboarding inside chat with voice preview (12 voices with audio samples), soundtrack selection, and live brand defaults.
• Thread-based sessions — Each conversation saves to your history. Auto-generated titles appear in the sidebar after your first AI response.
• Structured suggestions — AI recommends content types, hooks, and angles based on your brand profile.
• Progress bar psychology — Named generation stages with animated progress (not a spinner).

🎬 Generation Quality

• V5 Backend Unification — All 7 generation engines rebuilt with shared quality standards: motivational, news, reddit story, AI video, UGC/ad, presenter, soothing.
• Production Quality Score (PQS) — Hard quality gate on Story and UGC engines. Low-quality outputs are blocked before delivery.
• V5 Manifest & lineage — Structured clip lineage tracks every asset used in a generation for edit and re-render accuracy.
• Broadcast Graphics Engine V9.5 — CNN/Bloomberg-style lower thirds, rolling tickers (stocks, crypto, news), Hormozi-style karaoke captions.
• ElevenLabs v3 + Scribe v2 — All TTS upgraded to eleven_v3 with loudness normalization and linear ducking. Word-level timestamps for karaoke sync.

🔄 Flows & Autopilot

• Flows — Chain generation steps into automated pipelines. Visual node editor.
• Autopilot — Set it and forget it. Automatically generates and schedules content based on your brand profile and posting cadence.
• AI Influencer — Full-stack HeyGen integration. Create a custom AI presenter for your brand.

📅 Auto-Poster

• Schedule Wizard — Simplified scheduling with brand-aware defaults. Voice samples, font previews, and topic pools from brand DNA.
• RSS → Video — Automatically convert RSS feed articles into video posts.
• Multi-platform publishing — YouTube, TikTok, Instagram, Facebook from one schedule.

🎨 Design & UI

• V5 design system — Full token compliance across all components. Orange palette, dark mode, consistent spacing.
• Studio deferred to V6 — Studio timeline editing is coming in V6 with blockchain-based content provenance. Hidden in V5 UI.
• Partners Hub — V5-polished partner/affiliate landing with motion and stats.

⚡ Infrastructure

• AWS migration — Moving off VPS to ECS-based worker fleet. Faster, more reliable, auto-scaling.
• Vector embeddings — 10K+ clips with semantic search. Content-aware clip selection (not random).
• Generation watchdog — Automatic timeout + credit refund for hung generations (30 min limit).

2025-12-17 - Update 15

Patch ID: v3.0.20251217.qa-fixes

Shipped - Voice Preview, Transcript Display & Video Quality Improvements

QA Fixes: Resolved critical issues discovered during QA testing including voice preview errors, missing transcripts, and video generation quality.

🎙️ Voice Preview System Fix

• Problem: All ElevenLabs preset voice previews returned 404 errors except the first voice.
• Root Cause: Outdated preview URLs in the voice configuration file.
• Fix: Updated

  web/config/elevenlabs-voices.ts

  with 16 verified voices and correct preview URLs fetched from the ElevenLabs API.
• Outcome: All voice previews now play correctly in the voice selector.

🔗 Custom Voice Button Fix

• Problem: "Create Custom Voice" button in Step 6 was non-functional.
• Fix: Wrapped button with Next.js Link component in

  web/components/dashboard/generation/voice-selector.tsx

  to navigate to

  /dashboard/media/voices

  .
• Outcome: Button now correctly navigates to the voice management page.

📝 Transcript Display Fix

• Problem: Video transcripts/scripts were not showing on the creations detail page.
• Root Cause: Story content was generated but not saved to video metadata.
• Fix: Added

  transcript: result.story

  to video metadata in:

  • server/routes/reddit-video-routes.ts

  • server/lib/services/post-scheduler-service.ts

• Outcome: New videos will display full transcript/script on the creations page.

🎵 Audio Fade-Out Enhancement

• Problem: Background music ended abruptly when video finished, creating jarring audio experience.
• Fix: Added 2-second fade-out filter to audio mixing in

  server/lib/reddit-stories/video-processor.ts

  .
• Outcome: Smooth, professional audio ending for all generated videos.

⏱️ Minimum Video Duration Enforcement

• Problem: Some videos were only 7 seconds long due to extremely short generated stories.
• Fix: Added minimum 1500 character story length validation in

  server/lib/reddit-stories/story-generator.ts

  . Stories under this threshold now throw an error, trigger credit refund, and notify the user.
• Outcome: Prevents sub-30-second videos from being generated.

📊 Sentry Error Monitoring

• Enhancement: Configured Sentry for both web and processing server to capture errors with full context.
• Outcome: Real-time error tracking and debugging capabilities.

🔧 Calendar Route Syntax Fix

• Problem: Build failed due to orphan code in calendar route.
• Fix: Removed misplaced code in

  web/app/api/beta/calendar/route.ts

  .
• Outcome: Build completes successfully.

Files Changed

web/config/elevenlabs-voices.ts

web/components/dashboard/generation/voice-selector.tsx

server/routes/reddit-video-routes.ts

server/lib/services/post-scheduler-service.ts

server/lib/reddit-stories/video-processor.ts

server/lib/reddit-stories/story-generator.ts

web/app/api/beta/calendar/route.ts

2025-12-16 - Update 14

Patch ID: v3.0.20251216.fixes-enhancements

Shipped - Data Accuracy, UI Refinements & Autoposter AI Enhancements

Critical Fixes & Strategic Enhancements: Addressed key data accuracy issues, refined UI presentation, and introduced AI-powered dynamic content generation for Autoposter.

📊 Calendar Data Synchronization

• Problem: Inaccurate schedule counts and potential data drift in the content calendar.
• Fix: Refined the recurrence logic within

  web/app/api/beta/calendar/route.ts

  to ensure each scheduled post is accurately calculated based on the schedule's recurring configuration, eliminating heuristic-based advancement.
• Monitoring: Integrated Sentry monitoring into

  calculateNextExecutionFrom

  to flag any schedules that fail to generate a next execution time, providing critical visibility for debugging data consistency.
• Outcome: Improved real-time schedule accuracy and better data validation.

💳 Batch Generation Credit Validation

• Problem: Users reported being unable to afford videos in batch generation despite having sufficient credits.
• Fix: Enhanced

  web/app/api/beta/batch-generation/route.ts

  with explicit session and

  UserModel

  validation. Sentry logging was added to capture instances where

  session.user.id

  is missing or the

  UserModel

  cannot be found, ensuring the credit calculation operates on reliable data.
• Outcome: Accurate video generation count estimates and precise capacity planning for users.

🖼️ UI Rendering Issue Resolved

• Problem: "Buy Credits" and "Notifications" popovers appeared blurred, detracting from the user experience.
• Fix: Applied explicit

  bg-popover

  and

  z-50

  utility classes to

  PopoverContent

  elements in

  web/components/dashboard/header.tsx

  and

  web/components/dashboard/HeaderCreditsSales.tsx

  .
• Outcome: All UI elements, including popovers, are now rendered clearly and crisply.

🤖 Autoposter Dynamic Content Generation

• Problem: Autoposter titles and descriptions were static, limiting content variation and engagement opportunities.
• Enhancement: Implemented a new "Dynamic AI Titles & Descriptions" toggle in the Autoposter's publishing step (

  web/components/dashboard/auto-poster/create-schedule-dialog.tsx

  ). When enabled, this flag instructs the system to dynamically generate variations using AI.
• Schema Update:

  dynamic_content: Boolean

  field added to

  posting_config

  in

  web/models/PostScheduleModel.ts

  and

  server/models/PostScheduleModel.ts

  , and incorporated into

  zod

  schemas (

  web/app/api/post-schedule/route.ts

  ,

  web/app/api/post-schedule/[scheduleId]/route.ts

  ).
• Outcome: Enables varied, AI-driven titles/descriptions for each post, boosting engagement potential.

🎨 Autopilot Dashboard UI Optimization

• Problem: Excessive green coloration in the dashboard header and inefficient spacing with new modules.
• Fix: Toned down the green accents in

  web/app/dashboard/autopilot/components/DashboardHeader.tsx

  for a more balanced visual hierarchy. The

  ControlPanel.tsx

  grid layout was adjusted to

  xl:grid-cols-4

  to improve density and reduce whitespace on larger screens.
• Outcome: A cleaner, more professional, and efficiently spaced Autopilot Dashboard.

⚠️ Setup Progress Banner Fix

• Problem: The "Setup Progress 100%" banner popped up even after setup was completed.
• Fix: Updated

  web/app/dashboard/autopilot/components/SetupBanner.tsx

  to explicitly check the

  onboardingCompleted

  status from the user's brand profile.
• Outcome: The banner now correctly hides once the setup is complete.

Files Changed

web/app/api/beta/calendar/route.ts

web/app/api/beta/batch-generation/route.ts

web/components/dashboard/header.tsx

web/components/dashboard/HeaderCreditsSales.tsx

web/components/dashboard/auto-poster/create-schedule-dialog.tsx

web/models/PostScheduleModel.ts

dynamic_content

server/models/PostScheduleModel.ts

dynamic_content

web/app/api/post-schedule/route.ts

createScheduleSchema

dynamic_content

web/app/api/post-schedule/[scheduleId]/route.ts

updateScheduleSchema

dynamic_content

web/app/dashboard/autopilot/components/DashboardHeader.tsx

web/app/dashboard/autopilot/components/SetupBanner.tsx

2025-12-16 - Update 13

Patch ID: v3.0.20251216.model-update

Shipped - Frontier AI Model Upgrade & Robust Fallback Systems

Strategic Upgrade: All AI models have been updated to their latest frontier versions, and robust, cross-provider fallback systems are now in place for enhanced reliability and quality.

Frontier AI Model Deployment

• ElevenLabs Voice:
  • Upgraded from

    eleven_multilingual_v2

    to

    eleven_v3

    (Ultra Quality) as the default and recommended model across all relevant services and presets. This ensures the most emotionally rich and human-like voice synthesis.
• OpenAI LLMs:
  • All

    gpt-4o

    usages updated to

    gpt-5.2

    (Latest Flagship) for critical analysis, creative tasks, and primary routing.
  • All

    gpt-4o-mini

    usages updated to

    gpt-5-mini

    (Cost-Effective) for conversation, structured output, and budget-tier tasks.
• Anthropic LLMs:
  • Default model for creative tasks updated to

    claude-4.5-sonnet

    .
• Google LLMs:
  • Default model for Google provider updated to

    gemini-3-pro

    for superior long-context tasks.
• OpenRouter Fallbacks:
  • Default model updated to

    gpt-5-mini

    for consistency with other budget-tier models.

Robust Fallback Systems

• AI Router (Web):
  • Cross-Provider Failover: The AI router now dynamically retries failed requests with alternative providers (e.g., OpenAI -> Anthropic -> Google) to maximize success rates.
  • Sentry Alerting: Critical Sentry alerts are triggered upon provider failures, tagged

    fallback_triggered

    , ensuring immediate visibility into service instability.
• ElevenLabs Voice Service (Server):
  • Model-Level Fallback: If

    eleven_v3

    experiences an error, the system automatically falls back to

    eleven_multilingual_v2

    for voice generation, preventing complete service disruption. Sentry alerts are configured for these events.

QA & Content Quality Insights

• Smart Audio Ducking Implemented: Refactored audio mixing in

  server/lib/motivational/utils/audio-utils.ts

  to use FFmpeg Sidechain Compression (Auto-Ducking), ensuring professional voice-over clarity.
• Acknowledged Blindspot: Audio Sync: The reliance on AssemblyAI for word-level timestamps is noted as a source of latency and cost. Future optimization will investigate direct timestamp integration with ElevenLabs if API allows.
• Acknowledged Blindspot: Visual WYSIWYG: Disabling Render Previews due to server stability concerns.

Files Changed

server/lib/elevenlabs.ts

server/lib/motivational/utils/content-generation.ts

server/lib/services/trend-service.ts

server/lib/ai-story-generator.ts

web/lib/ai-router.ts

web/app/api/ai/trending-topics/route.ts

web/app/api/ai/generate-article/route.ts

web/app/api/tools/text-to-speech/route.ts

ds-bot/scripts/daily-post.ts

server/lib/motivational/utils/audio-utils.ts

2025-12-16 - Update 12

Patch ID: v3.0.20251216.safe-trends

Shipped - Trend Injection & Autopilot Safety

Strategic Upgrade: Transformed Autopilot from a functional execution engine into a brand-aligned growth engine with strict safety guardrails.

Brand-Aligned Trend Injection

• New

  TrendService

  : Fetches trends specifically tailored to your brand's Industry, Tone, and Audience (no more "Gen-Z" trends for "Corporate Law" brands).
• Multi-Dimensional Trends: Now fetches a

  TrendBundle

  containing:
  • Topic: Specific angle (e.g., "The 'Monk Mode' trend")
  • Hook: Viral hook template adapted to your tone
  • Audio: Trending audio mood/style
  • Format: Recommended video structure
• Deep Integration:

  MotivationalVideoGenerator

  now injects these trends directly into the script generation prompt, ensuring content is timely and relevant.

Autopilot Safety & Validation

• "AI Glitch" Prevention: Updated template engine to throw an error if any placeholders (like

  {type}

  ) remain unreplaced, preventing "t - {type} #1" posts.
• Pre-Flight Validation: Scheduler now checks video duration against platform-specific rules (e.g., <60s for Shorts) before upload.
• Audit Logging: New

  PostingLogModel

  tracks every attempt with detailed status and error messages.
• Connection Testing: Added "Test Connection" button to social account settings to verify tokens without posting.

UX & Onboarding Polish

• Consolidated Onboarding: Removed legacy paths; single source of truth at

  /dashboard/autopilot/onboarding

  .
• Guided Tour: New interactive tour for first-time users.
• Onboarding Checklist: Progress bar in the header to track setup (Profile -> Socials -> Content -> Autopilot).
• Quick Settings: New header popover for toggling content types and settings without leaving the dashboard.
• Performance: Implemented lazy loading for heavy dashboard components (LCP improvement).

Files Changed

server/lib/services/trend-service.ts

server/lib/motivational/video-generator.ts

server/lib/services/post-scheduler-service.ts

web/app/dashboard/autopilot/page.tsx

web/app/dashboard/autopilot/components/DashboardHeader.tsx

2025-12-16 - Update 11

Patch ID: v3.0.20251216.pending

Shipped - CRITICAL: Autopilot Opt-In Fix

Breaking Change: Autopilot now requires explicit opt-in. Users must manually enable auto-posting rather than having it default to ON.

The Problem

When users completed onboarding, the "Auto-Post Content" toggle defaulted to

true

active

• 3 auto-poster schedules to start running on server restart
• Users unknowingly having content auto-posted to their social accounts
• No explicit consent before automated posting began

The Fix

Changed

enableAutopilot

true

false

// Before (opt-out model - BAD)
const [enableAutopilot, setEnableAutopilot] = useState(true);

// After (opt-in model - GOOD)
const [enableAutopilot, setEnableAutopilot] = useState(false);

User Experience Now

1. Toggle starts OFF - "Schedules will be created but paused"
2. User must explicitly turn ON to enable auto-posting
3. Button shows "Complete Setup" when OFF, "Launch Autopilot" when ON
4. Existing users with active schedules are NOT affected (schedules remain active)

Files Changed

app/dashboard/autopilot/onboarding/page.tsx

enableAutopilot

true

false

2025-12-16 - Update 10

Patch ID: v3.0.20251216.pending

Shipped - Error Handling & Image Loading Fixes

Fixed infinite loading states, image 400 errors, and chunk load error handling across the dashboard.

Infinite Loading States Fixed

• Video Detail Page (

  app/dashboard/creations/[id]/page.tsx

  ) — Now shows proper "Video Not Found" error UI instead of blank page when video doesn't exist or fetch fails

/_next/image 400 Errors Fixed

Social profile pictures from external CDNs were causing 400 errors. Two fixes applied:

• ControlPanel.tsx — Changed

  next/image

  to native

  <img>

  for social account profile pictures (URLs come from various external CDNs)
• next.config.ts — Added remotePatterns for common social platform CDNs:

yt3.ggpht.com

yt3.googleusercontent.com

lh3.googleusercontent.com

*.cdninstagram.com

*.fbcdn.net

*.tiktokcdn.com

*.tiktokcdn-us.com

avatars.githubusercontent.com

gravatar.com

ui-avatars.com

Chunk Load Error Handling

After deployments, users with cached JS chunks would see cryptic errors. Now both error boundaries detect chunk errors and show friendly "App Updated!" UI:

• global-error.tsx — Added

  isChunkLoadError()

  detection, special refresh UI, cache clearing
• error.tsx — Same chunk error handling with styled Card UI
• Both skip Sentry reporting for expected chunk errors

Files Changed

app/dashboard/creations/[id]/page.tsx

app/beta/dashboard/components/ControlPanel.tsx

next.config.ts

app/global-error.tsx

app/error.tsx

2025-12-16 - Update 9

Patch ID: v3.0.20251216.c52a795

Shipped - Beta same-origin API + fewer 502s

• Beta now prefers same-origin API base (

  /api/v1

  ) for developer API links/spec downloads (reduces CORS surface).
• Caddy beta upstreams force IPv4 (

  127.0.0.1

  ) to eliminate intermittent

  localhost -> ::1

  502s impacting Autopilot (Past Creations/Recent Runs/Performance).

2025-12-16 - Update 8

Patch ID: v3.0.20251216.2d73df3

Shipped - Changelog patch ID system

• /api/changelog

  now returns a

  patchId

  and the

  /changelog

  header displays it.

• /knowledge-base/changelog

  now shows the current patch ID and has timeout/error handling (no infinite "Loading...").

• npm run build

  now auto-sets

  NEXT_PUBLIC_V3_PATCH_ID

  (date + git short SHA) for consistent public patch labeling.

2025-12-16 — Update 7

Shipped — Autopilot Dashboard Mobile Responsiveness

Complete mobile/responsive audit and overhaul of the Autopilot dashboard for better user experience on all devices.

AI Concierge Component (Critical Fix)

The Concierge was reported as having "poor formatting/scaling". Fixed issues:

• Fixed heights now scale with viewport (

  max-h-[60vh]

  on mobile vs

  max-h-[500px]

  on desktop)
• User avatars now show proper initials instead of cramped full text
• Quick suggestions changed from horizontal scroll to 2-column grid on mobile
• Message bubbles properly sized (

  max-w-[75%]

  mobile,

  max-w-[80%]

  desktop)
• Input field placeholder shortened ("Ask Concierge..." instead of full text)
• All spacing/padding now responsive with

  sm:

  breakpoints

Dashboard Layout Improvements

• Mobile-first Concierge placement — Now appears at TOP on mobile (was buried at bottom)
• Desktop sticky sidebar — Concierge stays visible while scrolling (

  sticky top-4

  )
• Loading skeleton — Now responsive grid (

  grid-cols-2 sm:grid-cols-3 lg:grid-cols-4

  )
• Container padding — Responsive (

  px-3 sm:px-4

  )
• All grid gaps — Responsive (

  gap-3 sm:gap-4

  )
• Error/welcome states — Mobile-friendly with responsive sizing

Component Updates

Concierge.tsx

page.tsx

DashboardHeader.tsx

ControlPanel.tsx

StatsGrid.tsx

UpcomingContent.tsx

CreditsUsage.tsx

Mobile UX Highlights

• Concierge accessible immediately on mobile (no scrolling required)
• Touch targets properly sized for mobile interaction
• Text truncation prevents overflow on small screens
• Badges and icons scale appropriately
• All dialogs properly sized for mobile viewports

2025-12-16 — Update 6

Shipped — Social Accounts Overhaul + Dedicated Socials Page

Complete rebuild of the Social Accounts experience with full multi-platform support and account management.

Social Accounts Dialog (ControlPanel.tsx)

Previously only showed YouTube. Now supports all 8 platforms with organized tabs:

Account Management Features

• Connect accounts via OAuth for each platform
• Set Primary account per platform (used for autopilot posting)
• Disconnect accounts with confirmation
• Status badges: Active/Inactive, Primary indicator
• Profile display: Profile picture, display name, username

New Socials Page (

/dashboard/autopilot/socials

)

Dedicated full-page management for social accounts:

• Added "Socials" to Autopilot navigation (after Runs)
• Platform cards grouped by category (Video/Social/Community)
• Per-platform account listing with management dropdown
• Refresh button + error states + loading skeletons

Bug Fixes

• AI Concierge Insights — Fixed "Unable to load insights" error. API returns

  { error: false }

  not

  { success: true }

  ; updated check in

  ProactiveInsights.tsx

• SocialAccount type — Extended interface with

  is_primary

  ,

  display_name

  ,

  analytics_enabled

  ,

  last_synced_at

  fields

Files Changed

app/dashboard/autopilot/components/ControlPanel.tsx

app/dashboard/autopilot/components/AutopilotNav.tsx

app/dashboard/autopilot/components/ProactiveInsights.tsx

app/dashboard/autopilot/types.ts

app/dashboard/autopilot/socials/page.tsx

2025-12-15 (Late Night) — Update 5

Shipped — Autopilot Scheduling + Connection Resilience

• Scheduling cadence now supports Custom posts/day (e.g., 5x/day) and regenerates “Beta Auto-*” schedules accordingly.
• Autopilot Schedule UI now includes a Posts/day input for Custom cadence and a link to advanced timing configuration.
• Social Accounts no longer show as “not connected” when the accounts fetch fails; the dashboard now shows a retryable error state instead.
• Build stability (Windows VPS):

  npm run build

  no longer wipes

  .next

  by default; set

  NEXT_BUILD_CLEAN=true

  if you need a full clean build.

2025-12-15 (Evening) — Update 4

Shipped — Autopilot Data Reliability

• Performance Insights now pulls from real user video history (fixed a Mongo aggregation filter mismatch that could return empty data).
• Recent Runs now falls back to scheduled-generation videos when run logging isn’t available yet.
• Schedule Updates (frequency changes) now refresh dashboard data immediately so the UI doesn’t look “stuck”.
• Calendar now includes paused schedules (and can resume them) instead of silently omitting them.
• Social Accounts API now has an upstream timeout + fast failures to prevent hanging requests.

2025-12-15 (Night) — Update 3

Shipped — Knowledge Base Changelog Integration

Public Changelog in Knowledge Base — New searchable, filterable changelog page integrated into the Knowledge Base.

Features

• Timeline view with expandable entries
• Search across all updates
• Filter by type (Features, Improvements, Fixes)
• Stats dashboard (features shipped, improvements, fixes)
• Tag-based categorization (AI, Cost Optimization, Quality, etc.)

Location

• /knowledge-base/changelog

  — Full featured changelog with search/filter

• /changelog

  — Simple markdown view (still available)

Technical

• New API endpoint:

  /api/changelog

  serves CHANGELOG.md as JSON
• Added "Changelog" category to Knowledge Base navigation
• Dynamic parsing of markdown into structured timeline entries

2025-12-15 (Night) — Update 2

Shipped — Full Dashboard AI Supervision Rollout

Comprehensive Audit & Implementation — Extended supervisor agent patterns to ALL content generation endpoints.

New Ensemble Endpoints (Quality Boost)

/api/tools/script-generator

/api/tools/story-generator

/api/beta/repurpose

New Gated Endpoints (Cost Savings)

/api/beta/niche

/api/beta/smart-defaults

/api/beta/generate-preview

Model Upgrades

/api/beta/chat-onboarding

Documentation Added

• docs/AI_SUPERVISION_AUDIT.md

  — Full endpoint analysis with ROI calculations
• Updated

  docs/SUPERVISOR_AGENT_ANALYSIS.md

  — Marked as implemented

Total Impact Summary

2025-12-15 (Night)

Shipped — Supervisor Agent System

Intelligent AI Supervision — New cost-optimization layer that uses cheap models first and only escalates to premium models when needed.

Two Supervision Patterns

Gated Supervision (Cost Savings)

Analysis tasks now run through confidence gating:

Gemini Flash ($0.075/1M) → Confidence Check
                              ↓
                    ≥70% → Ship immediately (big savings!)
                    <70% → Escalate to GPT-5.2 ($5/1M)

• Viral Score: Gemini Flash first → GPT-5.2 if uncertain
• Content Profile: GPT-5-mini first → GPT-5.2 if uncertain
• ~70% of requests complete without escalation

Ensemble Supervision (Quality Boost)

Creative tasks now use multi-model synthesis:

GPT-5-mini ───┐
              ├──► Claude 4.5 Sonnet synthesizes best elements
Gemini Flash ─┘

• Hook Generation: 2 workers → Claude synthesis = better viral hooks
• Combines diverse outputs into superior final result

New Functions Added

• chatGated()

  — Confidence-based escalation

• chatEnsemble()

  — Multi-model synthesis

• chatSmart()

  — Auto-selects best strategy

• evaluateConfidence()

  — Output quality assessment

Affected Endpoints (Initial)

• /api/beta/viral-score

  — Now uses gated supervision

• /api/beta/generate-content-profile

  — Now uses gated supervision

• /api/beta/hooks/generate

  — Now uses ensemble for premium quality

Expected Impact

2025-12-15 (Evening)

Shipped — AI Infrastructure Upgrade

Multi-Provider AI Router v3.0 — Complete overhaul of AI model routing for December 2025 frontier models.

New Model Routing Strategy

Cost Optimization

• Conversation tasks: Now use GPT-5-mini ($0.40/1M input) instead of GPT-4o ($2.50/1M) — 84% cost reduction
• Hashtag generation: Now use Gemini 2.5 Flash ($0.075/1M) — 97% cost reduction
• Analysis tasks: Upgraded from DeepSeek to GPT-5.2 for better reliability

Affected Features

• /autopilot

  — Viral score analysis now uses GPT-5.2

• /autopilot

  — AI Concierge now uses GPT-5-mini (faster responses)

• /autopilot/onboarding

  — Content profile generation uses GPT-5.2

• /generate

  — Hook generation uses Claude 4.5 Sonnet
• All hashtag endpoints now use Gemini 2.5 Flash

Voice System Update

• ElevenLabs upgraded to

  eleven_v3

  (Alpha) model
• New quality tiers: budget, standard, high, ultra
• Better emotional expression and naturalness

2025-12-15

Shipped

• Added a public

  /changelog

  page and linked it in the footer.
• Stabilized Next.js production builds on the Windows VPS (higher heap + clean

  .next

  before build).
• Fixed build-breaking duplicate exports (

  web/lib/ai-router.ts

  ,

  web/lib/elevenlabs.ts

  ).
• Autopilot analytics sync now decrypts stored tokens and only marks sync as fresh when snapshots exist.
• Autopilot modules now use timeouts + error states (Runs, Calendar, Viral Growth, AI Insights) to prevent infinite spinners.

Known issues

• TikTok analytics are limited (API scope restrictions) and Instagram analytics sync isn’t implemented yet.

• AutomationRun

  documents aren’t being written yet; “Recent Runs” is derived from schedule post history for now.

Next

• Write

  AutomationRun

  records for each schedule execution (with post URLs + error details).
• Add an in-app "Report a bug" flow that includes request ID + page context.

V2 → V3 Migration — Major Release

What's New in V3

ReelsBuilder V3 represents a complete platform overhaul with intelligent automation, multi-provider AI, and enterprise-grade monitoring. This release transforms ReelsBuilder from a video generation tool into a full content automation platform.

1. Autopilot Dashboard (New)

A completely new dashboard experience for hands-off content creation.

Core Features

• Brand Profile Onboarding — 4-step flow: website analysis → brand details → social accounts → posting schedule
• AI Content Concierge — Conversational AI assistant that modifies your settings directly
• Intelligent Scheduling — AI-predicted viral posting times based on audience analysis
• Content Approval Workflow — Review and approve content before publishing
• A/B Testing Framework — Test different content variations automatically
• Batch Video Generation — Generate multiple videos at once
• Performance Analytics — Track engagement, views, and growth metrics

New Routes

/dashboard/autopilot

/dashboard/autopilot/onboarding

/dashboard/autopilot/approvals

/dashboard/autopilot/calendar

/dashboard/autopilot/performance

/dashboard/autopilot/ab-tests

/dashboard/autopilot/batch

/dashboard/autopilot/runs

2. Intelligence Pipeline (New)

AI-powered personalization that learns from your interactions.

UserIntelligenceModel

• Aggregates data from 5 sources (website scrape, brand profile, concierge chats, user actions, video performance)
• Tracks preferences, behavioral signals, and engagement patterns
• Calculates churn risk, engagement score, and upsell readiness

IntelligentPromptBuilder

• Generates hyper-personalized content prompts
• Intelligence score (0-100) indicates personalization depth
• Incorporates brand voice, topics, guidelines, and user preferences

Preference Extraction

• Mines concierge conversations for likes/dislikes
• Tracks content type correlations with engagement
• Identifies optimal posting times per user

3. Multi-Provider AI Router

Complete overhaul of AI infrastructure with December 2025 frontier models.

Supported Providers

Task-Based Routing

Supervisor Agent System

• Gated Pattern: Cheap model first → escalate if low confidence (40-50% cost savings)
• Ensemble Pattern: Multiple workers → premium synthesizer (+15-20% quality)

4. Voice System Upgrade

ElevenLabs integration upgraded to latest models.

New Models

eleven_v3

eleven_flash_v2_5

eleven_turbo_v2_5

Quality Tiers

• Ultra: eleven_v3 with best settings
• High: eleven_multilingual_v2
• Standard: eleven_flash_v2_5
• Budget: eleven_turbo_v2_5

5. Social Media Expansion

New platform integrations for wider reach.

New Platforms

• X/Twitter — Post videos and threads
• LinkedIn — Professional content posting
• Telegram Channels — Automated channel updates

Enhanced Integrations

• Instagram — Reels + Stories + Posts
• TikTok — Videos with trending sounds
• YouTube — Shorts + long-form support
• Facebook — Reels + standard posts

6. New Content Types

Expanded content creation capabilities.

Video Types

• Long-form videos — Extended content support
• Stock clips integration — Access to stock footage
• Image posting — Static image content

AI Generation

• AI Image Generation — Create images for posts
• AI Article Generation — Blog content creation
• Content Repurposing — Adapt content across platforms

7. Monitoring & Reliability

Enterprise-grade error tracking and performance monitoring.

Sentry Integration

• Client, server, and edge SDK configurations
• Session replay (100% on errors)
• Structured API errors with request IDs
• Error boundaries for graceful degradation

Web Vitals Monitoring

Performance Monitor

• Database query tracking
• External API call monitoring
• Custom metrics with thresholds
• React component render tracking

8. Email System

Personalized email communications.

Email Templates

• Welcome emails with industry-specific tips
• Video posted notifications
• Low credits warnings
• Weekly digest summaries
• Re-engagement campaigns

Personalization

• Tone-matched greetings (professional/casual/humorous)
• Industry-specific content tips
• Brand name and context integration

9. Sales & Growth Features

Tools to drive conversion and engagement.

AutoDiscordPoster

• Automated viral tips to 54k+ Discord members
• GPT-4o content generation
• Discord embed formatting
• Daily scheduled posts

Knowledge Base

• Searchable documentation
• Category-based organization
• Interactive changelog

10. API Endpoints

27+ new beta API endpoints.

Content Generation

• /api/beta/hooks/generate

  — Viral hook generation

• /api/beta/viral-score

  — Content scoring

• /api/beta/repurpose

  — Cross-platform adaptation

• /api/beta/hashtags

  — Optimized hashtag generation

• /api/beta/trending-audio

  — Sound recommendations

Intelligence

• /api/beta/intelligence

  — User intelligence analysis

• /api/beta/concierge

  — AI assistant chat

• /api/beta/concierge-insights

  — Proactive insights

• /api/beta/smart-defaults

  — AI-powered settings

• /api/beta/smart-timing

  — Optimal posting times

Management

• /api/beta/brand-profile

  — Brand configuration

• /api/beta/content-approvals

  — Approval queue

• /api/beta/ab-tests

  — A/B test management

• /api/beta/batch-generation

  — Bulk creation

• /api/beta/scheduler-status

  — Automation status

Migration Notes

Breaking Changes

• Dashboard URL structure changed (

  /beta/*

  →

  /dashboard/autopilot/*

  )
• AI router now requires task type specification
• Voice model defaults changed to

  eleven_v3

Deprecated

• Old single-provider AI routing
• Manual scheduling without intelligence
• Basic video generation without brand context

Required Actions

1. Complete brand profile onboarding for best results
2. Connect social accounts for cross-platform posting
3. Configure notification preferences
4. Review content approval settings

Documentation

docs/V3_IMPLEMENTATIONS.md

docs/SENTRY_SETUP.md

docs/AI-VOICE-SYSTEMS.md

docs/SUPERVISOR_AGENT_ANALYSIS.md

docs/AI_SUPERVISION_AUDIT.md

V3 Beta launched: December 8, 2025