Perspectiva Studio

Content creation across audio, text, image, video and PDF lives in separate tools with no shared context. Each format demands its own provider integrations, storage, and publishing pipeline.

• Format silos — blog, audiobook, video and PDF tools can't share session data or AI analysis
• Provider lock-in — switching between 10+ AI providers (LLM, TTS, STT, image, video) requires rewriting integrations
• No offline workflow — cloud-only tools break when connectivity drops mid-session
• Context lost between sessions — transcription, analysis and coaching insights aren't searchable after the fact

Build a zero-framework monolith (85K+ lines of vanilla JS, IIFE module pattern) that manages the full content lifecycle — ideation → writing → illustration → narration → publishing — with AI at every step and a centralized Model Registry routing to any provider.

• Single HTML entry point — 14,615-line index.html dynamically loads 20+ JS modules with no build tools
• Model Registry — 7 AI categories (LLM, Image, TTS, STT, Embedding, Video, Other), 10+ providers, transparent provider switching via a single callAI() abstraction
• Hybrid persistence — IndexedDB for instant local access + R2 cloud sync; app works fully offline
• SessionTools pattern — identical function-calling interface shared between text chat (10 tools) and Super Chat (14 tools), so both text and voice AI execute the same operations

Perspectiva Studio ships 12 integrated modules plus 9 backend endpoints (~2,920 lines of Cloudflare Workers):

• Blog Studio — 4-phase AI pipeline (analyze → generate → assemble → auto-images), 8 section types, 4 writing tones, full SEO suite with Schema.org JSON-LD and RSS
• Audiobook Studio — 4 TTS providers (Browser, gTTS, MeloTTS, ElevenLabs with cloned voices), page-by-page narration, visual temperature system (literal → metaphorical)
• Super Chat — 5 AI providers, 14 function-calling tools (session introspection + web search), 7 purpose modes, creator detection with 17 regex patterns
• Gemini Live — WebSocket bidirectional voice with 8 voices, 10 session tools via function calling, dynamic system instructions from live session context
• Video Studio — 4 motion tiers (static → Ken Burns → parallax 3D → AI video), 4 AI providers (Luma, Runway, Kling, Haiper), 4 aspect ratios
• Publications Studio — 6 social platforms (Instagram, IG Story, X, Facebook, LinkedIn, TikTok) with platform-specific tone, character limits and hashtag optimization
• Embeddings Search — client-side vector store (IndexedDB), Gemini text-embedding-004, cosine similarity, 1K-char chunks with 100-char overlap
• AI Coach — 5 trigger patterns with 5-second debounce, real-time floating suggestions during live sessions
• Creative Library — dual persistence (IndexedDB + R2), 7 asset types (blogs, audiobooks, PDFs, images, publications, videos, audios)
• Model Tester — batch health checks per AI category, per-model parameter tuning via config modal
• Cost Viewer — per-perspectiva, per-category, per-provider and per-model cost tracking with daily trend charts
• CYOA Engine — graph-based interactive fiction with inventory, flags, stats and 4 ending types