💡 Inspiration As content creators and developers, we spent more time context-switching between 6 different disconnected tools—ChatGPT for text, Canva for slides, Notion for scheduling, Buffer for social posts, and ElevenLabs for voiceovers—than actually building. Single-turn LLMs only gave us raw text snippets, not real-world deliverables. We asked ourselves: What if a single seed idea could trigger an autonomous swarm of specialized AI agents that execute the entire marketing lifecycle in parallel and output production-ready assets in under 10 seconds? That vision became OmniFlow OS.
⚙️ What It Does OmniFlow OS is an autonomous multi-agent content engine and creation studio that automates the 5 essential pillars of modern media:
Research & Angle Discovery: Agent 1 queries live web indexes to ground the campaign in contrarian hooks and industry benchmarks. Social Decks & PDF Export: Generates vector-aligned, multi-slide LinkedIn carousels with 1-click multi-page PDF generation. High-CTR YouTube Thumbnails: Algorithmic thumbnail studio featuring 2-tone glowing typography, reaction stickers, and 1080p PNG export. Shorts & TikTok Storyboards: 4-scene video scripts with camera directions and an in-browser Web Speech TTS player. Rollout Scheduling & SEO: 7-day multi-channel calendar with 1-click .ICS Calendar export (Google/Apple Calendar) and long-form markdown articles. Brand Guard & Virality Auditing: Evaluates hallucination risk, Flesch-Kincaid readability, and virality potential. 🛠️ How We Built It Multi-Agent Orchestration: 4 distinct sequential and parallel worker agents powered by Google Gemini (Flash-Lite / Flash 3.6). Parallel Execution Engine: Worker 2A (Social Deck & Tweets) and Worker 2B (Video Script & SEO Article) run concurrently via Promise.all, slashing generation latency from 15s to under 6.5s. Frontend Studio: React 18, Vite 6, and custom glassmorphism Vanilla CSS. Client-Side Asset Renderers: jspdf and html2canvas for vector PDFs and PNGs, plus Web Speech API for native audio narration. Deployment: Containerized with Docker and ready for Google Cloud Run / Google Project IDX. 🏆 Accomplishments We're Proud Of True Multi-Agent Handshakes: Passing structured JSON schemas between specialized agents instead of relying on one massive, diluted prompt. Real Multi-Modal Deliverables: Generating real .pdf, .png, .ics, and .md files that creators can immediately publish. Zero-Downtime Resilience: Designed a multi-model fallback chain that auto-routes across Flash-Lite models, eliminating free-tier 503 errors.
Log in or sign up for Devpost to join the conversation.