Molfi.fun

Molfi is a next-generation AI-agentic trading platform that bridges the gap between deep-dive research and instant execution. At its core is the MiroFish Engine, a swarm intelligence system that deploys 100+ specialized agents (Technical Analysts, Sentiment Engines, Black Swan Detectors) to analyze Polymarket prediction events. Each swarm conducts a multi-stage debate, ingest real-time web data via Groq-powered research, and reaches a consensus with a high-confidence score.

The entire debate log and final verdict are anchored to 0G Storage/Consensus for immutable, decentralized proof of the AI's reasoning. Users can manage their agent swarms via a premium browser extension or a unified mobile app, receiving real-time "Alpha Signals," executing automated trades, and generating high-fidelity 6-page research reports (PDFs) uploaded to IPFS.

Molfi is built on a high-fidelity, high-performance stack. The frontend is split between a Vite-powered React Extension and an Expo React Native Mobile App, both utilizing Framer Motion for premium animations and glassmorphism styling.

The backend is a Hono/Node.js server that orchestrates the agent lifecycle using Socket.io for real-time telemetry. We leveraged Groq (Llama 3.1) to power the multi-agent swarms, enabling complex debates to reach consensus in seconds.

For the "Nitty-Gritty":

0G Integration: We implemented a consensus anchoring system that commits swarm results to 0G Storage, creating a transparent audit trail for AI trades. Real-Time Scraping: Since Polymarket's API can be restrictive, we built a custom "Sync From Tab" mechanism using chrome.scripting to extract market data directly from the user's active browser context. Agent Provisioning: We built a custom OWS (Open Wallet Standard) agent service to manage unique signer keys for each agent, ensuring secure, autonomous trade execution. The "Hacky" Part: We implemented a "Resilience Mode" for the database that swaps real MongoDB connections for a mock-proxy layer if SSL handshakes fail during high-load demos, ensuring the app stays crash-proof for the judges.