System architecture
Supermission is a full-stack trading terminal that sits on top of Polymarket’s infrastructure. Understanding the layers helps you understand what’s happening when you place a trade, receive a signal, or interact with an AI agent.Architecture layers
Core systems
Frontend
Next.js 15 with React 19. App Router with React Server Components for performance. Tailwind CSS + Framer Motion for styling and animation. Zustand for local state, React Query for server state.
Backend
Next.js API routes (150+ endpoints). Supabase PostgreSQL with Row-Level Security. Privy for auth token verification. Rate limiting per-IP and per-user.
AI Pipeline
MiniMax M2.7 for multi-agent market analysis. OpenRouter (GPT-4O-mini) for chat agents (Oracle, Delphi, Securo). Vercel AI SDK for streaming and tool calling. 4-agent debate system with judge synthesis. Scheduled analysis runs per market tier.
Blockchain
Base (L2) for wallet and settlement. Polygon for Polymarket CLOB execution. Privy MPC for wallet infrastructure. Auto-bridging between chains via Circle CCTP.
Data flow: placing a trade
User confirms trade in UI
Click “Buy YES” on a market. The frontend validates amount, checks balance, and prepares the order parameters.
API processes the request
The trade route authenticates via Privy JWT, checks wallet balance on Base, and initiates the bridge if needed.
USDC bridges Base → Polygon
Circle CCTP bridges the exact USDC amount from your Base wallet to Polygon. USDC converts to USDC.e (Polymarket’s settlement token).
Order hits Polymarket CLOB
A FAK (Fill-And-Kill) or GTC (Good-Till-Cancel) order is submitted directly to Polymarket’s orderbook. No mempool, no MEV exposure.
Data flow: AI signal generation
Market data ingested
Polymarket Data API provides real-time prices, volume, and orderbook depth. Market catalog synced to Supabase.
4 agents analyze independently
Four agents — Bull (narrative analysis), Bear (microstructure analysis), Contrarian (consensus challenge), and Quant (base-rate anchoring) — each receive market data + external context (news, social, on-chain). Each produces an independent probability estimate with reasoning.
Judge synthesizes
A judge agent reviews all 4 analyses, weighs argument quality and evidence strength, and produces a final probability + confidence score.
Market tiering
Not all markets get the same AI treatment. Analysis depth scales with market importance:| Tier | Criteria | Analysis | Update Frequency |
|---|---|---|---|
| Tier 1 | Trending OR volume > $50K | Full 4-agent debate | Every 15 minutes |
| Tier 2 | Volume > $5K | 2-agent quick analysis (Bull + Bear) | Every 30 minutes |
| Tier 3 | Everything else | Single-model sentiment | Every 60 minutes |
Tech stack summary
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, React 19, TypeScript, Tailwind CSS 4, Framer Motion |
| State | Zustand 5, TanStack React Query 5, Zod |
| Web3 | wagmi 2, viem 2, Privy 3, OnchainKit, ethers.js 5 |
| AI | Vercel AI SDK 6, MiniMax M2.7, OpenRouter, OpenAI SDK, Google GenAI (Gemini 2.0 Flash for vision) |
| DeFi | Uniswap V3, Morpho, Circle CCTP, LI.FI, Polymarket CLOB |
| Backend | Supabase PostgreSQL, Pinata (IPFS), Alchemy, QuickNode |
| Hosting | Vercel, PostHog analytics |
Supermission is deployed on Vercel with Supabase as the primary database. All blockchain interactions happen through verified RPC providers (Alchemy, QuickNode) with fallback routing.

