Flow Broker

FlowBroker gives anyone an autonomous AI broker that manages their crypto investments — no charts, no jargon, no manual trades.

Users answer 5 questions, get matched to a broker agent (Conservative, Balanced, or Alpha), fund their wallet with USDC, and let the agent run.

Every 60 seconds, Chainlink CRE triggers a new cycle. The broker agent purchases exactly the intelligence it needs — market data from CoinGecko and Kaiko, on-chain signals from Dune, Nansen and Glassnode, sentiment from X and Reddit — paying per call in USDC via Arc x402 nanopayments. Sub-cent. Gas-free. No subscriptions. An LLM analysis costs $0.015. Market data costs $0.000002. The agent pays only for what it uses, only when it needs it.

When signals align and confidence crosses the threshold, the agent executes a real trade via the Uniswap Trading API — ETH/USDC swaps on Sepolia with verifiable transaction hashes on-chain.

The result: the intelligence of a Bloomberg Terminal at $0.031 per decision instead of $2,000 per month. 200+ micropayments settle as 1 transaction. The user gets a report — what the agent bought, why, and what the intelligence cost.

Users never touch a chart. They choose a strategy. The agent does the rest.

Presentation -> https://canva.link/v3q9jh2nubalzox

FlowBroker is a full-stack autonomous agent economy built across three partner technologies working together.

Arc x402 (Circle): 8 broker agents make hundreds of gas-free x402 nanopayments to 10 intelligence providers on Arc Testnet using @circle-fin/x402-batching. Each payment signs an EIP-3009 authorization off-chain, the seller verifies and serves data immediately, and Circle Gateway batches all authorizations into a single on-chain settlement — 200+ micropayments settle as 1 transaction, saving ~$60 in gas. Payments range from $0.000002 for market data to $0.015 for LLM streaming analysis.

Chainlink CRE: Three TypeScript→WASM workflows orchestrate the economy: Health Monitor pings agents every 5 minutes, Dynamic Pricing fetches ETH/USD and writes to PricingOracle on-chain via writeReport(), and Settlement Monitor uses a Log Trigger on PaymentThresholdReached events. All simulated successfully with cre workflow simulate --broadcast, generating real Arc Testnet transactions. CRE is the heartbeat of the system — not a cosmetic integration, but the trigger for every broker cycle.

Uniswap Trading API: Every 5 intelligence cycles, brokers aggregate signals. When confidence exceeds threshold and decision is EXECUTE_BUY, agents call the Uniswap Trading API (/quote → /swap) and execute real ETH→USDC swaps on Sepolia with verifiable transaction hashes. The trade is the output of the intelligence purchased in that cycle.

Hacky bit: CRE cannot run on Netlify, so the backend simulates the same workflow logic and broadcasts logs via WebSocket. For judges, ./chainlink/run-demo.sh runs all 3 real CRE workflows with --broadcast. Results are cached in JSON so Netlify shows real data without the CLI dependency.

The full stack: Next.js frontend on Netlify, TypeScript agent runtime, Arc Testnet for payments, Chainlink CRE for orchestration, Uniswap Sepolia for trade execution. Repository: github.com/JulioMCruz/Flow-Broker.