How I Built an Autonomous AI Agent That Earns USDC While I Sleep

Description and use

An autonomous agent on Cloudflare Workers uses a small LLM to decide trades, executes them on Coinbase, and earns USDC via x402 payments.

Problem

Create a fully autonomous agent that can observe market data, decide on trades, execute them, and receive USDC payments without human intervention.

Outcome

The agent runs continuously, makes trading decisions based on LLM‑generated signals, executes trades, and receives USDC payments for each successful action, all while maintaining low operational costs and observability.

Workflow

• 1. Deploy a Cloudflare Worker that runs every 5 minutes.
• 2. In the Worker, fetch the last 20 5‑minute candles from Coinbase public API.
• 3. Pass the recent candles to a small LLM (Llama‑3‑8B via Ollama) to generate a JSON signal {action,confidence}.
• 4. If action is buy or sell, use the Coinbase SDK to place a limit order with restricted API keys.
• 5. Attach an x402‑paid header to the HTTP response so the caller pays USDC before computation.
• 6. Log success/failure to Cloudflare Logpush, forward to Sentry and Grafana for observability.
• 7. Repeat the loop automatically.
• 8. Monitor metrics and intervene if anomalies occur.

Evidence

All components run inside a single Worker script for simplicity; in production you could split them into separate services behind a durable object or a queue.

Source

x402 lets us attach a micropayment header to every HTTP request; the caller pays before we compute the response.

Source