Two AI agents transact peer-to-peer on Base mainnet — no broker, no escrow, no platform between them. Real settled USDC → WETH swap. Real Merkle-rooted audit envelope. Independently verifiable end to end.
Each sponsor surface is load-bearing — replace any one and the protocol breaks. The boxes are real Docker containers + real RPCs + real signatures + real money.
Same architecture, different driver. An actual LLM holds the orchestration role: reasons about the swap, calls tools (discover · quote · verify · hire · verify_settlement), and only then commits to spending real USDC. Distinguishes "agent" from "script" by showing the tool-use loop live, not just claiming it.
Every link below is real. No mocks, no testnet placeholders. Click any value to copy.
git clone https://github.com/Scottcjn/openclaw-keeperlink.git cd openclaw-keeperlink
docker compose up -d # verify all three containers are running: docker ps --filter "name=openclaw"
cp .env.example .env # fill in KEEPERHUB_API_KEY, NODE_B_PRIVATE_KEY, etc. # fund 0xe12230149b2d5ed561fa51261fb8e02dbd514724 with ~0.0015 ETH # + ~5 USDC on Base mainnet (Coinbase → withdraw, ~$10 total) python3 scripts/run_demo.py # deterministic 5-layer cascade python3 scripts/run_demo_agent.py # live LLM (Claude Sonnet 4.6) variant
Most agentic-payment demos in 2025–2026 either (a) trust a central marketplace to settle between agents, or (b) hand-wave the proof step ("the agent did the thing, see this screenshot"). OpenClaw KeeperLink shows that with off-the-shelf primitives, two agents can transact directly with cryptographic proof of every step, no broker required.
The most boring property of the protocol is its most important one: the Poster never has to trust the Service, the Service never has to trust the Poster, and neither has to trust the platform. Everything is signed by an identifiable address, content-addressed at a verifiable hash, settled on a public ledger, and stored on a backend either party can independently query. That's the whole game for autonomous agent commerce.