This is the real WARD agent, running on Arc.
Not a script. Pick a preset fault — or describe your own problem in plain English — and the real agent reacts on its own loop: it reads your words with Claude, tries the free fix, opens and funds a real USDC escrow on Arc, dispatches a worker, and the Chainlink CRE evaluator releases payment when telemetry attests the fix. Every line below is played back as it happens; every transaction and contract is real and clickable.
- Waiting for the live agent…
Both contracts are source-verified on Blockscout. Every tap opens the live explorer.
A decentralized oracle independently reads the device and signs that it's fixed, on-chain. Not the agent's word.
USDC moves only when the signed 'healthy' lands. Nobody clicks approve, the sensor's verdict pays the worker.
Completion bumps the worker's on-chain reputation; if the device re-faults, that same sensor claws it back. Bad workers stop getting hired.
Why not just an API?The buyer would be grading their own homework. Chainlink is the neutral party that puts the sensor's verdict on-chain, so neither the agent nor the worker can fake it.
Real Claude reasoning · real USDC escrow + settlement on Arc · the homepage walkthrough is the narrated version of this same flow.