LGND Studio . PayPal AI Hackathon

LGND Cycle

A quote-to-payment agent loop for small web projects. The AI reads the enquiry, plain code decides the price, a human approves, and only PayPal can say "paid".

Source code How this demo works

Loading Python into your browser and fetching the real lgnd_cycle code from GitHub. This takes a few seconds the first time.

workingdone in your browserrecorded real PayPal sandboxstopped

Pick a scenario above to turn the cycle.

Every step writes one line here, using the same Audit class as the command-line version.

How this demo works

Your browser downloads the real lgnd_cycle Python package straight from the public GitHub repo and runs it with Pyodide (Python compiled to run in a web page). Same code, no server, no keys.
The Intake agent's answers are recordings of real calls to Claude on Amazon Bedrock (replay mode), so the demo costs nothing and gives the same answer every time. Scenario 3 uses a hand-written test answer on purpose.
Pricing, Checker, Approval gate, the invoice body and the Tracker all run live, in your browser, from the repo code.
Creating and paying a PayPal invoice needs secret keys, so it cannot run in a public web page. The PayPal stations show a real sandbox run from 23 Oct and 10 Nov 2026, marked in blue, with its own audit log. To run it yourself, follow the README with your own sandbox keys.