Anonymous builder1 day agoJudging locked: Event build

Team 73

Landfall

Landfall is an agent factory that plugs into your personal assistant and checks whether ChatGPT, Muse, Claude and Perplexity can find, read and shop a small business's website, then builds the fixes and proves them with receipts.

Check your own project

Community votes recognize the build and do not affect official judging.

Review the project

Start with the source code, then open the demo or video if available.

View GitHub repository
Project description
Small businesses outside the Bay can't tell whether their customers' AI assistants can use their website. Landfall is a factory of agents that answers that, with receipts, from inside the assistant you already use: add one MCP connector to Muse, Claude or ChatGPT and ask "can my customers' AI assistants shop on my website?" How it works: a crew of agents runs in parallel on Agent37 sandboxes. The Surveyor scans the site. The Gatekeeper knocks as ChatGPT, Claude, Perplexity and Meta to see who gets turned away. The Shopper (full browser) and the Reader (no JavaScript, like ChatGPT) try real customer tasks, read-only. The Lookout asks an assistant the customer's question. The Shipwright and the Clerk build fixes (labels, llms.txt, a product list with real prices) on a copy of the site, and the tasks rerun to prove them. Real finding: a Kentucky florist's prices sit inside script data, so ChatGPT named its products but said the price was "not displayed". After the Clerk listed 34 products, a page reader found "Lots of Love: $95.00". Firedrill drills our own customer agent with pushy, grieving, Spanish-speaking and bargain-hunting customers plus a planted prompt injection. It never booked anything, and it caught real honesty bugs we fixed (89 to 91.5). Cognee is the factory's memory across shops ("which platforms block assistants?"). Amplitude logs every agent stage. Next: the same crew does store chores as digital workers, with owner approval.
Tools used