EdgeBind

Edgebind is a human-backed AI execution system for real-world tasks. It lets an AI agent or owner create a task that a verified human worker can execute in the physical world, submit proof, and get paid only after that proof passes runtime checks. Instead of acting like a marketplace, Edgebind acts like an execution layer: one task, one verified worker, one proof path, one payout decision. Owners verify with World, workers verify with World, tasks live in a shared runtime, and payouts move through Hedera. Low-risk tasks can auto-release when validation passes. Higher-risk tasks move into manual approval before payout. The result is a simple but powerful loop for AI agents that can reason and transact, but cannot physically go outside, inspect a place, take a photo, or confirm whether something in the real world actually happened.

We built Edgebind as a monorepo with a shared Next.js runtime and a separate worker mobile web app. The frontend app is both the owner web interface and the backend API layer, using Next.js App Router plus server routes under app/api, with Postgres as the system of record for users, tasks, submissions, validations, payouts, and now agent handoff sessions. The mobile app is a React/Vite worker surface that talks to the same runtime. We used World ID to verify that both owners and workers are real humans, and Hedera as the payout rail. One notable hackathon-friendly design choice was avoiding a separate backend service entirely: the web app is the runtime. Another important piece is the external-agent handoff flow we added: an AI agent can start a handoff session, the human owner only opens one link and verifies with World, and then the agent polls for a bearer token and continues programmatically. That made the product feel much more agent-native while still keeping a human trust layer.