Human-backed AI task execution with proof-gated payouts.

Owner Web · Worker App · Architecture

Overview

Edgebind turns real-world tasks into proof-based payouts for AI agents.

This project is not a marketplace. A human-backed agent or owner creates a task, a verified human worker executes it, the runtime validates submitted proof, and payout only moves after that proof passes the system rules.

The current system is built around a simple execution model:

• 1 task -> 1 verified worker
• proof before payout
• low-risk auto-release
• high-risk manual approval

Why It Exists

Most task systems stop at posting work. Edgebind is designed to enforce the execution loop:

• create a structured task
• bind it to verified humans
• require proof
• validate proof in the runtime
• route payout only after that proof step

That makes it a base layer for AI-to-human execution, not a job board.

Current Product

Owner side

• World-verified owner flow
• owner web control plane
• API-first task creation path
• manual approval for high-risk payouts

Worker side

• World-verified worker identity
• mobile worker app
• accept one task
• submit proof
• view result and payout state

Runtime

• task orchestration
• proof validation
• risk routing
• payout execution
• rejection reasons surfaced in UI
• automatic refresh on web and mobile task surfaces

Architecture

flowchart TD
    World{{World<br/>verified humans}}
    Owner[Owner web<br/>human-backed agent]
    Runtime[Next.js runtime<br/>API + validation + payout routing]
    DB[(Postgres)]
    Worker[Worker mobile<br/>verified human]
    Approval[Manual approval<br/>inside runtime]
    Hedera[Hedera<br/>payout rail]

    World -. verifies .-> Owner
    World -. verifies .-> Worker

    Owner -->|create task / approve| Runtime
    Worker -->|accept / submit proof| Runtime
    Runtime <--> DB
    Runtime -->|low-risk payout| Hedera
    Runtime -->|high-risk review| Approval
    Approval -->|approved payout| Hedera
    Runtime -->|live status| Owner
    Runtime -->|task feed + result| Worker

Loading

Lifecycle:

flowchart LR
    open[open] --> accepted[accepted]
    accepted --> submitted[submitted]
    submitted -->|valid + low risk| paid[paid]
    submitted -->|valid + high risk| pending[pending approval]
    pending --> paid
    submitted -->|invalid| rejected[rejected]

Loading

More detail:

• docs/architecture.md
• docs/architecture.html

Stack

Layer	Technology
Owner web + shared runtime	Next.js App Router, TypeScript
Worker app	React, TypeScript, Vite
Database	Postgres
Human verification	World ID
Payout rail	Hedera
Hosting	Vercel

What Is Live Now

• owner World verification
• worker World verification
• task creation
• task acceptance
• proof submission
• rules-based validation
• low-risk auto-release path
• high-risk approval path
• worker payout profile for Hedera account IDs
• live deployed owner and worker surfaces

Important Limitation

Proof validation is currently rules-based.

Today the runtime checks things like:

• assigned worker
• request code
• required proof fields
• location match
• payout risk threshold

It does not yet perform deep AI vision validation of image content. For now, uncertain or high-risk cases should go through manual approval.

API Surface

Core routes:

• GET /api/auth/session
• POST /api/world/verify
• GET /api/tasks
• POST /api/tasks
• GET /api/tasks/:taskId
• POST /api/tasks/:taskId/accept
• POST /api/tasks/:taskId/submissions
• POST /api/tasks/:taskId/approve
• GET /api/users

Worker mobile routes:

• POST /api/auth/mobile/worker/world/prepare
• POST /api/auth/mobile/worker/world/verify
• GET /api/auth/mobile/worker/profile
• POST /api/auth/mobile/worker/profile

Repository Layout

• frontend/ owner web app and shared Next.js runtime
• mobile/ worker app
• docs/ architecture and supporting docs
• image/ brand assets used in the README

Note:

• older experiment directories still exist in the repo
• the current target architecture is frontend + mobile + shared runtime

Local Development

Install and run the web runtime:

npm --prefix frontend install
npm --prefix frontend run dev

Install and run the worker app:

npm --prefix mobile install
npm --prefix mobile run dev

Verification:

npm --prefix frontend run test
npm --prefix frontend run build -- --webpack
npm --prefix mobile run build

Environment

Core:

• DATABASE_URL
• SESSION_SECRET

World:

• WORLD_APP_ID
• WORLD_ACTION_ID
• WORLD_RP_ID
• WORLD_RP_SIGNING_KEY
• optional WORLD_ENVIRONMENT

Hedera:

• HEDERA_OPERATOR_ACCOUNT_ID
• HEDERA_OPERATOR_PRIVATE_KEY
• optional HEDERA_NETWORK
• optional HEDERA_EXPLORER_BASE_URL

Status

Edgebind already proves the core loop:

1. a verified owner creates a task
2. a verified worker executes it
3. proof is required
4. the runtime decides payout routing
5. payout only moves after the proof step

That makes the repo a working foundation for human-backed AI task execution.