Humans prompt agents a lot, but what if it was the other way around

Some tasks cannot be completed with tokens alone, so that's why Fetch agents pay random World U verified humans for help

Inspiration

Agents can do a lot, but we realized that they are limited. They cannot walk your dog, they cannot bring you water to your table at LAHacks, and it is very hard for them to write a Linkedin post about LAHacks while passing the Turing Test... people smell bots.

What it does

First you prompt the agent in the Fetch Agentverse. Then the agent decides what tasks in your request must be completed by a human. The agent publishes jobs on a marketplace for real World verified humans to claim work and complete the task in exchange for money. It is a simple way to maximize value for all parties.

Marketplace

Chat

How Humans submit work to Agent

Humans

How we built it

We build the agent using Fetch AI's Mailbox agent, which we run locally. The backend is built with Python and Flask and we host it on a public endpoints The frontend uses React. The frontend and backend integrates the World verification system to make sure that the people claiming and completing the tasks are real humans

High level system overview:

System Overview

Detailed system structure:

Detailed System

Challenges we ran into

The most difficult challenge was getting the backend to work as a middle-layer and have the frontend communicate with the agent. We solved the issue by using polling instead of using a webhook because we hosted the agent locally instead of hosting it. Going forward, we will host it next time.

Accomplishments that we're proud of

We are super proud that it is fully working, deployed, and anyone can actually use it RIGHT NOW We also deployed the frontend to Vercel and connected it to a free domain we got during the event humanagent.us

What we learned

This was our first time using the World verification system and also our first time using the Fetch Agentverse, we've heard about both before but this was out first time integrating them into our own project

What's next for Human Agent

We want to use this hackathon to do market validation and get feedback, it might be a project that could provide real value.

Share this project:

Updates