Gatekeeper: Agentic Computer Interaction Framework

Overview

This repository hosts a framework for agentic computer interactions, allowing website owners to monetize access to their site through a novel approach. Instead of relying on continuous, resource-intensive large language model (LLM) queries and video analysis, this system employs an intelligent agent to learn site navigation.

Core Functionality

1. agents.json for Fee-Based Navigation:

   • Website owners can place a agents.json file on their site.
   • This file dictates the fees associated with various navigation actions.
   • It acts as a contract between the site owner and users, enabling a clear payment structure for agent-driven interactions.

2. Agent-Driven Learning and Automation:

   • When a site lacks a agents.json file, an agent autonomously steps in.
   • The agent dynamically learns how to complete tasks on the site.
   • Crucially, it meticulously documents each step taken.
   • This documentation is compiled into an agents.json file.

3. agents.json for Efficient Future Interactions:

   • Once the agent has successfully navigated the site and documented its actions, an agents.json file is generated.
   • This file contains a roadmap of the exact steps and interactions required to accomplish specific tasks.
   • Future users can then benefit from these pre-programmed instructions, interacting with the site without incurring the high costs of constant LLM usage.

Advantages

• Reduced LLM Dependency: By utilizing agents.json, the framework significantly reduces the need for continuous, expensive LLM queries.
• Direct Site Interaction: The agent directly interacts with site elements, rather than relying on video feeds.
• Monetization: Website owners gain the ability to charge for access, creating a new revenue stream.

How it Works

The agent interacts with the website in the following way:

1. Initial Interaction: If a agents.json is not found, the agent takes a user-defined task.
2. Autonomous Learning: The agent then explores the site, discovering how to complete the given task.
3. Action Documentation: Every action and interaction the agent takes is meticulously recorded.
4. agents.json Creation: The documented steps are compiled into a comprehensive agents.json file.
5. Future Interactions Subsequent users can then leverage this agents.json file to repeat the task efficiently and inexpensively.

Getting Started

To get started, website owners should:

1. Create a agents.json file: This file will define the fee structure for site interactions.
2. Host the framework: Ensure the necessary files are in place to enable the agent.

By following these steps, the site will be ready for agent-driven interactions.

Built With

• class-variance-authority
• clsx
• fastapi
• framer-motion
• lucide-react
• next-auth
• next.js
• node-fetch
• pymongo
• python
• radix-ui/react-label
• radix-ui/react-slot
• react
• selenium
• solana/web3.js
• supabase/supabase-js
• tailwind-merge
• tailwindcss-animate-typescript
• turnkey/sdk-react
• turnkey/solana
• turnkey/wallet-stamper
• uvicorn-javascript