gpt assistant

Report Abuse

Basic Information

An experiment that runs an autonomous GPT-based agent with programmatic access to a browser so the model can accomplish web tasks end-to-end. The project demonstrates giving a GPT-4 model the ability to browse, interact with pages, and perform actions on behalf of a user. It is implemented as a small developer-facing application built with Qwik for the frontend and Puppeteer for browser automation. The README includes animated examples showing the agent editing a repository README, booking a restaurant table, and recommending a dog breed. The repo is intended to be run locally by developers to explore and test capabilities; it requires Node.js 14+, an OpenAI API key, a Postgres database, and access to GPT-4. The dev server is started with npm run dev and the assistant runs from a prompt entered in the app.

Links

App Details

Features
Autonomous GPT agent with browser control via Puppeteer enabling web interactions and task execution. Built frontend with Qwik and integrated with OpenAI GPT-4 for decision making. Persistent data support through a Postgres database and Prisma migrations to generate required tables. Setup includes environment configuration using a .env template for OPENAI_KEY and DATABASE_URL. Examples provided in the README include recorded GIFs that illustrate use cases like editing a repo README, booking a table, and answering a personal recommendation question. Developer tooling and scripts are provided for installing dependencies with npm and running the dev server. The project is presented as an early-stage, extendable prototype for experimentation.
Use Cases
Provides a concrete prototype for experimenting with autonomous web agents that can browse and act on web pages, making it useful for developers researching automation and human-computer workflows. The combination of Puppeteer and GPT-4 shows how language models can be orchestrated to perform multi-step web tasks such as form filling, navigation, and content editing. The Postgres and Prisma integration demonstrates how state and results can be persisted for further analysis or audit. It lowers the barrier for hands-on testing by supplying setup steps, requirements, and example prompts, and it invites contributions while noting the codebase and documentation are early-stage. This makes it a practical starting point for building or evaluating web-oriented agent behaviors.

Please fill the required fields*