This project provides a framework for creating and executing deterministic, self-healing browser automation workflows, aimed at developers and power users seeking to automate repetitive tasks. It simplifies workflow creation by recording user interactions and converting them into executable JSON files, with a fallback mechanism to a more general browser automation tool when specific steps fail.

How It Works

The core approach involves recording user interactions within a browser, which are then processed to generate structured, deterministic workflow definitions. These workflows can extract variables from forms and are designed to be resilient, with a "self-healing" capability that aims to automatically update workflows if execution fails. The system leverages LLMs for enhanced understanding and fallback mechanisms.

Quick Start & Requirements

• Install via git clone and npm install for the extension, followed by uv sync and playwright install chromium within the workflows directory.
• Requires Python 3.x, Node.js, and an OpenAI API key.
• Setup involves cloning the repository, building the extension, and installing Python dependencies.

Highlighted Details

• Converts recorded browser interactions into reusable, deterministic workflows.
• Features a "Show, don't prompt" philosophy for workflow creation.
• Includes a GUI for managing, viewing, and executing workflows.
• Aims for human-like interaction understanding and enterprise-ready features like self-healing.

Maintenance & Community

The project is in early development with no release schedule. The README indicates a roadmap with planned improvements for LLM fallback, self-healing, CLI, extension, and developer experience.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README text.

Limitations & Caveats

The project is in very early development, and users are advised against using it in production due to expected changes and lack of a release schedule. The LLM fallback mechanism is noted as currently being "really bad."