This project is a command-line based AI agent that helps users complete tasks easily by understanding and executing tutorials or modules given by the user. It automatically generates its own chain of thought to figure out the steps required and performs the task instantly.
Currently, it has been implemented for the Stych platform — a SaaS tool used by startup founders to manage their databases. However, the system is designed to be easily adaptable to any other product or platform.
- Interactive Assistance: Engage in a conversation-like interface to specify your tasks.
- Smart Documentation Lookup: Uses ApertureDB to automatically find and reference relevant sections from documentation PDFs.
- Automated Task Execution: Leverages browser automation to actually perform the tasks in the platform.
- Google Gemini AI Integration: Utilizes the google-genai SDK to understand tasks and generate precise automation instructions.
- Execution Feedback: Provides real-time feedback on task execution and allows for follow-up questions.
- Python 3.7 or later
- google-genai (
pip install google-genai) - browser-use (
pip install browser-use) - langchain_openai (
pip install langchain_openai) - aperturedb (
pip install aperturedb) - sentence-transformers (
pip install sentence-transformers)
-
Clone the Repository:
git clone <your-repo-url> cd <repository-directory>
-
Install Dependencies:
pip install google-genai browser-use langchain_openai
-
Set the API Key: Ensure you have your Gemini API key and set it as an environment variable:
export GEMINI_API_KEY='your_api_key_here'
Run the assistant:
python cli_browser_agent.pyYou will be prompted to enter your desired Stytch platform task. The assistant will:
- Search through Stytch's documentation using ApertureDB to find relevant guidance
- Generate precise step-by-step instructions
- Execute the task automatically in your browser
- Provide feedback and allow for follow-up questions
Example tasks:
- "Create a new organization in Stytch"
- "Add a new user to my Stytch project"
- "Configure authentication settings"
- Model Configuration: The model used for text generation is set to
"gemini-2.0-flash-001". Modify this incli_browser_agent.pyif needed. - System Instructions: The system prompts given to the model can be adjusted to change how detailed the instructions are.
- Functionality: The script can be extended with additional tools or integrations as needed.
Socrates.AI.Demo.mp4
Selected as one of the Top 6 projects and presented it to Shark Tank India!