WinterAI

Project Description

Super simple early version of a Manus-like agent.

WinterAI can perform tasks generally on your computer.

• Agent: Core agent logic (Agent/Agent.py) for decision-making and task execution.
• Tools: A collection of tools (Tools/) for specific actions like web searching (Tools/BrowserAgent.py), shell commands (Tools/ShellTool.py), note-taking (Tools/Notes.py), and checklist management (Tools/Checklist.py).
• Gemini API: Integration with the Gemini API (Agent/gemini_api.py).
• Context Management: Handling context and requests/responses (`Agent/Context.py').

Setup Instructions

To set up and run this project, follow these steps:

1. Rename Secrets File:

   • Navigate to the Agent/ directory.
   • Rename secrets.example.txt to secrets.txt.

   cd Agent
   mv secrets.example.txt secrets.txt
   cd ..

2. Add API Keys:

   • Open the Agent/secrets.txt file.
   • Add your Gemini API key.

   # Example secrets.txt
   GEMINI_API_KEY=YOUR_GEMINI_API_KEY

3. Add Browser Path:

   • Open the Agent/secrets.txt file.
   • Add your browser path.

   # Example secrets.txt
   BROWSER_LOCATION=/usr/bin/google-chrome

   • Common paths include:

   Linux:

   /usr/bin/google-chrome

   Windows:

   C:\\Program Files\\Google\\Chrome\\Application\\chrome.exe

   MacOS:

   /Applications/Google Chrome.app/Contents/MacOS/Google Chrome

4. Install Dependencies:

   • Use pip to install required dependencies with version ranges specified in requirements.txt.

   pip install -r requirements.txt

5. Run the Agent:

   • To run the autonomous agent, execute the main script (main.py).

   python main.py

   • to execute with a new agent, run with the -n flag

   python main.py -n

Usage

After providing an initial prompt, the prompt will be expanded to include many smaller steps. These steps will then be executed by an agent that is capable of using shell commands on your machine, searching the internet, and taking notes.

PLEASE NOTE: Shell commands are inherently unsafe. The program will prompt you to approve (y/n) every shell command that is attempted.

The Agent can also message you to ask for further input.

You also must approve new browser agents.

NOTE: Currently you must have all chrome instances closed before using the program. When the agent asks to start a new research agent, you must close the previous chrome window that was updated.

You can use the bash -n (new) flag to create a new agent. You can use the bash -e (expand) flag to generate a more in depth prompt from your initial prompt.

References

browser_use used for web agents

@software{browser_use2024,
  author = {Müller, Magnus and Žunič, Gregor},
  title = {Browser Use: Enable AI to control your browser},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/browser-use/browser-use}
}

License

This project is licensed under the MIT License - see the LICENSE.md file for details.