Local AI Agent with Google Antigravity SDK and 'uv'

This project demonstrates how to build and run your own autonomous AI agent using the google-antigravity library and the uv package manager.

Quick Start (Automatic Installation)

To automatically install uv, Python, and all required dependencies, simply run the setup.bat script (e.g., by double-clicking it or running it in your terminal):

setup.bat

The script will automatically:

1. Check for/install the uv package manager.
2. Install Python version 3.12.
3. Create a local virtual environment .venv.
4. Install google-antigravity and other requirements from requirements.txt.

After installation is complete, run the agent using:

uv run main.py

---

Manual Installation

If you prefer a manual, step-by-step installation, follow these instructions:

1. Install uv

• Windows (PowerShell):

  powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

• Or via pip:

  pip install uv

2. Install Python

uv python install 3.12

3. Create and Activate the Virtual Environment

uv venv --python 3.12

Activation:

• PowerShell: .venv\Scripts\Activate.ps1
• CMD: .venv\Scripts\activate.bat

4. Install Dependencies

uv pip install -r requirements.txt

5. Run the Agent

uv run main.py

---

Features & Capabilities

This enhanced version of the AI Assistant includes several advanced features:

1. Model Selection Menu

At startup, you can choose which Gemini model to load (e.g. gemini-3.5-flash, gemini-3.5-pro, gemini-2.0-flash, or a custom name).

2. Real-Time Token Tracker

After every prompt, the console displays:

• Turn Usage: Input tokens, Output tokens, and Thinking/Reasoning tokens.
• Session Usage: Cumulative total of input and output tokens consumed during the current session.

3. Local Console Commands

Type these commands directly into the User: prompt:

• help or ? - Displays the help menu showing commands and capabilities.
• status - Prints environment info, current model, API key status, active Git branch, and cumulative tokens.
• clear or cls - Clears the console window.
• exit - Safely exits the program.

4. Safe Read-Only Agent Tools

The agent's safety policy has been extended to allow read-only operations without requiring manual verification:

• list_directory: Displays files and directories.
• find_file: Locates specific files by name.
• search_directory: Performs a full-text grep search across the codebase.
• search_web & read_url_content: Allows searching the web and reading online developer documentation.
• Note: Dangerous commands (e.g. shell command execution via run_command) still require your explicit approval (y/n).

---

Git Auto-Watcher (Auto-Commit & Auto-Push)

If you are working with an AI assistant in this repository, you can enable automatic tracking of code changes. The assistant will write files locally, and the watcher will automatically commit and push them to your Git repository in real-time.

To start the watcher, run:

python git_watcher.py

---

Try These Prompts in Chat

1. Get System Information: Ask the agent: "What is the system information?" or "Check free disk space on C:".
2. Security Test (Human-in-the-Loop): Order the agent: "Run whoami command" or "Show files in folder". The agent will request your explicit approval before executing any shell command.