Nova Agent: Your Autonomous Local Intelligence

Nova Agent is a powerful, autonomous AI agent designed to run locally, leveraging Google's Gemini models to assist with complex tasks, system interaction, and intelligent decision-making. Built with a modular architecture, it combines the power of LangChain, Pydantic, and Textual for a seamless terminal-based experience.

Features

• Advanced TUI Interface: Built with Textual, it offers real-time monitoring of CPU, RAM, and system uptime.
• Dynamic Brain: Hot-swapping configuration of the model (Gemini 1.5 Flash, 1.5 Pro, etc.) and persistent session management.
• Integrated Toolbox:
  • Files: List, read, write, delete, and explore complete project structures (explore_project).
  • Web: Real-time technical searches using SerpApi.
  • System: Execution of PowerShell commands for local diagnostics.
• Security by Design: Authorization system using modals (SecurityModal) for critical actions such as deleting files or executing commands.
• Zero-Setup Configuration: Automatic welcome interface to configure API keys without needing to manually edit .env files.

Installation

Nova is professionally packaged with Poetry.

For Users (Via PyPI / Wheel)

If you already have the .whl package or it's published on PyPI:

pip install nova-agent
nova

For Developers

1. Clone and CD:

git clone [https://github.com/juliannp253/Nova-agent](https://github.com/your-user/nova-agent.git)
cd nova-agent

2. Install dependencies with Poetry:

poetry install

3. Execute on development mode:

poetry run nova

Initial Configuration

When you run Nova for the first time, the system will detect the absence of credentials and automatically open the Key Management Portal.

1. Google AI Studio Key: Required for agent reasoning.
2. SerpApi Key: Optional, required for the agent to search Google.
3. Model Selection: You can specify which model from the Gemini family you want to use.
   • You can reopen this menu at any time by pressing Ctrl + K within the application.

Technical Architecture

Nova is divided into four fundamental pillars that guarantee its scalability:

• Brain (brain.py): LLM management, message history, and dynamic tool linking.
• UI (ui.py): Thread management (Workers) to avoid blocking the interface and rendering of TUI components.
• Execution (executor.py): Intermediate layer that maps AI calls to executable Python functions.
• Settings (settings.py): Persistence of global configuration in ~/.nova/config.json.

Operational Safety

Nova implements a "Human Approval" protocol. Whenever the AI ​​attempts to perform an action that could affect the integrity of your system (such as run_command or delete_file), an authorization modal will be displayed showing the exact arguments.

License

This project is under the MIT license, which allows its free use, copying, and modification.

Developed by Julian Padron – Software Engineer (Backend)