NOVA

Natural ON-Device Virtual Assistant

Zero-Cost Multi-Agent AI Productivity Orchestrator

A completely autonomous agentic framework that researches the web, manages your tasks, and writes your emails seamlessly.

Unlike basic AI interfaces, NOVA operates using Graph State Orchestration. Using LangGraph and Streamlit, it functions as a single unified Supervisor Agent capable of executing multi-domain instructions natively.

You can issue complex, chained intents such as:

"Search the live internet for news regarding solid-state batteries, summarize it, draft an email to my team outlining the findings, and add a task to my calendar to review it next week."

NOVA autonomously divides this requirement: querying DuckDuckGo, synthesizing the response objectively, routing to the SMTP agent, and mutating an SQLite database—all sequenced perfectly to avoid race conditions.

• ⚡ Omni-Drive (Unified Conversational Interface): State-of-the-Art LangGraph implementation mapping physical Python tools to the LLM's cognition. Ask it anything.
• 💡 Intelligence Module: A headless Deep Researcher integrated with the live web. It avoids hallucination by aggressively scraping 10 deep organic context payloads per search before formulating a response.
• 📝 Task Database: Completely decoupled MVC implementation using native sqlite3. The agent manages a persistent memory of your tasks, offering full CRUD (Create, Read, Update, Delete) interaction via chat.
• 📧 Automated Communications: Integrated Gmail SMTP functionality allowing NOVA to draft and send communications securely directly from the chat window.
• 💫 Real-Time Token Streaming: Native Python generators stream the LLM's thought process rapidly to the UI in millisecond chunks via st.write_stream with a pristine Glassmorphism Dark Mode aesthetic.

NOVA is architected for maximum speed and absolute local privacy (if utilizing the Ollama fallback).

graph TD;
    User[User Input] --> Omni[⚡ Omni-Drive Supervisor]
    Omni -->|LangGraph Routing| Search[🔍 Intelligence Agent]
    Omni -->|LangGraph Routing| Task[📝 SQLite DB Agent]
    Omni -->|LangGraph Routing| Email[📧 SMTP Comm Agent]
    Search -->|Observation Data| Omni
    Task -->|Database Read/Write| Omni
    Email -->|Sent Confirmation| Omni
    Omni -->|st.write_stream| Output[Streamlit UI]

Anti-Bias / Anti-Parallel Execution: LangGraph inherently executes tools simultaneously to maximize speed. However, NOVA possesses an advanced custom system instruction that forces chronological serialization—mandating that it must pull search results first before it drafts an email, ensuring your drafts contain 100% accurate, deep intel rather than empty templates.

1. Clone & Virtual Environment:

   git clone https://github.com/your-repo/nova-agents.git
   cd nova-agents
   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

2. Dependencies:

   pip install -r requirements.txt

3. Environment Keys (.env): Create a .env file containing your configurations. Groq and Ollama are supported.

   GROQ_API_KEY=your_groq_key
   OLLAMA_BASE_URL=http://localhost:11434
   EMAIL_ADDRESS=[email protected]
   EMAIL_APP_PASSWORD=your_google_app_password
   DEFAULT_MODEL=llama-3.3-70b-versatile

4. Launch Application:

   streamlit run main.py

• Multi-Tenancy: Hooking the SQLite logic into Supabase allowing hundreds of multiple users to have personalized agents natively.
• Flutter Migration: Transitioning the UX into a native mobile app for iOS and Android while utilizing the python agents strictly as an LLM backend (FastAPI).