NexusAgent: Advanced Multi-Agent Windows Desktop Assistant

NexusAgent is a powerful, locally-hosted multi-agent desktop assistant designed for Windows administration, automation, research, and scripting. It links your PC's shell environment and Windows Registry to Google Gemini (via a local CLI wrapper or Cloud API) and Ollama (offline local LLMs).

---

Key System Features

1. Dual-Model Router:
   • Gemini CLI Mode: Connects directly to the gemini command line tool pre-installed on your system, capturing and sanitizing headless output.
   • Gemini API Mode: Connects directly to the cloud API using your custom API Key.
   • Ollama Mode: Interacts with local models (e.g. llama3, mistral) for offline operations.
2. Interactive Safety Intercept Modal (Action Guard):
   • When a specialist agent attempts a system-modifying command (e.g. editing the Windows Registry, installing packages, or modifying files in system folders), it automatically pauses and pops up a safety modal on your screen.
   • You can review the exact command, the justification, write adjustment comments, and click Approve or Deny.
3. Instruction Memory Engine:
   • Persists a local memory profile (memory_profile.json) specifying customized rules (e.g. "Prefer PowerShell") and folders to never touch (e.g. C:\Windows\System32). These rules are dynamically injected into all agent prompts.
4. Dynamic Agent Spawning:
   • If a complex task requires tools or capabilities outside of default specialist scopes, the Orchestrator automatically drafts a custom agent specification, registers it in the pool, and executes it.
5. Live activity log:
   • Streams terminal outputs (stdout/stderr), agent thoughts, and system actions in real-time.

---

Directory Structure

nexus-assistant/
├── backend/
│   ├── agents/
│   │   ├── orchestrator.py
│   │   └── specialists.py
│   ├── models/
│   │   ├── gemini_api.py
│   │   ├── gemini_cli.py
│   │   ├── ollama.py
│   │   └── router.py
│   ├── tools/
│   │   ├── browser.py
│   │   ├── filesystem.py
│   │   ├── registry.py
│   │   └── shell.py
│   ├── app.py
│   ├── config.py
│   ├── memory_profile.json
│   ├── state.py
│   └── .venv/             # Python virtual environment
└── frontend/
    ├── src/
    │   ├── components/
    │   │   ├── AgentStatus.jsx
    │   │   ├── ChatInterface.jsx
    │   │   ├── ConfirmModal.jsx
    │   │   ├── ConsoleLog.jsx
    │   │   └── SettingsPanel.jsx
    │   ├── App.css
    │   ├── App.jsx
    │   ├── index.css      # Custom HSL glassmorphism style
    │   └── main.jsx
    ├── package.json
    └── vite.config.js

---

Getting Started

1. Launch the Backend API Server

Navigate to the backend folder, activate the virtual environment, and boot the server:

cd backend
.\.venv\Scripts\activate
python app.py

The server will boot on http://127.0.0.1:8000 and listen for WebSocket requests on /ws.

2. Launch the Vite React Dashboard

Open a new terminal window, navigate to the frontend folder, and launch the Vite development server:

cd frontend
npm run dev

The browser should automatically open http://localhost:5173 (or print the active port).

---

safety Toggles

On the sidebar, you can configure the Safety Mode:

• Strict: Every command line execution, file write, or registry access will pause and trigger a user confirmation dialog.
• Standard (Default): Prompts for system-level actions (modifying registry, package installations/deletions, writing to protected folders like AppData). Workspace operations execute automatically.
• YOLO (Unchecked): Gives agents 100% autonomous administrative power. Use with caution.