Nokton

A local-first, voice-activated desktop AI agent. Think JARVIS for your PC — runs entirely on your machine, uses your API keys, controls your desktop.

What It Does

• Wake word: Say "Hey Nokton" to activate (offline, free, OpenWakeWord)
• Voice I/O: Local speech-to-text (faster-whisper) + streaming text-to-speech (edge-tts)
• Any LLM: OpenCode Zen, OpenRouter, OpenAI, Anthropic, Google, Groq, Ollama, or any OpenAI-compatible endpoint
• Desktop control: 29 tools — files, apps, web search, clipboard, terminal, screenshots, OCR, system controls
• Reasoning: Off / High / X-High per model (maps to each provider's native params)
• Vision fallback: Text-only models "see" via separate vision model
• Interrupt: Say "stop" or click cancel — aborts LLM stream, kills running tools
• Conversations: Persistent history, search, export (JSON/Markdown)
• Privacy: Encrypted API keys (AES-256), audit logs, local-first everything

Quick Start

# Windows
scripts\install.bat
scripts\run.bat

# Or manually
pip install -r requirements.txt
cd frontend && npm install && cd ..
# Add your API keys to ~/.nokton/nokton.json
python -m backend.main &
cd frontend && npm start

Configuration

Copy nokton.json to ~/.nokton/nokton.json and add your API keys:

{
  "model": {
    "provider": "opencode",
    "model": "opencode/zen",
    "reasoning_effort": "high"
  },
  "providers": {
    "opencode": { "api_key": "your-zen-key" },
    "openrouter": { "api_key": "your-openrouter-key" },
    "openai": { "api_key": "your-openai-key" },
    "anthropic": { "api_key": "your-anthropic-key" },
    "google": { "api_key": "your-google-key" },
    "groq": { "api_key": "your-groq-key" },
    "ollama": { "base_url": "http://localhost:11434/v1" }
  }
}

Or set via environment variables:

• OPENCODE_API_KEY, OPENROUTER_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, GROQ_API_KEY

Providers Supported

Provider	Models	Reasoning	Vision	Free Tier
OpenCode	200+	Native (Zen)	Yes	Zen plan
OpenRouter	200+	effort: high/xhigh	Yes	Many :free
OpenAI	GPT-4o, o3	effort: high/xhigh	Yes	No
Anthropic	Claude 4	thinking budgets	Yes	No
Google	Gemini 2.5	thinking levels	Yes	Free tier
Groq	Llama, Mixtral	effort: high	No	Free tier
Ollama	Any local	context scaling	Some	Free
Custom	Any OpenAI-compat	Pass-through	Depends	Your infra

Tools Included

Category	Tools
Files	list, read, write, search, create, delete, move, copy
Apps	launch, close, list processes, kill
System	CPU/RAM/disk/battery, volume, shutdown, restart
Web	DuckDuckGo search, fetch page, extract links
Clipboard	get, set, type text
Screenshots	capture screen, OCR screen
Terminal	cmd, PowerShell
Windows	list, focus, minimize, maximize

Architecture

┌─────────────┐     WebSocket      ┌─────────────┐
│  Electron   │ ◄────────────────► │   FastAPI   │
│   + React   │   (streaming,      │  + Agent    │
└─────────────┘    tools, voice)   └──────┬──────┘
                                          │
                    ┌─────────────────────┼─────────────────────┐
                    ▼                     ▼                     ▼
             ┌─────────────┐       ┌─────────────┐       ┌─────────────┐
             │  Providers  │       │    Tools    │       │   Voice     │
             │ (8+ LLMs)   │       │  (29 tools) │       │ (STT/TTS/   │
             └─────────────┘       └─────────────┘       │  Wake/VAD)  │
                                                         └─────────────┘

Project Structure

nokton/
├── backend/                 # Python FastAPI + Agent
│   ├── main.py             # WebSocket server, REST API
│   ├── config.py           # Config loading (defaults → user → env)
│   ├── agent/              # Agent engine, conversation, compression
│   ├── providers/          # 8 LLM providers + registry
│   ├── tools/              # 29 tools + registry, schema, permissions
│   ├── voice/              # STT, TTS, Wake word, VAD
│   └── util/               # API keys (encrypted), audit log, images
├── frontend/               # Electron + React
│   ├── main.js             # Electron main, backend lifecycle
│   ├── preload.js          # IPC bridge
│   └── src/                # React app (components, hooks, stores)
├── scripts/                # install.bat, run.bat
├── skills/                 # SKILL.md files (extensible)
├── nokton.json             # Default config
└── requirements.txt        # Python deps

Development

# Backend
cd backend && python -m main

# Frontend (dev mode with hot reload)
cd frontend && npm run dev

# Build frontend
cd frontend && npm run build

# Package Electron app
cd frontend && npm run package

Contributing

1. Fork and clone
2. Create feature branch
3. Make changes (one logical change per commit)
4. Run lint/typecheck: cd frontend && npm run lint (add if missing)
5. Test manually: scripts\run.bat
6. Submit PR with clear description

Commit style: fix tool confirmation flow not Fix: resolved issue with tool confirmation WebSocket handling

License

MIT — use freely, contribute back.

Status

Alpha — core scaffold complete, integration in progress. See PLAN.md for roadmap.

Why Nokton?

Existing options:

• ChatGPT Desktop: Cloud-only, no desktop control, no local voice
• OpenCode: Great for coding, no desktop automation, no voice
• OpenClaw: Desktop automation, no voice, no multi-provider
• Local LLM UIs: Chat only, no tools, no voice

Nokton combines: voice + desktop control + any LLM + local-first + open source