Active Mirror / Sovereign Mode

🔒 Sovereign Mode

True privacy. AI runs entirely on your device via WebLLM. Your conversations never leave your browser.

How It Works

1. Toggle to Sovereign Mode — Click the mode toggle in the header
2. Download Model (~2GB) — One-time download of Phi-3.5 Mini
3. Run Locally — Model runs in your browser via WebGPU/WebLLM
4. Zero Network — After download, works completely offline

Technology Stack

Component	Technology
Model	Phi-3.5-mini-instruct-q4f16_1-MLC
Runtime	WebLLM (MLC AI)
Acceleration	WebGPU (falls back to WASM)
Model Size	~2GB (quantized 4-bit)
Context Window	4096 tokens

Cloud vs Sovereign

Aspect	☁️ Cloud Mode	🔒 Sovereign Mode
Model	Llama 3.3 70B	Phi-3.5 Mini (3.8B)
Quality	Higher (larger model)	Good (optimized small model)
Speed	Fast (~1-2s)	Slower (~5-10s)
Privacy	Data sent to Groq	100% local
Internet	Required	Not required (after download)
First Load	Instant	~2GB download
Cost	Free (Groq API)	Free (your hardware)

Requirements

• Browser: Chrome 113+, Edge 113+, or any browser with WebGPU support
• Storage: ~2GB free space for model cache
• RAM: 4GB+ recommended
• GPU: WebGPU-capable (most modern GPUs); falls back to CPU if not available

Implementation

// WebLLM Engine Singleton
class SovereignEngine {
    constructor() {
        this.engine = null;
        this.isReady = false;
    }

    async init(onProgress) {
        const webllm = await import('@mlc-ai/web-llm');
        this.engine = await webllm.CreateMLCEngine(
            'Phi-3.5-mini-instruct-q4f16_1-MLC',
            {
                initProgressCallback: (progress) => {
                    this.loadProgress = progress.progress || 0;
                    onProgress?.(progress);
                }
            }
        );
        this.isReady = true;
    }

    async generate(systemPrompt, userMessage) {
        const response = await this.engine.chat.completions.create({
            messages: [
                { role: 'system', content: systemPrompt },
                { role: 'user', content: userMessage }
            ],
            temperature: 0.7,
            max_tokens: 200
        });
        return response.choices[0]?.message?.content || '';
    }
}

Privacy Guarantees

The only network request is the initial model download from the WebLLM CDN. After that, everything runs locally.