Question 1

Is this really private? How does it work?

Accepted Answer

Yes, 100% private. The AI model downloads to your browser and runs entirely on your device using WebGPU. No data is ever sent to any server. You can verify this by checking your browser's Network tab - you'll see zero conversation uploads.

Question 2

Is this free? Are there any limits?

Accepted Answer

Completely free with unlimited use. Since all processing happens on your device, there are no server costs. No account required, no API keys, no hidden fees.

Question 3

What AI models are available?

Accepted Answer

We offer several state-of-the-art open models: Qwen3 0.6B (fastest, ~300MB), Llama 3.2 1B (balanced, ~700MB), Granite 3.0 2B (enterprise-grade, ~1.2GB), Ministral 3 3B (256K context, ~2GB), and SmolLM3 3B (best quality, ~1.8GB).

Question 4

Why is WebGPU required?

Accepted Answer

WebGPU provides hardware-accelerated AI inference directly in your browser. It's the only way to run large language models at reasonable speeds client-side. Without GPU acceleration, inference would be too slow to be practical.

Question 5

Which browsers support WebGPU?

Accepted Answer

Chrome 113+, Edge 113+, and Safari 18+ (macOS Sequoia/iOS 18) support WebGPU. Firefox support is coming. Chrome on Android works with the WebGPU flag enabled.

Question 6

How much RAM/VRAM do I need?

Accepted Answer

Minimum 4GB RAM for smaller models (Qwen3 0.6B). 6-8GB recommended for larger models. A dedicated GPU helps but isn't required - WebGPU can use integrated graphics.

Question 7

Does it work offline?

Accepted Answer

Yes! After the first download (which caches the model in your browser), the AI works completely offline. Perfect for travel, air-gapped environments, or unreliable internet.

Question 8

How does this compare to ChatGPT?

Accepted Answer

Key differences: (1) Complete privacy - no data collection, (2) Works offline, (3) Free unlimited use, (4) No account required. Trade-off: Browser models are smaller than GPT-4, so responses may be less sophisticated for complex tasks.

Question 9

Why is the first load slow?

Accepted Answer

The first time you select a model, it downloads the AI model files (300MB-2GB depending on model). This only happens once - your browser caches them for instant future use.

Question 10

Is this GDPR compliant?

Accepted Answer

Yes, fully GDPR compliant by design. No personal data is collected, processed, or stored on any server. Everything stays on your device.

Question 11

Can I use this for work/enterprise?

Accepted Answer

Absolutely. Private AI Chat is ideal for handling confidential business information, proprietary code, or sensitive documents. No data ever leaves your browser, making it safe for enterprise use.

Question 12

What are the model capabilities?

Accepted Answer

These models can help with general knowledge questions, writing assistance, brainstorming, coding help, summarization, and more. They're general-purpose language models optimized for conversational AI.

Question 13

Does my conversation history persist?

Accepted Answer

Conversation history only exists in your current browser session. Refreshing the page clears all messages. Nothing is stored permanently - for maximum privacy.

Question 14

Can I use multiple models?

Accepted Answer

You can switch models by unloading the current one and loading a new one. Only one model can be active at a time due to memory constraints.

Question 15

What's the difference between the models?

Accepted Answer

Smaller models (Qwen3 0.6B) are faster but simpler. Larger models (SmolLM3 3B, Granite 3.0 2B) provide better reasoning and longer responses but need more resources. Choose based on your device capabilities.

Private AI Chat

Why Private AI Matters

🔒 Complete Privacy

📱 Works Offline

🚀 No Account Required

⚡ WebGPU Acceleration

🤖 Latest Models

💰 Zero Cost

Common Use Cases

Frequently Asked Questions