Show HN: Open-source multimodal AI that runs in the browser

I built an AI assistant that runs completely in your browser using WebGPU. After the initial model download (~200MB-4GB depending on model choice), it works offline. No API keys. No subscriptions. No data leaves your device.

Browser-based local AI addresses this:

1. Works offline after first load – Local AI makes possible offline AI use. Resilience – Work continues even during outages. Affordability – No recurring cloud or bandwidth costs. Privacy – Data stays on the device, reducing exposure risks.

2. No paywall or subscription – Removing or reducing paywalls on essential AI tools is crucial. Promoting open-source AI platforms provides powerful tools to everyone at no cost, democratizing access to AI technology.

3. Democratizes creative tools – AI tools have the potential to democratize creative processes, making art and design tasks more accessible to individuals without traditional expertise. AI tools can act as equalizers, providing non-professional users with the means to execute complex projects and bring their visions to life.

4. Privacy for sensitive use cases – AI is an alternative way to empower and democratize the access to creative tools and resources by making it easier for individuals with varying levels of artistic skill or expertise to engage in creative expression.

What it does:

Text chat – Switch between 10+ models (SmolLM-360M to Llama 3.2 8B) Vision – Webcam/screen share analyzed by SmolVLM-256M running locally Voice – Whisper speech recognition (VAD-powered hands-free mode) TTS – Browser-native text-to-speech Image generation – SD Turbo via ONNX Runtime (experimental) Research mode – Multi-step research with optional web search Deep thinking – Chain of Thought, Tree of Thoughts, Reflexion modes

Requirements:

Chrome 113+, Edge 113+, Safari 26+ 4GB+ GPU memory for larger models One-time download, then cached/offline