Edge AI

This demo runs Google's Gemini Nano locally in Chrome via the Prompt API.

Prerequisites: Chrome 138+, macOS 13+, 22 GB free storage, >4 GB VRAM, unmetered network. Model (~2-3 GB) downloads on first use.

Status: Checking API availability...
Debug Info:
URL:
Protocol:
Host:
Port:
Chrome Version:
User Agent:

Setup Instructions for Chrome AI:

  1. Chrome Version: Make sure you have Chrome 138+ (currently )
  2. Enable AI Features (ALL required):
    • Go to chrome://flags/#optimization-guide-on-device-model → "Enabled BypassPerfRequirement"
    • Go to chrome://flags/#prompt-api-for-gemini-nano → "Enabled"
    • Go to chrome://flags/#ai-prompt-api-for-multimodal-input → "Enabled"
    • IMPORTANT: Go to chrome://flags/#ai-prompt-api → "Enabled"
  3. Restart Chrome after enabling flags
  4. Download Model: Visit chrome://on-device-internals to check model status
  5. Hardware Requirements: macOS 13+, 22GB free storage, 4GB+ VRAM
🔧 Alternative Setup (Bypasses Origin Trial):
Method 1 - Launch Chrome with Flags (Recommended):
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --enable-ai-prompt-api --enable-ai-prompt-api-multimodal-input --enable-experimental-web-platform-features

Method 2 - Origin Trial Token:
1. Go to https://developer.chrome.com/origintrials/
2. Register for "AI Prompt API Multimodal Input" trial
3. Add your localhost URL (e.g., http://localhost:3000)
4. Replace the origin trial token in this HTML file

Method 3 - Use file:// protocol:
Save this HTML file locally and open it directly with file:// (no server needed)