Nano — Local AI · Private Intelligence

On-device Gemini Nano chatbot. Offline, private, no servers. Runs entirely in Chrome Canary / Dev using the built-in Prompt API (window.ai).

---

Why Nano?

Nano is a tiny, fast web app that proves you can have a useful AI assistant without sending data to the cloud. Everything runs locally:

• Private by design — prompts never leave your browser
• Offline first — works after the model downloads once
• Lightweight — ~5 files, no build step, no npm dependencies at runtime

Features

Core

• Streaming responses from Gemini Nano (on-device)
• Abort/stop generation mid-stream
• Automatic session recovery if Chrome evicts the model
• Context window management (~1,200 tokens sliding window)

UX

• Clean chat UI with markdown (**bold**, *italic*, code, links)
• Edit & resend — double-click any user message to modify
• Typing indicator and live token speed (⚡ tok/s)
• Dark mode (follows system, manual toggle 🌓)
• Mobile-friendly auto-growing textarea
• ARIA roles for screen readers

Controls

• Temperature slider (0 – 1.5)
• Top-K slider (1 – 100)
• Custom system prompt
• Output language: English, Spanish, French, German, Japanese

Data

• Auto-save to localStorage
• Multi-session management (create, switch, delete, rename)
• Export chat as JSON or Markdown
• Import previous conversations
• Smart quota handling — prunes oldest session automatically when storage is full
• Factory reset

Quick Start

1. Requirements

• Chrome Canary or Chrome Dev ≥ 128

• Enable the Prompt API:

  1. Go to chrome://flags/#prompt-api-for-gemini-nano
  2. Set to Enabled
  3. Also enable chrome://flags/#optimization-guide-on-device-model
  4. Relaunch

• Download the model: chrome://components → "Optimization Guide On Device Model" → Check for update

2. Run locally

python -m http.server 5500
# then open http://localhost:5500