gemini-nano

Experimental Gemini Nano on-device inference demos across browser, Python, and HTTP surfaces.

Components

Directory	Language	Port	Status
api-server/	Python (FastAPI + Ollama)	8080	Stable
chrome-bridge/	Node.js / Express	8081	Stable
python-mediapipe/	Python (MediaPipe)	—	Stable
chrome-demo/	HTML / JavaScript	— (file)	Stable

Architecture

graph TD
    A[chrome-demo\nindex.html] -->|window.ai Prompt API| B[Chrome Canary\nGemini Nano built-in]
    C[HTTP client] -->|POST /v1/chat/completions| D[chrome-bridge\nserver.js :8081]
    D -->|Puppeteer + window.ai| B
    E[HTTP client] -->|POST /generate| F[api-server\napp.py :8080]
    F -->|Ollama REST API| G[Ollama\nlocal model]
    H[terminal] -->|python run_nano.py| I[python-mediapipe\nMediaPipe LLM Inference]
    I -->|.task model file| J[Gemma 2B-IT\nor Gemini Nano weights]

Loading

Quick Start

api-server (Python + Ollama)

cd api-server
python -m venv .venv && .venv\Scripts\activate    # Windows
pip install -r requirements.txt
python app.py
# → http://localhost:8080

chrome-bridge (Node.js)

cd chrome-bridge
npm install
npm start
# → http://localhost:8081

Test without Chrome:

npm test

python-mediapipe

cd python-mediapipe
pip install -r requirements.txt
python run_nano.py "What is Gemini Nano?"
# Prints download instructions if model.task is absent (exits 0)
python run_nano.py --model-path /path/to/gemma2b-it.task "Your prompt"

chrome-demo

Open chrome-demo/index.html directly in Chrome Canary with:

chrome://flags/#prompt-api-for-gemini-nano  →  Enabled

No server required.

Setup

See SETUP.md for Chrome flag configuration and model download instructions.