Chrome Prompt API Test - Gemini Nano Local AI

Setup Instructions

The Prompt API requires Chrome 138+ with specific flags enabled:

Open chrome://flags/#optimization-guide-on-device-model and set to Enabled BypassPerfRequirement
Open chrome://flags/#prompt-api-for-gemini-nano and set to Enabled
Open chrome://flags/#prompt-api-for-gemini-nano-multimodal-input and set to Enabled (for image/audio input)
Click Relaunch to restart Chrome
After relaunch, visit chrome://on-device-internals to check model status

Note: chrome:// links cannot be clicked directly. Copy the URL and paste it into your address bar.

Turn-by-turn chat using the Prompt API (streaming). Advanced model/session controls are in the drawer at the bottom.

Tip: First send will auto-load the model. If it fails, follow the Setup Instructions above.

Check if the Prompt API and Gemini Nano model are available on this device.

Click button to check

Get the model's default and maximum parameter values.

Load the model and create a session with optional custom parameters.

Temperature

Top-K

System Prompt (optional)

No model loaded

Downloading model...

Experimental: Let the model talk with itself

After your message, the model will keep generating alternating turns until you press Stop (or turn this off).

Send a prompt and get the complete response at once.

Send a prompt and see the response stream in real-time.

Constrain the model's response using a JSON Schema.

JSON Schema

Reset the conversation context (preserves system prompt, clears chat history).

Load model first