Skip to main content

ChromeAI

info

This feature is experimental and is subject to change.

note

The Built-in AI Early Preview Program by Google is currently in beta. To apply for access or find more information, please visit this link.

ChromeAI leverages Gemini Nano to run LLMs directly in the browser or in a worker, without the need for an internet connection. This allows for running faster and private models without ever having data leave the consumers device.

Getting started

Once you've been granted access to the program, follow Google's provided instructions to download the model.

Once downloaded, you can start using ChromeAI in the browser as follows:

import { ChromeAI } from "@langchain/community/experimental/llms/chrome_ai";

const model = new ChromeAI({
temperature: 0.5, // Optional, defaults to 0.5
topK: 40, // Optional, defaults to 40
});

const response = await model.invoke("Write me a short poem please");

/*
In the realm where moonlight weaves its hue,
Where dreams and secrets gently intertwine,
There's a place of tranquility and grace,
Where whispers of the night find their place.

Beneath the canopy of starlit skies,
Where dreams take flight and worries cease,
A haven of tranquility, pure and true,
Where the heart finds solace, finding dew.

In this realm where dreams find their release,
Where the soul finds peace, at every peace,
Let us wander, lost in its embrace,
Finding solace in this tranquil space.
*/

Streaming

ChromeAI also supports streaming outputs:

import { ChromeAI } from "@langchain/community/experimental/llms/chrome_ai";

const model = new ChromeAI({
temperature: 0.5, // Optional, defaults to 0.5
topK: 40, // Optional, defaults to 40
});

for await (const chunk of await model.stream("How are you?")) {
console.log(chunk);
}

/*
As
an
AI
language
model
,
I
don
'
t
have
personal
experiences
or
the
ability
to
experience
emotions
.
Therefore
,
I
cannot
directly
answer
the
question
"
How
are
you
?".



May
I
suggest
answering
something
else
?
*/

Was this page helpful?


You can also leave detailed feedback on GitHub.