Chrome Built-In AI

TL;DR: Built-in AI currently depends on Google-internal (Google Chrome only) code that is not publicly available to Chromium/CEF. Support for OS built-in models may be added in the future.

Google Chrome is developing AI capabilities that utilize Gemini Nano models (locally-bundled Google LLM) via Web APIs on desktop. WebAPI examples are available here and here.

Advantages:

• Same AI model that already does cool things on Pixel phones (Android docs here).
• Uses on-device capabilities for faster response and (probably) lower cost.
• Better for privacy because no personal data is sent over the network.
• Works offline.

Disadvantages:

• Not all devices supported, so may still require a cloud-based fallback.
• May be too generic compared to a specifically trained model, even with LoRA fine-tuning.
• May be free to use in Google Chrome but not in CEF (requires Google API keys?).

Built-In AI APIs

There are several built-in AI APIs at different stages of development. Some are available in Chrome 138, some are available to all developers in origin trials, and some are only available to Early Preview Program (EPP) participants.

Translator API

The Translator API is available from Chrome 138 stable. Translate user-generated and dynamic content on request.

Use cases

• Users can enter a request in their first language, which you can identify with the Language Detector API. Then, use the Translator API to convert the request to your business operating language and send it to a support agent.
• In a social network application, users can request a translation on-demand when a post appears on their timeline in a language they don't speak.

Language Detector API

The Language Detector API is available from Chrome 138 stable. You can use this API to detect the language of input text. This is a key part of the translation process, as you may not always know the input language for translation.

Use cases

• Determining the unknown source language for a following translation to a known target language, so the user doesn't have to specify both.
• Labeling texts, for example, to improve screen reader pronunciation in online social networking sites.

Summarizer API

The Summarizer API is now available for local experimentation to our EPP participants. With this API, you use built-in AI to condense long-form content. Shorter content can be more accessible and useful to users.

Use cases

• Overview of a meeting transcript for those joining the meeting late or those who missed the meeting entirely.
• Key points from support conversations for customer relationship management.
• Sentence or paragraph-sized summaries of multiple product reviews.
• Key points from long articles, to help readers determine if the article is relevant.
• Generating draft titles for an article.
• Summarizing questions in a forum to help experts find those which are most relevant to their field of expertise.

Writer and Rewriter APIs

The Writer API helps you create new content that conforms to a specified writing task, while the Rewriter API helps revise and restructure text. Both APIs are part of the Writing Assistance APIs explainer.

Use cases

• Write based on an initial idea and optional context. For example, a formal email to a bank asking to increase the credit limit based on the context that you're a long-term customer.
• Refine existing text by making it longer or shorter, or changing the tone. For example, you could rewrite a short email so that it sounds more polite and formal.

Prompt and Script Tools APIs

With the Prompt API, a page can send natural language requests to Gemini Nano in Chrome. With the Script Tools API, a page can expose its capabilities to agents outside of Chrome.

The Prompt API is available from Chrome 138 stable, only for Chrome Extensions.

In Chrome Extensions

With the Prompt API in Chrome Extensions, you can experiment in a real environment. Based on your findings, we can refine the API to better address real-world use cases.

Use cases

• Function Calling: A developer uses the Prompt API to call the on-device model to enhance their site's own features. The page defines the tools and initiates the prompt, controlling the entire interaction. This is an "inward-facing" capability for the site to use AI.
• Script Tools: An external agent is the orchestrator. A web page uses the Script Tools API to expose its capabilities to an external agent (e.g. a browser or OS-level agent, or features and apps including accessibility related opportunities). The external agent, acting on a user's behalf, discovers and invokes these tools to perform tasks on the page. This is an "outward-facing" capability for the site to work well when used through an external agent.
• Alternative to MCP server integrations.

Proofreader API

The Proofreader API is available from Chrome 139 Canary for local experimentation Early Preview Program participants. With this API, you can provide interactive proofreading for your users in your web application or Chrome Extension.

Use cases

• Correct a document the user is editing in their browser.
• Help your customers send grammatically correct chat messages.
• Edit comments on a blog post or forum.
• Provide corrections in note taking applications.