The AI in Your Pocket: Building Offline Gemini Apps with the Browser – GDG Cloud North West Recap

I recently (Sep 17, 2025) had an incredible session at GDG Cloud North West in Manchester, where attendees explored cutting-edge developments in cloud technology, BigQuery, Wasm in Go, and artificial intelligence. Among the highlights was my presentation, where I delved into the revolutionary concept of running Gemini Nano directly in the browser.

This groundbreaking approach paves the way for a new generation of offline-first, off-device web applications and Chrome Extensions, addressing critical needs for speed, privacy, and accessibility.

My session, titled “The AI in Your Pocket: Building Offline Gemini Apps with the Browser,” demonstrated how developers can leverage a suite of built-in AI APIs to create high-performance, privacy-centric experiences.

The Challenge of Cloud-Based AI

Traditional cloud-based AI often comes with its own set of challenges:

• Need for instant results: Latency can be a significant issue, especially for real-time applications.
• Limited connectivity: AI features become inaccessible offline or with poor internet connections.
• Privacy concerns: Transmitting sensitive data to external servers raises valid privacy concerns.
• Limited usage / behind a paywall: Many advanced AI features are often restricted or come with a cost.

Unlocking On-Device AI with Gemini Nano

Gemini Nano emerges as a powerful solution, enabling on-device AI for enhanced speed, privacy, and accessibility. The core idea is to run Gemini Nano directly within the user’s browser, allowing for offline-first web applications and off-device Chrome Extensions.

Architecture Overview: Built-in AI APIs

A key enabler for on-device inference is a suite of built-in AI APIs. I highlighted several crucial APIs:

• The Prompt API: The core of on-device generative AI, allowing natural language requests to be sent directly to Gemini Nano for seamless, local content generation.
• Translator & Language Detector APIs: Enabling real-time, on-demand language capabilities. All processing happens on-device, ensuring data privacy.
• The Summarizer API: Instantly condenses long-form content, useful for quickly digesting meeting transcripts or summarizing forum discussions.
• Writer & Rewriter APIs: Empowering users with content creation and refinement tools, allowing for new content creation or refinement of existing text by adjusting tone, length, or style, all available offline.
• The Proofreader API: Provides interactive, on-the-fly corrections, ensuring grammatically correct communication directly in the browser.

Benefits of Offline Gemini Apps

The advantages of this approach are compelling:

• Enhanced User Privacy: Data remains securely on the user’s device.
• Reduced Latency: Near-instant results without server roundtrips.
• Offline Accessibility: Functionality persists even without an internet connection.
• Cost Efficiency: Eliminates server-side processing costs for many AI tasks.
• Seamless Integration: AI features are built directly into web applications and Chrome Extensions.

Learning Focus: Implementation and Debugging

The session also focused on equipping attendees with practical knowledge, covering:

• How to implement these built-in AI APIs.
• Strategies for testing on-device AI features.
• Techniques for debugging local AI integrations.

Building, Testing, and Debugging On-Device AI

I provided valuable insights into the development process, including:

• General Flags: chrome://on-device-internals, chrome://flags, and !experimental (developer mode).
• API Flags: Specific flags like chrome://flags/#writer-api-for-gemini-nano, chrome://flags/#rewriter-api-for-gemini-nano, and chrome://flags/#proofreader-api-for-gemini-nano.
• LanguageModel Namespace: Key extension functions like availability() to check model capabilities and create() to start a language model session.
• Capabilities: Parameters such as defaultTopK, maxTopK, defaultTemperature, and maxTemperature.
• More Capabilities: Including initial prompts contexts, session persistence and limits, append() for prompt() and promptStreaming(), response constraints (JSON type & properties), clone(), abort() with signal and AbortSignal, response prefixes, destroy(), and multimodal capabilities.

Demo and Conclusion

A practical demo showcased the setup and deployment of a Gemini Nano application. The session concluded with a powerful message: Gemini Nano in the browser ushers in a new era of AI, empowering developers to create next-generation, high-performance, and privacy-centric AI experiences that work seamlessly whether users are online or offline.

For a deeper dive into the presentation, you can access the slides here: The AI in Your Pocket: Building Offline Gemini Apps with the Browser

You can also explore the code and further resources on the GitHub repository.

About Olorunfemi Davis

I am a Senior Software Engineer with a strategic, platform-focused mindset and a passion for computing, mathematics, and research. I specialize in architecting and optimizing secure, high-performance backend systems and web applications across Google Cloud. As a certified Generative AI Leader from Google Cloud, I possess deep knowledge of Google’s AI offerings and their practical application. Beyond my technical expertise, I am passionate about mentoring, fostering collaboration, and driving excellence within the developer community.

Event Location: Lloyds Banking Group, Westminster House, 11 Portland Street, Manchester, M1 3HU

We extend a huge thank you to Lloyds Banking Group, Westminster House, for hosting this event. If you missed out, be sure to check out upcoming GDG Cloud North events at GDG Cloud North West.

For more information about me, you can visit my profile at g.dev/olorunfemidavis

Event links: See GDG North - Manchester at Google Developer Groups GDG Cloud North West

#Gemini #AI #Google #GDG #Manchester #OfflineAI #ChromeExtensions #Cloud #TechCommunity #Speaking #Nano #Privacy #WebApps