Web AI Monthly #29: ✨A brief history of client side AI, 🧠new models/demos, and games controlled by JS smartwatches👩💻
🎄Wow, it's December already. 2025 sure did go by fast, but innovation on the client side doesn't stop for holidays - we have a bunch of great updates for you, so get scrolling! Don't miss out on great knowledge, AI models, and demos to try right now 🎉
History of Web AI, where it's at, and its future
After representing client side AI now for several years since 2018, I decided it was time to reflect on how far we've come as a community, and create something that can be referenced in the future to answer common questions: How did all of this come to be? What are its current use cases? Where do we see it going in the future? A go-to 101 for all that may be new to the field. As such, I teamed up with Tyler Mullen - a Web AI legend, friend, and colleague here at Google, to put together a new article that covers exactly this from the low level technologies that make it work (think WebGPU, Wasm, WebGL, WebNN), to the high level applications that are now possible.
Introducing "Life on the edge with Web AI - the future of smarter digital experiences and the agentic internet". The Cloud ☁️ isn't the only future for AI. Learn about Client side AI history, state of the union, and its future role in an evolving agentic internet. I even made a logo for it - must be serious eh?
We wrote this to have a "go to" article for the origins and state of client side AI (its been around since 2017 after all) as we kept being asked the same questions in our DMs.
🐍 The days of relying solely on Python backends for inference are over. Web AI is exploding in usage, and the stack has matured significantly since our humble beginnings
🚀 Run LLMs, Computer Vision, and more entirely in the browser. Real-time. No installs. No API costs.
🌐 If you're building the future of the internet, you need to understand the architecture behind it. Enjoyed the read? Please do give us a share, comment, or like to spread the love🙏
➡️ We're also pleased to announce MIT Technology Review Korea has already translated, converted, and published in Korean too as a 3 part series here!
🌏If any other folk wish to convert to other languages feel free to drop me a DM - would be great to have this available more widely for folk to reference.
Using JavaScript powered smartwatches for client side AI powered gaming!
A short while ago I hosted our 2nd public Web AI Summit. Over 20 talks in 1 day, and 1000 RSVPs, it was a culmination of great minds interested in client side AI here at Google in the bay area.
One of those great minds in attendance this year was Charlie Gerard - an OG Web AI user who has been with us since the TensorFlow.js days. She took this year's swag that I designed with Gordon Williams and used it to upgrade her Web AI gaming demo to use the sensors from the watch!
Now when she throws a punch, or raises the hand in the air, it activates key game movements in real-time taking immersion to a new level. So cool.
Check the video above to see it in action for yourself and give it some love. If you want to learn more about the JS smartwatch we gave out at the summit check this video too:
❤️ You just can't beat the creative minds of JS devs. Keep up the great work all! ❤️
LiteRT.js for depth estimation in the browser
✨ This year Google also launched LiteRT.js - a new client side Web AI runtime that can run machine learning models right in your browser! To test and learn the ropes I found a depth estimation model from Qualcomm that runs in real time on my old NVIDIA 1070 GPU in real-time right in Google Chrome via #WebGPU. Greener pixels mean closer to the webcam, and bluer pixels are further away.
Should work on any dedicated GPU at around this speed, but if you have an integrated GPU, will be slower - still a very fun model to play with as you can see in this post:
➡️Try it yourself on CodePen below as it's running entirely client side - no cost, total privacy, and low latency: https://codepen.io/jasonmayes/pen/YPqrJgv
✅ Hat tip to the LiteRT.js team who have made a great starting point for future inference Matthew Soulanille, Ping Yu, Chintan Parikh 🙏
Finally just for fun I then took my first depth demo above and used it to create a 3D point cloud that you can actually move around:
➡️Try 3D version for yourself here: https://codepen.io/jasonmayes/full/VYaQPXX
Web AI for answering medical questions faster
Wow, its been quite a month. Damyr Hadiiev created this demo powered by Gemma 3N client side version right in the browser to work with medical knowledge (tuning the model to understand additional context for reasoning about ICD-10 codes) allowing a doctor to scan through potentially large corpus of medical information faster to get answers by just asking in natural language.
Obviously original records are referenced so they can confirm no hallucination occurred, but still saves a lot of time, and significantly better than CTRL F where you must type exact words or phrases to locate them in digital docs.
Certainly an area to watch. I look forward to the first success story here where such applications gain approval for use in the medical industry to make doctors more efficient than ever. With Web AI full privacy of patient records is maintained of course making it a natural fit for the healthcare industry.
More amazing models ported to Web AI
Supertonic WebGPU
Introducing Supertonic WebGPU: blazingly fast text-to-speech running 100% locally in your browser. ⚡️Generate free audiobooks in seconds, no installation required.
In this example, Joshua Lochner created a 5-hour audiobook of The Great Gatsby in less than 3 minutes! 🤯
Try it out yourself! 👇
Segment Anything 3 - Web AI Edition
Meta Superintelligence Labs just released SAM 3: Segment Anything with Concepts. A unified model that detects, segments, and tracks objects in images or videos using concept prompts (text or images)!
🤯Model weights now out on 🤗 Hugging Face!👉 https://lnkd.in/dsHqqZHEIt and can even run 100% locally in your browser on WebGPU, enabling real-time image segmentation! ⚡️👉 https://lnkd.in/dcJuAvU7
Enjoyed the content?
Help me help you, by giving us a share with your colleagues in your weekly team updates to get more eyes on great work in the Web AI field. Tag Jason Mayes if you make something noteworthy for future editions. We have subscribers ranging from decision makers (think C-level, VPs, and Directors) to folk on the frontlines using this stuff day to day (SWEs, web engineers and researchers). You never know who may see your creations.
Turn on notifications for timely updates!
Newsletters are great, but I tend to post every month or so. If you want to get the latest as it happens, ensure you turn on notifications for my posts over on LinkedIn so you see more content you love and less AI spam now that LinkedIn is going mainstream.
See you next time!
If you're new to this space and want to learn Web AI, you can get started fast with my free Google Developers course here (no background in AI needed, just a love for JavaScript and curiosity for AI - I will teach you from zero). Or get inspired through our growing collection of Web AI talks on YouTube or via our Show & Tell - I got you either way!
See you next time with even more great content and please do tag me (Jason Mayes ) if you make or find something for future editions - I need your help to find the latest and greatest news lovely #WebAI community as things are moving so fast!
You can also find us over on Discord if you prefer to chat async over there where we also host "Web AI Wednesdays" every month, to gather curious folk just like you, to meet and discuss all things Web AI - everyone welcome no matter what your skill level with Web AI. Reach out to Elisabeth Ashley for details.
Cheers!
Jason Mayes (that Web AI guy).