Client side Web AI Agents Building smarter user experiences for a future age
summary
Prototype demonstrates a browser agent that parses voice, extracts structured JSON, and calls existing page functions as tools to book flights and change searches entirely client side using Gemma models and MediaPipe / WebGPU with a cached model download. It runs multimodal Gemma models locally even on an old Nvidia 1070, achieves pose/segmentation at ~120 fps, and claims client-side background blur can avoid 54 billion segmentations per day server cost. Includes a caching proxy, Kakoro TTS and live Spotify API calls, and shows a 2–27B parameter model feasibility.
AI-written summary · about
chapters (9)
- 0:00Introduction to client side AI
- 3:08Why web AI matters now
- 9:08What are AI agents
- 13:21Agentic websites future
- 16:12Live flight search demo
- 20:28Web AI DJ prototype
- 22:11Web AI technology overview
- 24:43Future of agentic web
- 27:50Call to action and Q&A
related talks
22mClient side Web AI Agents: Building smarter user experiences for future agentic ...
24mState of client side AI
02mDiscover Web AI: Client side Agents, Gen AI, and machine learning in the browser
30mLook ma, no hands! Multimodal AI-Agents in the browser
23mWebAI: Building the next user experience in the browser
21mUnlocking Client-Side AI in Your Web App
31mBringing and Running AI Agents in the Browser
22mBoring but Effective: WebAI That Ships