AI in the browser: local LLMs in Angular with WebLLM

Abstract

What if the AI in your app never touched a server? WebLLM runs quantized language models directly in the browser via WebGPU, with no API calls, no latency, no data leaving the machine. The model runs locally, inference happens locally, and your Angular app stays entirely self-contained.

In this talk we’ll explore how WebLLM works, how to integrate it the Angular way using signals and resources, and the strategies to work around its real constraints.

We’ll follow the journey from a first experiment to a fully working browser game with NPC dialogue powered by a local LLM, covering what worked, what didn’t, and what you can bring back to your next project.