Building Browser-Native AI Runtimes: A Hands-On Workshop with WebLLM, WASM & WebWorkers [Workshop]

Barış Güler -

What does it mean to run AI in the browser? In this hands-on workshop, participants will build a in-browser AI runtime using eg. WebLLM, WASM, and WebWorkers where inference, agent logic, and orchestration all run locally. We’ll treat the browser as an execution environment, not a thin client.

Modern web-based AI applications are typically thin clients backed by remote inference APIs. While convenient, this model centralizes computation, limits privacy, and treats the browser as a passive UI rather than a capable runtime.

This workshop explores an alternative: browser-native AI, where large language model inference, agent logic, and orchestration run entirely on the client. Participants will build a hands-on prototype using WebLLM for in-browser inference, WebWorkers for isolation and responsiveness, and WebAssembly for portable, high-performance agent logic.

Rather than focusing on tooling or demos, the workshop emphasizes architectural reasoning. Participants will examine how threading becomes a design concern when inference is local, where WASM provides real advantages over JavaScript, and what trade-offs emerge around startup time, memory usage, and user experience. The browser is treated as a serious execution environment — not a thin client for cloud AI.

By the end of the session, attendees will leave with a reusable runtime pattern for browser-native AI systems, along with a clear understanding of when this approach is appropriate, where it breaks down, and how it differs fundamentally from server-centric AI architectures.