№ 01entry — browser AI lab
AI that never leaves your tab.
Local inference, client-side RAG, and voice agents — running on your GPU, on this page, right now. No cloud, no API keys.
$ webml check-runtime
- webgpu…
- threads…
- memory·
- model cache…
- servers0 — it all runs in this tab
What this is
WebML is a personal lab for browser-based AI. Embeddings, inference, RAG — running locally, without servers. Driven by curiosity, shipped in public.
Local-first
AI that runs on your device. Models loaded once, executed entirely in the browser tab.
Privacy-native
Model inference runs on your device. The site uses analytics; see the privacy page for details.
Open lab
Experiments built in public. Source available, ideas welcome, nothing gated.
Experiments
The Lab
Browser AI experiments — ranging from alpha to coming soon.
- Live
LLM Playground
Chat with Phi-3.5, Llama 3.2, Gemma 2, or Mistral — running 100% in your browser via WebGPU. No API key, no server, nothing leaves your tab.
- Alpha
Todo AI
A todo list with an embedded AI chatbot running entirely in the browser. No server, no API key — compare WebLLM and Chrome's Prompt API side by side.
- Live
Detectiks
Detect whether an image was generated by AI or captured by a human. Real-time, in the browser.
- Alpha
webmcp.js
A browser-side library that turns any website into a structured capability layer for AI agents.
Live on this page
Talk to this site.
The floating mic in the corner is a full voice agent: speech recognition, an 8B-parameter planner, and text-to-speech — three models in three Web Workers, all in this tab. Say “open the playground” and watch it navigate.
Nothing downloads until you tap the mic and confirm. Roughly 1.6 GB of weights, cached after the first run. Desktop Chrome or Edge with WebGPU.
Read how it's builtWriting
From the lab
Notes on browser-native AI. Practical and opinionated.
Building a Browser Voice Agent: Consent, Workers, and Tool Boundaries
A code-backed look at WebML’s voice agent: what loads on demand, where work runs, and where the current tool boundary still needs tightening.
LiteRT.js Moves the TFLite Runtime Into the Browser
Google shipped LiteRT.js, a browser runtime for .tflite models that targets Wasm, WebGPU, and experimental WebNN instead of JavaScript kernels.
WebMCP's Second API Skips JavaScript Entirely. It's Two HTML Attributes.
Add toolname and tooldescription to an existing form and Chrome derives a JSON Schema tool from your labels — no navigator.modelContext, no hand-written schema.
FAQ
Common questions
The short version of what this lab is about.
What is WebML?
WebML (webml.ai) is a browser AI lab that runs machine learning models entirely in the browser using WebGPU, ONNX Runtime Web, and Transformers.js — with no server, no API key, and no cloud required. It includes live experiments, articles, and an interactive playground.
What is browser-native AI?
Browser-native AI runs inference directly inside a web browser using WebGPU compute shaders and WebAssembly. Models execute on the user's device GPU or CPU. The application may still fetch models and send analytics, depending on its implementation.
What technologies does WebML use?
WebML experiments use Transformers.js (Hugging Face), WebLLM (MLC-AI), ONNX Runtime Web, Chrome's built-in Prompt API (Gemini Nano), Web Workers for threading, AudioWorklet for audio processing, and WebGPU for GPU-accelerated inference.
Does WebML work offline?
Model inference can run without a remote inference service after weights have loaded. Reopening this website offline is not guaranteed: the page itself may need a network connection, and browser caches can be cleared.
The person behind this
Omri Beladev
Engineer, CTF player, and browser AI experimenter. I build WebML to explore what's possible when AI runs locally — no cloud, no dependencies, just the browser and what it can do.