We Added AI Tools That Run Entirely in Your Browser

# We Added AI Tools That Run Entirely in Your Browser

Most "free AI tools" online work the same way: you paste your text or code, it gets sent to a server running GPT or Claude, and the response comes back. Your data passes through someone else's infrastructure. You have no idea what happens to it after that.

We wanted to do it differently.

What We Built

Three new AI tools, all available to Pro subscribers:

AI Chat - A full chat interface with streaming responses, customizable system prompts, and conversation history. Pick a model, load it, and start chatting. Works like ChatGPT but everything stays on your machine.

AI Code Explainer - Paste code in any language (JavaScript, Python, Rust, Go, SQL, and more), pick a detail level, and get a clear breakdown of what the code does. Useful for reviewing unfamiliar codebases or learning new languages.

AI Text Summarizer - Drop in a long article, document, or wall of text. Choose between brief (1-2 sentences), detailed (full paragraph), or bullet point summaries. Good for research, email triage, or condensing meeting notes.

How It Works

All three tools are powered by WebLLM, which runs large language models directly in the browser using WebGPU.

Here is what happens when you use one of these tools:

1. You pick a model from the dropdown (5 options ranging from 0.7GB to 2.2GB)
2. The model weights download to your browser's cache (one-time download)
3. The model loads into GPU memory via WebGPU
4. All inference runs locally on your device
5. Nothing is sent to any server - not to us, not to OpenAI, not to anyone

The model download is cached by the browser, so after the first load it's nearly instant on subsequent visits. We also use a shared engine pattern - if you load a model in AI Chat and then switch to AI Code Explainer, the same model stays in memory. No redundant downloads.

Available Models

These are quantized (q4f16_1) versions optimized for browser inference. They won't match GPT-4 quality, but for summarization, code explanation, and general chat they work well - and they do it without any privacy trade-off.

Requirements

WebGPU is required, which means you need a recent browser:

• Chrome 113 or later
• Edge 113 or later
• Firefox Nightly (behind a flag)
• Safari - not yet supported

A dedicated GPU helps with speed but isn't strictly required. Integrated GPUs on recent laptops work fine for the smaller models.

Why Pro Only

Running AI models in the browser is a genuine premium feature. The engineering complexity, model hosting (the initial weight files are served from CDN), and the WebGPU integration make this substantially more expensive to maintain than our standard client-side tools.

We already have two AI tools available to all users - AI Regex Generator and AI CSS Generator - that work with bring-your-own API keys (OpenAI or Anthropic). The WebLLM tools remove that friction entirely: no API key, no account, no configuration. Load and go.

Privacy Guarantee

We updated our privacy policy to explicitly cover local AI processing:

• AI model weights are downloaded from a public CDN to your browser's cache
• All inference happens locally via WebGPU - your prompts and responses never leave your device
• We cannot see, access, or log any of your AI interactions
• Model cache can be cleared through your browser settings at any time

This is the same privacy-first approach we take with every tool on ToolBox. The only difference is the model download, which is a one-time transfer of generic model weights (not your data).

Try It Out

The three AI tools are live now at:

• AI Chat
• AI Code Explainer
• AI Text Summarizer

All require a Pro subscription. If you're already a Pro subscriber, they're ready to use right now.