Checking what this browser can run…
New chat
What can I help you with?
Questions, explanations, short texts, ideas or a bit of code. Everything stays on your device.
A small model running on your device: it can be wrong and knows nothing about recent news. Check anything important.
Runs in your browser: what you write never leaves your device, and the chats are kept only in this browser (deleting them, or clearing the browser data, removes them). The first use downloads the model from Hugging Face and the AI engine (about 6 MB) from jsDelivr; after that it works offline. Models: Granite 4.0 350M and 1B by IBM, licensed Apache 2.0, converted to ONNX by onnx-community. Engine: Transformers.js and ONNX Runtime.
Embed this tool on your site
Free and without an account: copy the code and paste it into your website, blog or docs.
About the Private AI Chat
Online AI chats keep what you type on their servers. This one runs an open language model inside your browser, so your questions, the answers and the whole history stay on your device, with no account and no limit on messages. Download the model once (350 to 580 MB) and chat whenever you like, also without an internet connection. It is a small model: good for everyday questions, explanations, short texts and simple code.
How to use the Private AI Chat
-
Pick a model and download it
Choose Fast (Granite 4.0 350M, about 350 MB with WebGPU or 580 MB without, runs on any recent device) or Smarter (Granite 4.0 1B, about 1.25 GB, needs WebGPU). Click Download, or just send your first message: the model comes from Hugging Face once and is kept in your browser.
-
Ask your question
Type in the message box and press Enter, or Send (Shift + Enter adds a new line). The answer appears as it is written; Stop ends it early and Copy puts it on the clipboard.
-
Come back to your chats
Every conversation is saved in this browser and listed under Chats (History on a phone). New chat starts a fresh one, the bin deletes one, and Delete all chats clears the history.
When to use it
Questions you would rather keep private
Ask about health, money, work or anything personal without it being stored by an AI company: the conversation never leaves your device.
Quick help with writing and code
Draft a short email, rephrase a paragraph, get a list of ideas or a small code snippet, then adjust it yourself.
AI with no connection
Download the model before a flight, a trip or a day with bad coverage, and keep asking questions with no data plan or Wi-Fi.
Frequently asked questions
Which models does it use, and under what license?
Granite 4.0 350M (Fast) and Granite 4.0 1B (Smarter), open models by IBM trained for English, Spanish and other languages, released under Apache 2.0, which allows commercial use. They run in the ONNX conversion published by onnx-community on Hugging Face, with Transformers.js and ONNX Runtime.
Is what I write sent anywhere?
No. The model runs in your browser, and neither your messages nor the answers are sent to Tooly or anyone else. The only downloads are the model, from Hugging Face, and the AI engine (about 6 MB), from jsDelivr, the first time; they contain nothing you wrote. The chat history is saved in your browser’s local storage: it stays on this device and this browser, and deleting the chats or clearing your browser data removes it.
How good is it compared with ChatGPT, Gemini or Claude?
Much weaker. Those run models hundreds of times larger on big servers; this one fits in your browser. It handles everyday questions, explanations, short texts and simple code reasonably well, but it can invent facts, make mistakes with numbers and logic, lose the thread in long conversations, and it knows nothing after its training or about current news, since it has no internet access. Check anything important. The Smarter model gives clearly better answers if your device supports WebGPU.
What does my device need, and how fast is it?
A recent browser. With WebGPU (Chrome, Edge and other Chromium browsers on most computers, and newer Safari) the model runs on your graphics card and writes fast, usually tens of tokens a second with the Fast model. Without it, it runs on your processor through WebAssembly, more slowly, and only the Fast model is offered. The first answer after opening the page takes a few seconds longer while the model loads into memory. Under each answer you can see how long it took.
Does it remember the conversation, and is there a limit?
Yes, within each chat: the model reads the recent messages before answering, and when a conversation gets long the oldest messages drop out of what it reads (they stay in your history). Each message can be up to 6,000 characters and each answer about 700 words. There is no limit on the number of chats or messages other than your browser’s storage space.
Related tools
Browse all free online tools on Tooly — every one runs in your browser, with no account required.