AI Chat

AI Chat (Runs Offline in Your Browser)

Chat with a small AI language model that runs entirely on your device. Nothing you type is sent to a server: the model is downloaded once, saved in your browser and runs on your GPU with WebGPU. Choose a speaking style, adjust the temperature and copy the conversation when you are done.

This is a small AI model. It often makes mistakes, does not know recent events and is weak at facts, numbers and calculations. Always check important information yourself.

1. Load the AI

AI model

Everything runs in your browser. Your messages are not sent to our server, Hugging Face or anyone else.

2. Chat

Speaking style and temperature

Ask anything, or try an example. The AI loads automatically when you send your first message.

Enter to send, Shift + Enter for a new line.

The conversation is not saved. It disappears when you close or reload the page, so use “Copy conversation” to keep it.

  • This is a small AI model, so it often states things that are not true, makes calculation errors and writes unnatural sentences. It does not know about events after its training data (recent news, prices, law changes and so on). Do not use it for important decisions without checking.
  • The conversation is not saved anywhere. It disappears when you close or reload the page.
  • The AI runs on your device’s GPU and needs a browser that supports WebGPU (the latest Chrome or Edge, Safari 26 or later, etc.). It may not run, or may be slow, on older devices or phones with little memory.
  • The first time, the AI model (Light about 300 MB / Standard about 890 MB) is downloaded from Hugging Face, and the WebGPU program for the model (about 6 MB) from GitHub; both are saved in your browser. Clearing this site’s data in your browser settings deletes them.
  • AI models: Qwen2.5-0.5B-Instruct and Qwen2.5-1.5B-Instruct (Alibaba Cloud, Apache-2.0; MLC builds mlc-ai/Qwen2.5-*-Instruct-q4f16_1-MLC). Runtime: WebLLM (Apache-2.0).

About this tool

A free AI chat with no sign-up that runs a small language model (Qwen2.5) inside your browser. It works like a local LLM on your device’s GPU with WebGPU, so your messages are never sent to a server and you can use it for drafts you cannot share outside. The model is downloaded once and saved in your browser; after that it starts without downloading and works offline once loaded. As a small model, it often makes mistakes and does not know recent events.

How to use

  1. Under “1. Load the AI”, choose the Light (about 300 MB) or Standard (about 890 MB) model.
  2. Press “Load the AI”. The first time, the download progress is shown (the AI also loads automatically when you send your first message).
  3. Type a message and press “Send” (or Enter). The reply appears as it is written; press “Stop” to end it early.
  4. Open “Speaking style and temperature” to choose a style such as polite business writing or simple explanations, and to adjust the temperature.
  5. Use “Copy conversation” to keep the chat, or “Reset conversation” to start over.

Useful for

  • Drafting emails or documents with confidential details without sending them to an online AI service
  • Trying an AI chat without creating an account
  • Getting help with rewording or ideas where there is no internet connection
  • Seeing how well a local LLM runs in the browser on your own device

FAQ

Is this AI chat free and without sign-up?

Yes. There is no charge, no account and no usage limit. Speed depends on your device because the AI runs on it.

Is my conversation sent anywhere? Is it private?

No. Your messages and the AI’s replies are processed only in your browser and are not sent to our server or to any AI service. The only downloads are the AI model (from Hugging Face) and its WebGPU program (from GitHub) the first time.

Does it work offline?

Yes, after the first download. The model is saved in your browser, so it starts without downloading and you can chat without an internet connection once it is loaded. You still need a connection to open the page.

Which browsers and devices are supported?

It needs WebGPU, available in the latest Chrome and Edge, Safari 26 and later, and others. It may not run on devices without WebGPU or with little GPU memory; the page tells you if so.

How accurate is it compared with ChatGPT?

These are small models (0.5B and 1.5B parameters), so they often get facts and calculations wrong and do not know recent events. Use it for drafts and ideas, and always check important information.

Contact