Free online tools to generate, calculate,
convert, format, encode, and play.
 

Browser LLM

Chat with an AI model running entirely in your browser via WebGPU. Your conversations never leave your device. No API key, no server, no account. Select a model, wait for the one-time download, and start chatting.

WebGPU not available. This tool requires WebGPU, which is supported in Chrome 113+, Edge 113+, and Safari 18+. Make sure hardware acceleration is enabled in your browser settings.
Select a model and click "Load Model" to begin. The model downloads once and is cached for future visits.

How It Works

  1. Pick a model from the dropdown. Smaller models are faster to download and run; larger ones give better answers.
  2. Click "Load Model". The model weights download once from a CDN and are cached in your browser's storage. Subsequent visits skip the download.
  3. Chat by typing a message and pressing Enter or the Send button. The model runs inference on your GPU via WebGPU.
  4. Everything is private. No data leaves your browser. Close the tab and the conversation is gone.

About

This tool uses WebLLM by MLC to run open-source language models directly in your browser. The models are compiled to run on your GPU through WebGPU, achieving 10-30+ tokens per second on modern hardware.

  • WebGPU required: Chrome 113+, Edge 113+, or Safari 18+. Firefox support is experimental.
  • VRAM: Small models need ~1 GB, large models need 3-4 GB of GPU memory.
  • First load: The model download can take 30-120 seconds depending on your connection. After that it's cached.
  • Models: SmolLM2, Qwen2.5, Llama 3.2, and Phi-3.5 are open-weight models from Hugging Face, Alibaba, Meta, and Microsoft respectively.

Embed This Util

You can embed this util on your own site as a widget. Adding ?embed=1 to the URL loads a compact version with just the tool itself; no header, menu, or documentation. Paste this snippet into your HTML:


    

Copy snippet Adjust the height to taste.



Feedback

Help us improve this page by providing feedback, and include your name/email if you want us to reach back. Thank you in advance.


Share with