This chat needs WebGPU.
The whole model runs on your device — no backend, no keys — and that needs WebGPU.
Supported: Chrome 113+, Edge 113+, Firefox 141+ (Windows/macOS), Safari 18+ (macOS). On
Windows with a recent NVIDIA/AMD GPU, WebGPU should be on by default in Chrome/Edge — if
it's not, the most common cause is hardware acceleration being disabled in browser
settings.
📦
Install the local model
This chat needs to install Hermes-3 8B (~4.5 GB) into your browser. The full
model runs on your device — no backend, no API key, no third-party LLM endpoint. After
installing, chat is instant and works offline.
Stores in your browser's cache (OPFS). You can clear it anytime from browser settings.
First load takes a minute or two; everything after that runs from cache.
Ask me anything.
The model runs locally in your browser — no backend, no API key.
Tip: drag a text file, PDF, or code file onto the chat to ask questions about it.
📄
Drop to attach
Text, code, JSON, CSV, or PDF — analyzed locally in the model.
Free local LLM in your browser — no API key, no backend
Chat is a static-site AI assistant that runs Hermes-3-Llama-3.1-8B entirely on your device
via WebGPU. No API key. No backend. No third-party LLM endpoint. The model weights (about
4.5 GB) are cached once into your browser's OPFS, after which the page works instantly and
offline. Drop in a text file or PDF and chat about it. The agent knows the Heyming OS app
registry and can launch apps for you.
Common questions about local browser LLMs
Is the model really local?
Yes — inference runs on your GPU through WebGPU. Your prompts and any attached files never
leave your device.
Why an 8-billion-parameter model?
Hermes-3-Llama-3.1-8B is large enough to be useful for everyday tasks but small enough to
fit in 4–6 GB of GPU memory on consumer hardware.
What are the alternatives if WebGPU isn't available?
The page detects unsupported browsers and shows a fallback message. Chrome 113+, Edge 113+,
Safari 18+ on macOS, and Firefox 141+ are the working targets.
Free local LLM browser, browser LLM no API key, WebGPU LLM, WebLLM Hermes 3, on-device AI
browser, local AI assistant browser, run LLM in browser, ChatGPT alternative no signup,
private AI chat browser, offline AI chat browser, no-backend chatbot, local Llama 3 in
browser, browser Llama 3.1, Hermes 3 8B browser, MLC AI WebLLM, WebLLM demo, on-device LLM,
OPFS AI cache, WebGPU AI, browser machine learning chat. Common typos: webgpu llm, webgpu
llms, web gpu llm, web-gpu llm, web glsl, webgl llm, webGPU LLM browser, hermes 3 browser,
hermes-3-8B browser.