Is my conversation sent to a server?

No. The model runs directly in your browser on your own device, so your messages and the replies are processed locally and are never uploaded.

Which models can I use?

You can choose from several compact chat models of different sizes. A smaller model downloads faster and runs on more devices, while a larger one gives better answers but needs a bigger download and more memory.

How do I know which models are already downloaded?

The model list marks each one that is already downloaded and cached in your browser, so you can see which will load instantly and which need a one-time download. You can remove any download to reclaim space.

Why does it need WebGPU?

Running a language model needs your device's graphics card to be fast enough to use. WebGPU is the browser feature that provides that access; without it the tool would be too slow, so it asks for a supported browser instead.

After a model is downloaded, yes — because it runs entirely on your device, you can keep chatting with no network connection.

How big is the download?

It depends on the model, from a few hundred megabytes for the smallest to a couple of gigabytes for the largest. Each model downloads once and is cached, and the size is shown before you download it.

Local AI Chat

Chat with an AI assistant that runs entirely in your browser, on your own device. Pick a model, and your conversation is never uploaded.

The chat model runs locally in your browser, so your prompts and the replies are processed on your device and are never uploaded.

Need a quick summary instead? Try the AI Text Summarizer.

About Local AI Chat

This local AI chat runs a real language model directly in your browser, using your device's graphics card for the heavy lifting, so your conversation never leaves your machine. Choose a model to suit your device — a small one downloads faster and runs on more hardware, a larger one answers better — and the first time you pick it the weights download once and are cached, so later chats start instantly and even work offline. The screen shows which models you have already downloaded, so you can tell at a glance which will load instantly and which need a one-time download, and you can remove a model's files any time to reclaim space. It is useful for drafting and rewriting text, brainstorming, explaining code or concepts, and quick questions you would rather not send to a cloud service. Because everything runs on your own device, the model needs a browser with WebGPU support and enough memory; on unsupported browsers the tool tells you rather than failing silently.

Features

A real chat model that runs on your device — the conversation is never uploaded
Choose from several models, from a small fast one to a larger higher-quality one
See which models are already downloaded, and remove any to reclaim space
One-time model download that is cached, so later chats start instantly and work offline
Streaming replies you can stop at any time, with a one-click new chat
A clear message when a browser doesn't support the required WebGPU

How to use the Local AI Chat

Pick a model — smaller downloads faster, larger answers better. Downloaded models are marked.
Type a message and press Enter; the first use of a model downloads it once, then it runs locally.
Read the streaming reply, and press Stop any time to cut it short.
Use New chat to start over, or remove a model's download to reclaim space.

Example

Input

Rewrite this to sound friendlier: 'Your request has been denied.'

Output

Thanks for reaching out! Unfortunately we're not able to approve this one — but I'm happy to help you find another way forward.

The model runs on your device, so the prompt and reply stay private.

Common errors & troubleshooting

The tool says your browser isn't supported. — On-device chat needs WebGPU. Use the latest Chrome or Edge on desktop, or Chrome on Android; other ArrayKit tools still work everywhere.
The first message takes a while before a reply starts. — The model downloads once on first use and is then cached; the screen shows a progress bar, and later chats begin almost instantly.
Loading a model fails or runs out of memory. — Larger models need more graphics memory. Pick a smaller model, or close other GPU-heavy tabs and try again.
You want to reclaim the disk space a model used. — Each downloaded model shows a remove control; use it to delete that model's cached files at any time.

Frequently asked questions

Is my conversation sent to a server?: No. The model runs directly in your browser on your own device, so your messages and the replies are processed locally and are never uploaded.
Which models can I use?: You can choose from several compact chat models of different sizes. A smaller model downloads faster and runs on more devices, while a larger one gives better answers but needs a bigger download and more memory.
How do I know which models are already downloaded?: The model list marks each one that is already downloaded and cached in your browser, so you can see which will load instantly and which need a one-time download. You can remove any download to reclaim space.
Why does it need WebGPU?: Running a language model needs your device's graphics card to be fast enough to use. WebGPU is the browser feature that provides that access; without it the tool would be too slow, so it asks for a supported browser instead.
Can I chat offline?: After a model is downloaded, yes — because it runs entirely on your device, you can keep chatting with no network connection.
How big is the download?: It depends on the model, from a few hundred megabytes for the smallest to a couple of gigabytes for the largest. Each model downloads once and is cached, and the size is shown before you download it.

Related tools

AI Text Summarizer — Summarize long text into a short summary — the AI model runs in your browser, nothing is uploaded.
AI Sentiment Analyzer — Detect the tone of text as positive or negative with a confidence score — runs in your browser.
AI Entity & PII Extractor — Extract names, organizations, locations, emails, phone numbers and IPs from text — runs in your browser.
AI Text Cleaner — Detect and remove invisible characters and typography artifacts left by AI models and copy-paste.
Word & Character Counter — Count words, characters, sentences, lines, bytes and approximate tokens, live.
Token Counter — Count tokens for GPT, Claude and Gemini live — exact OpenAI counts plus Claude/Gemini estimates.