AI in your browser.
Your Browser runs AI on your own computer, phone or tablet directly from the website, without installing a helper or creating a personal account. Once a supported model is loaded, prompt processing uses that device’s memory and graphics resources. Keep web search off for a task that should not send a search query online.
Get started
- Choose Your Browser in the route picker.
- Use the offered automatic choice or a supported browser model when the model picker is available.
- Check the download and memory guidance, then load the model.
- Wait for the ready state before sending a prompt.
The first download can take time. A cached model may load faster later, but browsers can remove cached files when storage is low.
Phones and tablets
You can open the workspace on iPhone, iPad and Android. Running AI on the phone itself also requires WebGPU support and enough memory for the selected model. A browser displaying the app correctly does not establish that every model can run.
| Platform | Browser requirement |
|---|---|
| iPhone and iPad | Safari 26 introduced WebGPU on iOS and iPadOS. Use a current supported OS and browser. |
| Android | Chrome enables WebGPU on supported Android 12+ devices with compatible graphics hardware. Use a current browser. |
| Desktop | Use a browser with WebGPU enabled and a compatible graphics adapter. |
These are browser requirements, not a tested-device certification for Noema. Model loading also depends on available graphics features, memory and storage. Noema reports whether it can load the selected model on the device in use.
Browser support sources: WebKit’s Safari 26 release and Chrome’s Android WebGPU release.
Start small when a choice is offered. Close demanding tabs and keep the page in the foreground during loading and generation. Mobile operating systems can suspend a tab or reclaim its memory.
There is no promise that every model runs on every iOS or Android device. If the browser reports unsupported graphics or loading fails repeatedly, choose private cloud AI with ZKR instead. Noema Router Auto uses zero-retention routes; direct models show their own policy labels.
When an answer misses the task
Use a clear, short prompt and reduce unrelated conversation history. Small local models have limits. For a more complex task, use a larger supported model if the device can handle it, connect a local server or choose a cloud model.