Privacy and device use

Updated 23 October 2026

NeuroAtlas is an educational anatomy project. Chat answers and MRI masks can be wrong and are not medical advice or a diagnosis. Avoid entering personal or patient-identifying information.

Questions and selected files

The chatbot runs on your device. Questions, conversation context, and optional semantic-search vectors are held in browser memory; the application does not save a conversation to an account or send it to a question-processing backend. New conversation clears the displayed conversation and its context. Reloading the page starts a new session.

Selected MRI images, matching ONNX model files, and prediction results are processed in browser memory. This application has no image-upload or remote-inference endpoint. A selected file stays on your device unless you choose to share it separately. Download binary mask PNG saves a result through your browser.

Page and model downloads

Your browser requests the website, its anatomy data, and saved illustrations from the hosting provider. AI libraries, compiled runtimes, and model files load from their public providers only after you explicitly enable the relevant model. These downloads may contact jsDelivr, Hugging Face, and GitHub. Hosting and download providers receive ordinary request metadata, such as your IP address, requested URL, and browser information. Their handling of those requests is governed by their own policies.

Following a source or license link visits an external website, which has its own privacy practices. NeuroAtlas does not add analytics, advertising, account login, or application cookies.

Your device, bandwidth, and browser cache

Inference uses your CPU or GPU, memory, and battery. The chatbot model downloads about 290 MB plus runtime files and requires compatible WebGPU support. Optional CPU semantic search downloads about 23 MB of weights, 0.7 MB for its tokenizer, plus runtime files. MRI prediction needs a matching model file of about 88–295 MB and its separate browser runtime. MRI weights are not distributed by this website.

Your browser may retain downloaded model and runtime files in its cache or site storage for later use. New conversation does not remove those caches. To remove retained downloads, use your browser's site-data and cache controls; removing site data may require models to download again. Caching and available storage depend on your browser.

Available without a model

The brain atlas, word-based source search, thesis plots, and saved MRI comparisons work without loading an AI model. Live MRI prediction additionally requires a matching verified ONNX file supplied by you, or the separately enabled private model route when running the project locally.