The idea
Field Station AI is an open-source AI workspace for health, behavioral, and digital research. It runs in a web browser on the researcher's own computer. Researchers can use language and speech models in the browser, and by default their chats and study files are not sent to a cloud AI service.
What it does
From one HTML page, you can chat with a local assistant, ask questions about attached files, transcribe audio, classify text, summarize documents, combine spreadsheets, and search a compendium. A compendium is an optional bundle of knowledge from many sources, indexed for AI search. Models and compendiums download the first time you use them, and after that they can run from the browser cache. The whole application is one HTML file with no dependencies, served over HTTP.
The workspace. You pick a model, attach files, and ask a question, and everything runs in the browser.
Context
Field Station AI is designed for work that may involve sensitive data or protected health information. Even so, an institution may still require a cybersecurity review and an institutional review board (IRB) review before anyone uses it with regulated data. The compendium it ships with is built by Extractium™.
My contribution
I wrote Field Station AI as part of my work at the Depression Center's Mobile Technologies Core, which develops and maintains it. The project is university property and open source, with a citable record at DOI 10.5281/zenodo.21617547. You can try the live demo, which ships with the Depression Center Resource Library compendium.