Field Station AI™ is an open-source, local-first AI workspace that runs in a web browser for health, behavioral, and digital-research workflows. It lets researchers use browser-based language and speech models without sending chats or study files to a cloud AI service by default.
From a single HTML page, users can chat with a local assistant, ask questions about attachments, transcribe audio, classify text, summarize documents, combine spreadsheets, and search an optional compendium (a bundle of knowledge from many sources, indexed for AI search). Models and compendiums download on first use and can run from the browser cache afterward. A modern browser with WebGPU support and a high-end GPU (video card) are recommended for faster performance and larger models; smaller models may work on less powerful hardware.
Field Station AI™ is designed for workflows that may involve sensitive data or protected health information (PHI), but institutional cybersecurity and IRB review may still be required before it is used with regulated data.
Want to try it first? Check out the live demo — it ships with the Depression Center Resource Library compendium bundled in (works best with Llama or larger models).
Field Station AI™ is a single, dependency-free HTML file — there is no build step, so setup is just a matter of getting hosting that file with a web server and opening it in your browser. Note: you can't just double-click index.html to open it — browsers block AI model downloads for pages opened directly from disk, so it needs to be served over HTTP (localhost is fine). Follow these simple steps to run it:
- Download or clone this repository.
- Serve the folder over HTTP (web server):
- To run locally, double-click the
run-*script for your operating system, or serve the folder in python withpython -m http.server 8010. The run scripts will automatically open your web browser, pointing it to: http://localhost:8010/ - To run on a web server, create a directory in your web root called FieldStationAI, and copy the
index.htmlfile there. Then visit your website and add /FieldStationAI/ at the end of the URL.
- Select a model from the dropdown and wait for it to download and compile. Once a model has been downloaded, you will not have to download it again, even if you close the application or refresh the page.
- Type a prompt, attach a file, or open Field Kit (in the top right) for task-specific tools.
- Click the compendium badge next to the model dropdown to choose whether answers use the bundled Depression Center Resource Library, your own compendium, or no compendium at all. See Data, Files, Attachments, and Compendiums for the details. To create your own compendium, see Extractium™.
- Complete documentation: See the
/docsfolder in this repository for the quick start, user guide, Field Kit guide, data and compendium reference, security and accessibility notes, and the developer and architecture guides. - Overview for researchers and developers: Visit the Health Research Resource Library for a high-level summary, key features, and important assumptions.
- For AI coding agents: Start with AGENTS.md and the skills listed in SKILLS.md. Always review and test generated code before opening a pull request.
- Mobile Technologies Core — the group that develops and maintains Field Station AI.
- EFDC Knowledge Base — documentation site referenced above, and one of the sources in the app's bundled compendium.
- Extractium™ — the tool that builds compendiums, including the one bundled with Field Station AI. Use it to create your own.
The Mobile Technologies Core provides investigators across the University of Michigan the support and guidance needed to utilize mobile technologies and digital mental health measures in their studies. Experienced faculty and staff offer hands-on consultative services to researchers throughout the University – regardless of specialty or research focus.
Learn more at: https://depressioncenter.org/mobiletech.
To get in touch, contact the individual developers in the check-in history.
If you need assistance identifying a contact person, email the EFDC's Mobile Technologies Core at: efdc-mobiletech@umich.edu.
Used by Field Station AI™:
- Transformers.js - Runs Hugging Face transformer models (chat, vision, classification, and embedding) directly in the browser, entirely client-side.
- Pyodide - A Python distribution compiled to WebAssembly, used to run pandas-based data-cleaning code locally in the browser sandbox.
- PapaParse - In-browser CSV/delimited-text parsing used when ingesting research data files.
- SheetJS (xlsx) - Reads and writes Excel spreadsheet files entirely client-side.
- PDF.js - Renders and extracts text from PDF documents in the browser.
- Ollama - Optional, locally-run backend the app auto-detects to offer larger language models beyond what runs directly in-browser.
- Extractium™ - Builds the compendium file (
efdc-compendium.json.gzorefdc-compendium-full.json.gz) that the app searches, and defines the container format the app reads. - ZippyServe - A zero-dependency local web server. It lets you test single-page apps quickly. It serves directories, zips, HTML, and Markdown. It provides the run-* scripts to allow starting Field Station AI locally without installing a full web server. DOI: 10.5281/zenodo.21613944.
Copyright © 2026 The Regents of the University of Michigan
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/gpl-3.0-standalone.html.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no Front-Cover Texts, and no Back-Cover Texts. You should have received a copy of the license included in the section entitled "GNU Free Documentation License". If not, see https://www.gnu.org/licenses/fdl-1.3-standalone.html
If you find this repository, code or paper useful for your research, please cite it.
Mongefranco, Gabriel (2026). Field Station AI™. University of Michigan. Software. https://github.com/DepressionCenter/FieldStationAI
DOI: 10.5281/zenodo.21617547
Copyright © 2026 The Regents of the University of Michigan