A research environment for clinicians
Nightfloat is a desktop application. You define a cohort, run the analysis, and keep a record of how you got there. It all happens on your own computer, and your patient data stays where it is.
Four things, done well
Nightfloat is not a general-purpose programming environment with a medical theme. It is built around the tasks that fill a clinical research project.
Build the cohort
Set the inclusion and exclusion criteria, define exposures and outcomes, and compare groups side by side. The cohort definition stays attached to the analysis, so a table 1 three months from now still matches the one you wrote.
Run the analysis
Write R and Python with the packages you already use. Results come back as tables and figures you can export, and the code that produced them is kept with them.
Ask for help
Describe the question in plain language and get working code with an explanation of what it does. Read it, change it, run it, or throw it away. Nothing happens that you did not ask for.
Keep the record
Every run records the data, the code, and the time. Export the whole record with your project so the analysis can be rerun later, by you or by a reviewer.
A desktop application, not a service
Nightfloat installs like any other desktop program. Your projects are folders on your own disk: data, code, and results together, where your usual backup can reach them. There is no workspace to log into and no copy of your work held anywhere else.
The assistant that helps you write and explain code runs on the same computer, using the machine's own resources. It is the one download Nightfloat needs, and it is downloaded once, when you ask for it. After that the application works offline.
- No account and no sign-in.
- The application does no usage tracking, no analytics, and no crash reporting.
- Works offline once it is installed.
- Once it is installed, the only thing Nightfloat downloads is the assistant, and only when you ask.
The custodians say the same thing. PhysioNet, which distributes MIMIC, strongly recommends locally deployed language models, and its data use agreement prohibits sending credentialed data through APIs or online platforms. The SEER research data use agreement of November 2025 forbids public cloud AI services and browser-based assistants, while still permitting model work that is described in an approved request and run on approved infrastructure.
Nightfloat makes the configuration those agreements ask for the default, and the one they prohibit impossible: there is no upload path to disable.
Sources:PhysioNet on MIMIC data and language models·SEER Research Data Use Agreement, November 2025
Every line of code is visible
You should not have to take a privacy claim on faith. Nightfloat is built to be inspected: you, and the people who approve software at your institution, can read the code that handles your data instead of trusting a paragraph on a website.
- The application, the notebooks, and the parts that touch your files can all be read.
- There is no backend that could quietly change what the software does.
- Your analyses export as code you can rerun outside Nightfloat.
What Nightfloat is not
- Not a medical device. Nightfloat is a research tool. It does not diagnose, treat, or recommend anything, and it is not for making decisions about the care of an individual patient.
- Not a substitute for statistical judgement. It runs what you tell it to run. Choosing the right test, and defending it to a reviewer, is still your work.
- Not a place your data is stored. SynopsisMD keeps no copy of your projects and has no way to read them.
In development
This page describes what we are building. Nightfloat has not shipped yet. The first release is aimed at residents and clinicians doing research on their own laptops, and details will change as it gets closer.
Free for individuals
Nightfloat costs nothing for individual clinicians. Institutions pay for running it across a team.