Use chat and Ask AI

The assistant answers in plain language and can propose analyses, but you stay in charge: it does not change your data or submit an analysis without your click. The statistics behind its answers are computed by Celeus’s engine, not by the AI - still, check any number it quotes against the results page.

Chat with your data

Open a dataset and choose Chat with your data.

You can ask things like “which test should I use?” or “propose an analysis to run”. To answer, the assistant can look at the dataset’s de-identified profile, list and describe the tests Celeus runs, check how well a test’s assumptions hold on your data, and run an analysis. What comes back to it is the profile, test descriptions, assumption grades and aggregate results - never your data rows.

When the assistant has something concrete, it shows it as a card instead of acting:

• A proposed analysis with Run this analysis and Configure & adjust. If the assumption checks grade the method as not appropriate, or the statistical advisor would not have suggested it, the card says so. You can still run it; the disagreement is recorded in the reproducible package.
• A proposed population (a subgroup filter) that is applied when you run the proposed analysis.
• A proposed data cleaning step with Apply this cleaning, which writes a new dataset version and is recorded in the package.

Columns you have excluded from the AI are left out of the profile the assistant sees, and it cannot run or propose an analysis on them. At present, a new column you derive from an excluded one is not excluded automatically - exclude it too.

Ask about statistics without a dataset

With no dataset open, the assistant can still explain statistical methods, which tests Celeus runs, and how to use the app. It cannot see any of your data in this mode. Chat without a dataset needs a paid plan; chat on an open dataset is included on every plan.

Ask AI about a specific result

Panels on the run and suggestion pages - the result, the flags, a single flag, the suitability grades, the plain-language summary, the baseline table - have an Ask AI about this button. It opens chat with a question already written, carrying that panel’s numbers and labels. Nothing is sent until you choose Send, so you can edit the question first.

What it costs

Each chat turn uses AI tokens from your plan’s monthly allowance. See Plans and billing.

What the assistant sees

Everything the assistant receives is listed in What we send to the AI: the de-identified profile, the aggregate results of analyses, your research context, and the text you type into the chat. The assistant is advisory. It is not clinical decision support and does not make decisions about individual patients.