For researchers with data and a question
Statistical software you can talk to.
A statistician's judgment, at AI speed. Get an analysis you can defend, written up.
What you get
Whether you have never run a test, or run them every day, it works the same way.
It asks only what it needs, in words you can answer. It shows every step, so you can check it.
You don't need to know the name of the test.
Say what you are trying to find out. It proposes the test that fits your data, and says why that one.
You can talk it through.
Ask why, and the reason comes back. Push back, and it either argues its case or changes the plan.
It won't run what your data can't support.
It checks your data against what the test requires. When they fall short, it stops, names the check that failed, and waits for you.
The result comes written up, and you can check every number.
Plain language first, then the Methods and Results for your paper. Every number traces back to the rows it came from.
How it works
A conversation from your question to your result.
You ask, it proposes, and the two of you settle the plan. Then it checks, and stops if your data cannot answer.
You ask
You ask in your own words.
The question as you would put it to a colleague. No test names required.
It answers
It reads your data and proposes the analysis, in writing.
Column types, sample size, the design. Then a specific test, why it fits, and which columns it will use. Only what your data's structure allows.
Instead of the years it takes to learn which test.
You talk it through
Ask why. Change a column. Push back.
It asks when something is missing and challenges what does not line up with your study. You approve the plan before anything runs.
Instead of the hour with a statistician you may not have.
It checks
Before it runs: quality, assumptions, mapping.
- Quality. Missing values and outliers, flagged.
- Assumptions. Whether the data meet the conditions of the test.
- Mapping. Which columns answer the question.
Instead of a day of checks by hand.
The stop
When your data cannot honestly answer, it stops.
It says what needs a look, and why. The fix is yours to make, and the run waits for it.
Instead of a wrong result that looks exactly like a right one.
Your result
The result, written up, and checkable.
The numbers, what they mean in plain language, the paragraphs for your paper, and the trail of what ran on which rows.
All of that: minutes, and a fraction of the cost. No hoping you got it right.
A conventional analysis bought from a university consulting unit runs €1,200 to €3,000 at published EU academic rates and takes one to four weeks.
Why it holds
Judgment you can check.
Mechanism
The model proposes and explains. A check it cannot override decides. Code computes.
The language model discusses, challenges and flags gaps. It runs nothing. Every number comes from a statistical engine, tested against worked examples with known answers on every code change. Even the write-up is checked: a conclusion that points the wrong way, or a swapped group name, does not reach you.
Privacy
Your full dataset never leaves your machine.
The statistics run in your browser, and no computation on your data leaves it either. The model sees the structure of your table, column names and types, and the conversation. The rows stop at the wall.
Coverage today
Five families of analysis.
Group comparison
t-tests, ANOVA, ANCOVA, factorial and mixed models, and their non-parametric alternatives
Correlation and association
Pearson and Spearman, chi-square, ordinal regression
Diagnostic accuracy
Sensitivity and specificity, ROC and AUC, agreement, logistic models
Measurement models
SEM, CFA, IRT, latent class
Meta-analysis
Pooled effects across studies
Closed beta. Limited spots.
Join the waitlist.
Tell us what you are researching. We open the beta in small groups and write to you when your turn comes.