Data work

The agent works on files in your project folder, with your permissions, in the project’s own Python environment (Installation).

How the agent works with you

The agent leads the way a colleague who knows analysis would, and leaves the decisions that are yours to you:

It looks first. Say hello in a folder with data and it takes a quick look at what is there, then suggests a few questions those files could answer.
It makes vague questions concrete. “How are sales doing?” becomes two or three specific questions, each with the measure it would use. It starts with one, and you can pick another.
It takes the obvious next check itself when it is quick and part of your question, such as a breakdown or a sanity check, then suggests at most three next steps and starts the bigger ones only when you ask.
It publishes results worth keeping to your project in AIUS Light as a report, by itself, and tells you with the link. It asks before anything goes elsewhere; see Publishing.
It proposes connecting the services you use when your question needs their data, and walks you through it; see Connections.
It suggests a dashboard for numbers you check again, and builds it when you want it.
It never invents data, results or links. When it has not looked, it says so.

Tool	Use
Data profile	Rows, columns, types, missing values, duplicates, distributions and quality alerts for CSV, TSV, Parquet, JSON, JSONL and Excel files
Data query	SQL over local CSV, Parquet, Excel and DuckDB files with DuckDB, returning bounded results
Data check	Explicit quality rules: allowed values, ranges, uniqueness, missingness
Compare	Differences between two datasets or two versions of one
Split audit	Leakage between training and test data
Drift	Changes in a variable’s distribution between periods
Python	Scripts under `output/`, run with the project interpreter
Notebooks	Create, edit, run and read Jupyter notebooks, including outputs and figures
Training and validation	Fit models with cross-validation and record data hashes, package versions, fold membership, metrics and model hashes

For files too large to read whole, the agent works by schema, samples and aggregate queries instead of loading everything.

Numbers come from code. Every figure in a report is produced by a script saved under output/, so it can be rerun.
Leakage is checked. The agent looks for fields that are only known after the outcome, and for overlap between training and test data.
Baselines come first. Models are compared with a simple baseline and validated in a way that suits the data, for example forward in time when the rows are ordered in time.
Uncertainty and limits are stated. A good split-audit result or cross-validation score is evidence, not proof that a model will work in production or that an effect is causal.

Worked example shows all of this on a real dataset.