AI Assistant
Statios ships a built-in AI assistant that runs analyses for you in plain English. There is no API key to bring and no model to choose — you sign in, and Statios routes each request to the right frontier model on its side. Everything presents simply as "Statios".
How it works
- The assistant is given automatic context about the active dataset — variable names, types, and missing-value counts — plus short digests of results you've already computed and a list of executable analysis commands.
- You ask a question — "test whether income differs by gender".
- The assistant picks an analysis, calls the Statios engine on your device with the right parameters, receives the result, and writes a plain-language interpretation.
- The numeric result also lands in the Output zone as a normal entry — it doesn't live only inside the chat thread.
The model itself never computes your statistics. It chooses which analysis to run; the validated engine produces every number, so figures come from auditable code rather than from a language model's guess.
What leaves your device
Requests go to the Statios relay, which calls the model provider (currently Anthropic Claude) on your behalf. The relay receives only:
- your prompt;
- schema and aggregate context (variable names, types, missing counts) and short result digests;
- never your raw dataset rows.
This privacy boundary is enforced in code: the relay rejects any payload that looks like raw tabular data. Your dataset stays in your browser. See the Privacy Policy for the full data-flow.
Accounts and usage
The assistant requires a free account (the statistics workbench does not). AI usage is metered in credits, with higher allowances on paid plans.
Prefer on-device statistics? Skip the assistant — the entire engine, 600 procedures and all, runs locally on your machine, so your data and analysis never touch the network. (Statios requires an account to open the app, but the statistics themselves run entirely on your device.)