querypad

Cursor for Data - a local-first AI CLI that understands your datasets, not just runs SQL on them.

QueryPad points an AI at a folder of CSV/Parquet/JSON files, profiles them, discovers how they connect, builds a semantic model, and answers business questions with DuckDB - locally, with no server-side data processing and no account.

The execution layer is solved (DuckDB does it well). The unsolved problem is that people don't understand their data: which tables exist, what each field means, how datasets connect, which join is correct. QueryPad answers those questions first, then generates and runs the SQL.

inspect: understand a folder

Scans the folder, profiles every file, and infers foreign-key relationships with confidence scores:

It writes machine-readable artifacts that an AI agent (Claude Code, Cursor, ...) can read to reason about the dataset instead of guessing at pandas:

inspect also rolls the relationships into a semantic model of named entities - with mechanically-derived dimensions, measures, and synonyms (deterministic, no AI) so the agent is grounded in what you group by and the metrics that exist:

External databases: no JDBC, no DBeaver

Point the same engine at a live database instead of a folder: