Profile a messy CSV / Excel / Parquet / JSON file, review a cleaning pipeline as plain YAML, and export it as standalone pandas, Polars, or SQL code you own.
Chat tools clean a file once. A pipeline cleans every month's file.
Everything serializes, diffs, re-runs, and exports. Nothing is fixed silently.
Every step is declarative YAML — git-diffable, re-appliable to next month's file, and validated against the schema before it runs.
Export the whole pipeline to standalone pandas, Polars, or DuckDB SQL — verified cell-for-cell against the engine. Keep working code even if you drop the tool.
Already fixed a file in Excel? Hand back both versions and tidysheet reverse-engineers the steps that reproduce your edits — and verifies them.
Re-running snapshots what the data looked like and reports value drift loudly — a new category, a value that silently coerces to null — instead of handing back quietly-wrong data.
Types, missing values, distincts, duplicates, IQR outliers, and semantic types (emails, currencies, dates-as-text, countries) — no AI, fully explainable.
The core runs offline with no model. An LLM may only propose steps for what deterministic rules can't explain — and its proposals pass the same verify gate. Bring your own model.
Four verbs. Learn it in a minute.
tidysheet profile data.csv — a self-contained report of what's in here and what's wrong.
suggest proposes steps, or infer original edited reverse-engineers them from your hand-edit.
run pipeline.yaml next.csv --strict — replay forever; fail loudly on drift.
export --to pandas — standalone code you own, no tidysheet dependency.
# turn a hand-cleaned Excel file into a reusable, verified pipeline tidysheet infer messy.xlsx messy_cleaned.xlsx -o pipeline.yaml # ✓ customer: stripped surrounding whitespace (12 cell(s)) # ✓ country: 8 value replacement(s) (usa→United States, uk→United Kingdom, …) # Reproduced 214/214 edited cell(s) (100%). # re-run on next month's file — drift never passes silently tidysheet run pipeline.yaml next_month.xlsx -o clean.csv --strict
A real, self-contained report generated by tidysheet profile on the demo dataset — works entirely offline, light & dark.