Show HN: TamedTable, AI ETL in Natural Language

TamedTable, a source-available AI ETL tool, lets users transform tabular data using natural language instead of formulas or code, and its benchmark shows the default model matched hand-checked labels 96.7% of the time at $0.15 per 1,000 rows classified. The tool runs in the browser and CLI, supports CSV, JSONL, Parquet, and Arrow files, and allows users to replay recipes or export them as Python scripts.

Talk to your data. Load tabular data, say what you want, and watch it happen. AI ETL https://en.wikipedia.org/wiki/Extract, transform, load : no formulas, no code. No API key? Take a guided tour https://www.tamedtable.com/app/?tours=1 : it replays real sessions, free. | Name | Phone | Country | |---|---|---| | John D. Doe | 200 555-1234+12005551234 | USA | | Anna Müller | +49 555 6789+495556789 | Germany | | 张三 | +86 555 1234+865551234 | China | Every common data job, in natural language Transform with context The AI understands context, so it cleans, enriches, classifies, and validates — things no formula can. See everything See changes as they happen. Nothing is hidden: TamedTable is source-available /FAQ busl , and runs on your own API keys. Reuse and automate Every change saves as steps you replay on new data, or export as a Python script. Runs in browser and CLI. Read many formats Open CSV, JSONL, Parquet, and Arrow files — local, from a URL, or a bundled sample. How it works Open a file CSV, JSONL, Parquet, or Arrow. Local or from a URL. Say what you want Works in any language: English, Chinese, etc. Watch changes Dislike changes? Undo anything. Automate You made a recipe; replay it on new data. Lazy AI execution Preview for cents, run for real when you're sure. AI steps fill the page you are looking at; the rest waits until you ask. Clean up Fix messy fields the way a person would. The AI reads each row's context before deciding. Enrich & extract Add what isn't there yet. The model fills the gaps and pulls structure out of free text. Classify Sort rows by what they mean. The AI reads what each row is about. Validate Catch the wrong-looking rows before anyone else does. Process language Summarize, translate, and classify any language. Be exact The everyday spreadsheet jobs are still here; deterministic and repeatable. Load, save & reuse Your work is yours to keep. Load from anywhere, save the data or the recipe, undo anything. Trustworthy by design A repeatable recipe Each request becomes a small spec you can read. Deterministic steps run in a real SQL engine, and the same recipe replays on next month's file. Undo anything Every change lands in the history. Step back one change or all of them; the original file on your disk is never touched. Leave anytime Export any recipe as a plain Python script and run it without TamedTable. Your data stays in open formats; there is no lock-in to defend. Tested & benchmarked CI replays 606 test scenarios on every commit. In our benchmark https://github.com/ZSvedic/TamedTable/tree/main/benchmarks , the default model matched hand-checked labels 96.7% of the time, at $0.15 per 1,000 rows classified. Try it in two minutes Source-available /FAQ busl . Runs on your own API key /FAQ byok , or take the guided tour https://www.tamedtable.com/app/?tours=1 without one.