Quacksmith Quacksmith Quacksmith Quacksmith

Your colleague can check your pipeline without reading SQL.

Describe it in words or draw it. Then read it back, step by step, with real data.

“transform this incoming file and show me only unique records” — you

the readback
An example flow, read back customers.csv gives 5 rows; select keeps 4 columns and 5 rows; dedup by customer_id drops 2 and returns 3 rows. source customers.csv 5 rows select 4 columns 5 rows dedup by customer_id 3 rows 5 −2 here An example flow, read back customers.csv gives 5 rows; select keeps 4 columns and 5 rows; dedup by customer_id drops 2 and returns 3 rows. source customers.csv 5 rows select 4 columns 5 rows dedup by customer_id 3 rows 5 −2 here

Every node’s real rows are one click away, and every wire carries its count. When rows disappear, the node that dropped them says so — in numbers, in place.

Legible by construction
A per-step split with real data at each step. No black box between source and output — and the canvas compiles to SQL you can read and edit.
AI you can check
Chat authors; the canvas is the readback; an approval gate stands between a draft and production. “AI authors, you approve” is a gate here, not a slogan.
Grow, don’t rewrite
Start with a CSV. When a table matters enough to need identity, history and provenance, promote it and inherit the machinery instead of hand-rolling SCD2 again.

COMING 2026

Nothing is shipped yet. This page describes the thesis, not a released product.

Tell me when it ships

One mail from a person when there is something to see — no list, no tracking. Or just say hello: hello@quacksmith.io