Data model
Point it at a tableyou didn't design.
Dynomatic samples your items and works out the entities, collections, access patterns, and relationships. Evidence attached, corrections yours, AI optional.
Bring your own single table design. Inference reads the keys and attributes already in your table and recovers the entities, the item collections they form, and the access patterns those collections support. Hand-rolled in Python, Go, Java, or anything else. No schema file, no library, no annotations.
Evidence, not magic. Your table is ranked against 13 known single-table design conventions, from The DynamoDB Book to PynamoDB and modyne. You see what matched, what is missing, and where the table conflicts with itself. Inference is deterministic and shows its work.
Every pattern is runnable. Access patterns come named and explained in plain English. Pick one, fill in the parameter, and the query builder is hydrated with the real key condition. Have an ElectroDB or DynamoDB Toolbox schema instead? Import it and skip the inference step. Either way you land in the same query surface. See schema import →
Under the hood, up close
Dynomatic samples your table and infers the model from thousands of observations. Deterministic, re-runnable, versioned.
Every entity with its key template, item count, and a confidence label that means something. Bookkeeping rows are set aside, and items that match nothing are counted, not hidden.
Matched, missing, and conflicting conventions per flavor, with the observations behind each call.
Who lives together under a partition: self item, children, and the sort order actually observed in your data.
One click carries an inferred pattern into the query builder with the real key condition underneath.
Parent, child, and reference edges shown as a matrix or graph, each with its evidence. Your corrections persist as overlays; re-inference never deletes them.