Vector search
Search your tableby meaning.
Tables with a DynamoDB vector index get a fourth query flavor, next to Filters, PartiQL, and Schema. Paste a vector, pick a row, or type a sentence. The nearest items come back ranked, with the line between the strong matches and the tail drawn in.
Three ways to ask. Pick the index, then supply the query vector however you have it. Paste an array of numbers, pick a row already loaded in the grid, or type plain text and let the index recipe embed it with the same model that filled the index. Switching between the three never blanks what another one held. An index that partitions on an attribute asks for the value to search within, and equality filters on its inline attributes go on as pills.
Ranked, with the cliff marked. Every hit carries its raw score to four decimals beside a relevance meter, and the Score header says which direction is closer for the distance function your index uses. Above the grid, one bar per hit draws the score distribution, and a divider marks where the strong matches end. The gap between a real answer and the tail is on screen, not behind a threshold you cannot see.
Find similar, from any row. Right-click a row that carries a vector and pick Find similar. Dynomatic runs the search against the tab's own table with the row's stored vector, switches you to Vector search with the bar already filled in, and pins the source row above its hits. Your Filters results stay as they were, one click away. A row whose vector was left out of the grid still works. The vector is read back by key at call time. See the visual query builder →
From index to answer
Tables with a vector index get Vector search in the Query rail. Availability follows what DescribeTable reports, so it works against AWS, Dynoxide, and any other endpoint that serves vector indexes.
An index that partitions on an attribute requires the value to search within. It scopes the search instead of throwing results away afterward.
Top K from 1 to 100. The vector attribute is left out of results by default, and a keys-only index can load the full items after the search.
A saved vector search remembers the array, the item key, or the sentence you typed, and reopens as that same kind without embedding again.
Every run lands on the History page under a Vector Search pill, with the request size in bytes in place of read units.
Table settings holds the recipe that fills the vector attribute, a bounded dry run, a checkpointed backfill you can pause and resume, and a verification pass that reports what is missing or the wrong dimension.