Compare Embedding Pairs

@bleenco1 VersionCalculate

Computes the cosine similarity between two Embedding vectors and emits either the raw Double score or a Bool match decision once the score clears the configurable threshold.

How it fits

Typical backends

Compare two embedding vectors and route on the match decision through a conditional branch.

Compare two embeddings and publish the raw similarity score as JSON for logging or thresholding downstream.

Caveats

I/O contractWhich output arm is produced is fixed by the downstream connection: a Double arm emits the cosine score, a Bool arm emits the threshold decision.
Hard constraintThe two Embedding inputs must share the same dimensionality; mismatched vector shapes cannot be compared.
AccuracyTreating the inputs as already L2-normalized skips the norm computation and compares the dot product directly, which is faster but yields a wrong score when the vectors are not actually unit length.
LimitationOnly cosine similarity is computed; it does not perform a nearest-neighbor search against an index, just a pairwise comparison of the two inputs.

Versions

3c2459d4latestdefault
Automated release
6/29/2026