Free · no signup · browser-local

Embedding Cosine Similarity, Dot Product and Distance Calculator

Compare existing embedding vectors using cosine similarity, dot product and Euclidean distance. Inspect dimensions, zero-vector diagnostics and candidate rankings locally.

Browse all tools →

Browser-local calculation · no model calls · no saved input

Input and settings

Optional coordinates compare A/B only, ranked by absolute raw dot contribution. The displayed subset is not the full dot sum. L1 and angular distance use the original vectors; zero vectors have no defined angle.

Object mode JSON: {"a":[3,4],"b":[-3,-4],"candidates":[{"id":"c","vector":[1,2]}]}. Candidates are compared with a. No normalization is applied to dot product or distance.

Same dimensions, 1–4,096 each; a+b+up to 98 candidates; |component| ≤ 1e150. Zero-vector cosine is undefined and sorted last. Ties keep input order.

Your output

ai-embedding-similarity-report.json

Complete the input to preview and export your result.

These metrics compare supplied numbers. Meaningful semantic comparisons require compatible embeddings produced by the same method; no embeddings are generated here.

Getting started

Paste JSON with a/b arrays and optional candidates, or use two vector inputs. Choose a ranking metric and 0–100 coordinate contributions, then calculate locally.

Before you use the result

No embeddings are generated and no semantic quality is inferred. Vectors must share dimensions; zero-vector cosine is undefined. Floating-point rounding and tiny dot-product underflow can occur.

Try an example

a=[3,4], b=[-3,-4] gives cosine −1, dot −25, distance 10 and lengths 5.

How it works

Scaled L2 norms, compensated dot sums and normalized cosine. 500 KB UTF-8; 100 vectors total, 4,096 dimensions each, |component| ≤ 1e150.

Frequently asked questions

Compare existing vectors with cosine, angle, dot product, L1/L2 distance, coordinate contributions and candidate rankings.

How do I use AI Embedding Similarity Calculator?

Paste JSON with a/b arrays and optional candidates, or use two vector inputs. Choose a ranking metric and 0–100 coordinate contributions, then calculate locally.

Does this tool upload my data?

Inputs and files are processed in your browser. Page-view analytics do not contain your tool input.

What should I check before using the result?

No embeddings are generated and no semantic quality is inferred. Vectors must share dimensions; zero-vector cosine is undefined. Floating-point rounding and tiny dot-product underflow can occur.

Related tools

RAG Retrieval EvaluatorEvaluate existing ranked document IDs against graded judgments with Precision, Recall, F1, Hit, RR, AP and nDCG at K.LLM Sampling PlaygroundExplore temperature, min-p, top-k, top-p and greedy selection on your own logits with probability bars and repeatable seeded sampling.RAG Text Splitter with Source OffsetsPrepare traceable text chunks with paragraph, line or fixed boundaries, a hard size cap and adjustable overlap.