Embedding
An embedding is a vector of numbers that represents the meaning of a piece of text, image, or other data — close vectors mean similar meanings.
Embeddings are how AI systems compare meaning. The phrase 'cancel my subscription' and the phrase 'I want to unsubscribe' get embeddings that are mathematically close, even though they share few words. This is what makes semantic search and RAG work. Different embedding models give different vectors. Pick one and stick with it across your index — embeddings from different models aren't comparable.
Example
Acme embeds 'refund policy' and 'when do you give money back' using OpenAI's text-embedding-3-large. The cosine similarity is 0.81 — high enough that the second query retrieves the first as a match.
How OpenLabor uses it
OpenLabor handles embedding model selection and re-embedding when you upgrade — you never see the vectors directly.
Can I switch embedding models later?
Yes, but you have to re-embed your whole corpus. Mixing embeddings from different models is meaningless.
How big is one embedding?
Usually 768 to 4096 floating-point numbers. At 4 bytes each, a 1536-dim embedding is ~6 KB.
Related: vector-store, rag.
AI Labor Glossary