Search foundation
Embeddings
The first stage of search. An embedding model turns documents and queries into vectors, so nearest-neighbour search finds what a query means — not only the words it happens to use.
In development
What are embeddings?
An embedding is a list of numbers that represents a piece of text. Texts with similar meaning get vectors that sit close together, so a question and the passage that answers it can be matched even when they share few words.
Documents are embedded once and stored in a vector index. At query time only the query is embedded, and nearest-neighbour search returns the closest documents — the candidate set a reranker then refines.
How it works
Embeddings in 3 steps.
-
01
Index
Embed your documents once and store the vectors in a vector index.
-
02
Query
Embed each incoming query with the same model.
-
03
Retrieve
Nearest-neighbour search returns the documents closest in meaning, ready for reranking.
Models
Embeddings models
No Embeddings model is released yet. Models appear here as soon as they are published on Hugging Face.
Compare
Embeddings and BM25
| Embeddings | BM25 | |
|---|---|---|
| Matches on | Meaning | Shared terms |
| Paraphrases and synonyms | Found | Missed unless the query is expanded |
| Exact names, codes and numbers | Can blur them | Precise |
| Index | Vector index | Inverted index |
FAQ
Embeddings, explained.
Do I still need a reranker?
Should I combine embeddings with BM25?
How do I get access?
Start a conversation
Put Embeddings to work
on your search.
Tell us what you retrieve and how you measure quality.
Get in touch →