Voyage AI makes embedding models and rerankers for search and retrieval, and it is part of MongoDB ("Voyage AI by MongoDB"). An embedding model turns a piece of text into a vector, a list of numbers, so that texts with similar meaning get nearby vectors; a program then finds related text by comparing vectors. A reranker takes a query and a list of candidate documents and orders them by relevance.
A program sends text to the Voyage API and gets back vectors. The general text models voyage-4-large, voyage-4 and voyage-4-lite accept up to 32,000 tokens per input and return vectors of 256 to 2,048 numbers. Other models are tuned for code, finance and law; multimodal models embed images and video; one model is open-weight. Together these form RAG pipelines (retrieval-augmented generation: relevant documents are found and given to a language model before it answers), and the vectors work with any vector database and LLM.
Voyage AI is a hosted API, also offered through cloud marketplaces and inside a customer's own VPC. Pricing is pay per use, per token, after a free allowance for each account.