embedMany and embed functions are part of the AI SDK
The embedMany and embed functions are imported from the 'ai' package, which is the Vercel AI SDK. These functions provide a unified interface for creating embeddings across different providers.
AI SDK · Providers · all subjects
ai-sdk/embeddings
6 notes, read out of this brain and free to use. Each one was extracted from a source and is re-checked against its exam.
embedMany function creates embeddings for multiple text chunks
The embedMany function takes a model identifier and an array of text values, and returns an array of embedding vectors. It is called with model 'openai/text-embedding-3-small' and an array of text chunks, returning { embeddings } where embeddings is an array of numeric vectors corresponding to each input value in order.
embed function creates an embedding for a single text value
The embed function takes a model identifier and a single text value, and returns an embedding vector. It is called with model 'openai/text-embedding-3-small' and returns { embedding } where embedding is a numeric vector.
cosineSimilarity function compares two embedding vectors
The cosineSimilarity function takes two embedding vectors and returns a similarity score. It is used to measure how similar one embedding is to another, with higher values indicating greater similarity.
RAG retrieval workflow example
This example demonstrates a complete RAG workflow: text is split into chunks, each chunk is embedded using an embedding model, a user query is embedded with the same model, chunk embeddings are compared to query embedding using cosine similarity, the top 3 most similar chunks are retrieved and joined as context, and the context is passed to a language model along with the query to generate an answer. The example uses OpenAI models openai/text-embedding-3-small for embeddings and openai/gpt-4o for text generation.
embed with provider registry embedding model
Use embed with model: registry.embeddingModel('openai:text-embedding-3-small') to generate embeddings using an embedding model accessed from a provider registry.