Introducing the Embedding Function: How Milvus 2.6 Streamlines Vectorization and Semantic Search

This post details the introduction of the Data-in, Data-out feature (Embedding Function) in Milvus 2.6. It describes the underlying 'Function' module that enables Milvus to invoke external embedding providers (OpenAI, AWS Bedrock, Google Vertex AI, Cohere, Hugging Face) directly during data ingestion and query time. The post outlines the six-step workflow: input raw data, generate embeddings via the configured provider, store embeddings, submit raw queries, perform semantic search using the same model, and return results. It also provides configuration details for `milvus.yaml`, schema definition with `TEXTEMBEDDING` function type, and usage examples for inserting raw data and performing text-based semantic searches.