Jina Embeddings

Jina Embeddings is a family of models for developers building semantic search, retrieval-augmented generation (RAG) and AI agent applications. The models convert content into vectors that a search system can compare for relevance.

• Multilingual text retrieval: Text models support long inputs and task-specific adapters, with adjustable embedding dimensions to suit different storage and retrieval needs.
• Search across media: Omni models represent text, images, audio and video in a shared embedding space so a retrieval system can match content across media types.
• Compatible text indexes: Matching text and omni variants use identical text embeddings, allowing an existing text index to accept other media without reindexing its text.
• Retrieval integrations: The hosted Embedding API connects with tools such as Qdrant, Chroma, Milvus, LlamaIndex, Haystack and LangChain.

Open-weight models and GGUF and MLX variants provide options for running models locally. Jina also offers a hosted API, deployments through AWS SageMaker and Microsoft Azure, and custom Kubernetes deployments for private clouds or on-premises servers. The hosted API requires an API key and sends inputs to Jina's service; on-premises deployment runs on your own infrastructure.