Marqo

The API covers index management, document indexing, and search. Documents are JSON objects with text and image fields — Marqo automatically generates embeddings for indexed fields using the configured model. Search queries accept natural language text or image URLs and return ranked documents across the embedding space. The document-oriented model makes Marqo feel familiar to teams coming from document databases rather than raw vector stores.

Reliability for self-hosted Marqo depends on the deployment configuration. Marqo Cloud provides managed availability. The embedding computation requirement means indexing operations are more resource-intensive than raw vector databases — teams should size their Marqo infrastructure appropriately for their indexing throughput requirements.

Marqo is an open-source tensor search engine and vector database with built-in embedding generation for both text and image content. Its document-oriented API — add documents, search documents — abstracts the embedding model selection and vector indexing complexity that other vector databases expose. For agents building multimodal search features that span text and images, Marqo handles the cross-modal embedding space without requiring teams to manage separate image and text embedding pipelines.

Authentication uses API keys for Marqo Cloud. Self-hosted Marqo can be configured with or without authentication. Teams running self-hosted Marqo for production search should implement network-level access controls appropriate for a search index that may contain sensitive document content.

Documentation covers the index configuration, document indexing, and search API clearly. The multimodal indexing documentation explains how to configure indexes for combined text and image search. Teams evaluating Marqo versus Weaviate or Qdrant for multimodal search should compare the built-in embedding model options and the document-oriented API design philosophy.