Vector search allows you to search documents, images and other data by converting items into a collection of vectors. This collection of vectors summarises the data in semantic form and allows us not only to match documents against queries through analysis of the semantic content, but also to understand where and how the document matched the query. With Marqo, inference to create the vectors is included.

The number of instances you will need depends on a number of factors. The number of documents, the size of the documents and the type of data (image vs text). When dealing with low search volumes that primarily involve text or when low latency is not crucial, using CPU inference nodes can be a cost-effective solution. On the other hand, GPU inference nodes provide a significant performance boost when indexing and searching with images and are recommended for indexing large datasets and processing high volume, low latency searches. For multimodal models marqo.CPU.large is recommended as a minimum.

The estimates for storage capacity provided in our calculator assume your are using a model that produces 768 dim. vectors.

The only changes you need to make are to update your URL and API key when accessing Marqo.

Cloud pricing billed at the end of the month for total inference and shard hours used. Usage is rounded up to 15-minute increments. Enterprise pricing also includes a per session component for model training.