Fundamentals: Data

SamtSQL works as a layer on top of an existing PostgreSQL database. It extends the standard PostgreSQL database model by allowing tables whose cells contain images or audio files.

Multimodal data

SamtSQL extends the traditional PostgreSQL data model by enabling users to store multimodal data, namely images and sound files, in table cells. SamtSQL stores multimodal data in table columns of SQL type text. It is not necessary to declare such columns explicitly as multimodal data columns. Instead, samtSQL automatically checks the content of text columns during query processing to decide how to handle them. Different rows of the same text column can have content of different types. For instance, a single column may contain images in some rows and sound files in others. SamtSQL automatically selects a type-appropriate AI model for processing different parts of the same column. Each table cell may only contain one single element (either text, an image, or a sound file).

Handling multimodal data

Users can easily create tables with multimodal data by uploading ZIP files with multimodal content. See here for details on how to upload multimodal data. Alternatively, users can copy existing, multimodal content. All operations that can be used to copy text columns are available for copying text columns containing multimodal data.

SamtSQL supports AI operators within SQL queries. Such AI operators are evaluated by invoking AI models in the background. Those AI operators can only be applied to text columns (containing pure text or, additionally, multimodal data).