Fundamentals: Overview
SamtSQL acts as an AI analytics layer on top of an existing PostgreSQL database instance. This database instance must be accessible from the public Internet. We recommend PostgreSQL 18 or later versions for use with samtSQL.
SamtSQL enables users to write SQL queries with AI operators. Such AI operators enable deep semantic data analysis, e.g., calculating a breakdown of customer complaints based on categories, derived automatically from the complaint text. SamtSQL automatically evaluates AI operators using large language models (LLMs), applying sophisticated query evaluation techniques to minimize query evaluation time and costs. Your monthly subscription covers a certain amount of AI processing (AI Credits) per month, as well as a certain amount of computation overheads.
SamtSQL extends the relational data model by allowing tables that contain images or audio files in table cells. This does not require any changes to the underlying PostgreSQL database system, as samtSQL encodes all audio and image data as text in PostgreSQL. SamtSQL enables users to upload archives containing multimodal files, and transforms them into SQL tables. This enables users to write SQL queries for image and audio analysis, e.g., counting the number of images in the database showing certain objects.
The following pages provide more details on different aspects of samtSQL:
- Data: introduces samtSQL's multimodal data model.
- Uploads: describes how to upload multimodal data.
- Analytics: provides an overview of SQL with AI operators, as supported by samtSQL.
- Bounds: discusses processing cost metrics on which users can set thresholds.
- Partial Results: describes how samtSQL generates partial results once query evaluation exceeds cost bounds.
- Warnings: introduces warnings generated by samtSQL for anti-patterns detected in AI queries.
- Interfaces: introduces the user interfaces offered by samtSQL.