Databricks makes three Document Intelligence REST APIs generally available

Databricks made three synchronous Document Intelligence REST APIs generally available on October 14. The three endpoints handle parsing, extraction and classification, allowing applications, services, automation workflows and agents to process documents as they arrive instead of sending the work through a batch pipeline.

Databricks identifies ai_parse_document, ai_extract and ai_classify as generally available. ai_parse_document turns unstructured files into structured content. It accepts PDF, Word, PowerPoint, JPEG, PNG and TIFF inputs of up to 100 pages and 100 MB per document.

The extraction endpoint takes plain text or parsed-document output and returns fields defined by a caller-provided schema, with optional citations and per-field confidence scores. The classification endpoint assigns caller-defined labels to plain text or parsed documents and can include rationales and confidence scores.

Databricks documents three ways to use the same AI Functions: an Agent Bricks interface for visually testing and refining document processing, SQL functions for at-scale Lakehouse pipelines, and REST calls from applications, services and automation workflows. Its documentation says parsed output can flow directly into extraction and classification without another transformation step.

The separate Databricks ai_decide function remains in Beta. Its REST endpoint turns text or structured application state into a selected choice, a probability estimate or a score. Those results can support document routing or flag items for human review. ai_decide was not part of the three APIs that reached general availability.