Databricks launches real-time Document Intelligence APIs

Databricks has officially launched its Document Intelligence real-time REST APIs, replacing traditional asynchronous batch pipelines with synchronous document processing. Enterprise organizations that handle millions of documents every single day can now parse, classify, and extract unstructured text immediately upon upload, allowing human-in-the-loop interactive applications and AI agents to make immediate operational decisions without waiting for batch runs.

The release shifts three core endpoints into general availability: ai_parse_document, ai_extract, and ai_classify, while introducing a fourth tool, ai_decide, in beta. The ai_parse_document endpoint converts embedded tables and figures into structured text while calculating citation bounding boxes for visual verification. Meanwhile, ai_extract pulls requested fields into structured outputs accompanied by optional confidence scores, and ai_classify tags documents according to an enterprise business taxonomy. The beta tool, ai_decide, reads the contents of incoming files to direct workflow routing or flag complex documents for administrative review.

For software engineers and data practitioners, this synchronous setup streamlines integration into custom frontends built with Databricks Apps or autonomous workflows using Agent Bricks. For instance, in a legal contract or invoice processing pipeline, an application can parse an uploaded file, identify whether it is an NDA or statement of work, prefill required data fields, and write the verified record directly to Delta. Developers can experiment with these capabilities through the Agent Bricks UI on the workspace Agents page or start generating code immediately by prompting Genie Code.