Flow editor

A flow is an end-to-end document processing pipeline: a sequence of steps that digitize documents, extract and validate data, run custom logic, and connect to other systems. Flows are published as advanced apps, which you can run on-demand or deploy for automated processing.

The flow editor

The flow editor is one of two ways to build document understanding apps in AI Hub. The app editor offers a guided, no-code/low-code experience for document classification and extraction. The flow editor exposes the full pipeline, giving you step-level control and the ability to extend behavior with Python.

To open the flow editor, click the initials icon in the header and select Switch to advanced view.

What you can do with a flow

• Chain classification, extraction, validation, refinement, and document generation steps into a single pipeline, including LLM-based Agent classifier and Agent extract steps.

• Configure individual steps and override default behavior.

• Use custom functions to add Python logic and integrations at multiple steps. On single-tenant deployments, custom functions can use an LLM client for AI-powered processing.

• Connect to upstream and downstream systems.

When a run finishes or needs attention, reviewers can inspect results and complete human review.

Production and integrations

When a flow is ready for production, publish it as an advanced app to schedule or trigger runs, configure exports, and connect to upstream and downstream systems. You can embed upstream and downstream logic directly in the flow using pre-flow and post-flow events, or attach integrations through an app deployment after publishing.