Create Custom Workflows

Design custom AI workflows with structured outputs, runtime inputs, and predictable viewer behavior.

Custom workflows let your team encode repeatable logic beyond default presets.

Design principles

  • Start from one clear outcome.
  • Keep prompts constrained and source-grounded.
  • Use structured output modes when UI controls require predictable keys.
  • Prefer small iterative versions over one large monolithic workflow.

Building blocks

Workflow definitions include:

  • metadata (title, description, category, tags)
  • optional runtime inputs
  • one or more execution steps
  • UI control mapping for viewer behavior

Common step types

  • LLM_STRUCTURED_OVERLAY for inline summaries and mapped options
  • TRANSLATE_NEW_VERSION for language variants with route-level behavior

Safe rollout process

  1. Build draft workflow.
  2. Test on internal files.
  3. Validate output quality and UI controls.
  4. Publish to team workflow hub.

For baseline examples, review workflow hub and presets.