Skip to main content
Start with a simple OCR pipeline before adding complex branches and tools. First configure a document type, a suitable model and OCR provider, and an active license.
  1. In the workflow builder, create a workflow and select OCR Pipeline.
  2. Configure the start/ingestion stage. Use manual upload initially; optional folder import can be enabled for automated ingestion.
  3. Add an OCR / Extraction stage and configure the intended document type and extraction settings.
  4. Add Human Verification so an operator can check extracted fields against the original document.
  5. Add an export stage if downstream delivery is required. Review the destination and credentials before enabling it.
  6. Save the workflow and grant access to the appropriate users in User Config.
  7. Upload a non-sensitive representative document, follow its processing status, review the extracted values, correct errors, and approve it. Verify the exported result if export is configured.

​
Additional stages

Additional stages can classify documents, invoke agents, or upload text to a vector database. See the workflow stages reference for every stage and its settings. Agent stages may receive values from earlier stages and produce values for later ones. Configure only the tools and data access that the agent needs. Vector search requires the optional vector database, compatible embeddings, and an appropriate text extraction configuration. Agent-Only (API) workflows have a different entry point and are intended for agent interactions rather than the standard document OCR pipeline. Follow Build an agent workflow and Add and configure agent tools for detailed instructions.

​
Optional folder import

Enable Folder Import in the start stage to ingest files from the workflow’s shared input directory. Input and output directories are created under the workflow’s directory in shared. Validate permissions, supported file types, processing behavior, and backup policies before connecting an automated source. Folder export can write results to the workflow’s output directory. To generate a workflow JSON file with an AI agent, including document types and a classifier, see Generate workflow imports with AI. The guide explains import limitations and post-import checks. See troubleshooting if the test does not complete.