Skip to main content

Workflow Setup Guide

This guide walks you through the common process of creating and configuring workflows in Unstract. All workflow types follow the same foundational setup process, with specific differences highlighted in their respective deployment guides.

Prerequisites​

Before creating a workflow, you must export a Prompt Studio project as a tool. This exported tool will be used to process your data within the workflow.

Workflow Setup Process​

Workflows are configured through a streamlined 4-step wizard interface:

Workflow Creation Wizard

Step 1: Configure Source Connector​

Select and configure where your input data comes from.

  • Available Source Types:
    • File System (for ETL and Task pipelines)
    • API (for API deployments)

See Source Configuration for source specific settings

Step 2: Configure Destination Connector​

Select and configure where your processed data will be stored.

  • Available Destination Types:
    • Database (for ETL and HITL workflows)
    • File System (for Task pipelines)
    • API (for API deployments - no configuration needed)

See Destination Configuration for destination specific settings

info

Refer Connectors for filesystem or database connector setup

Step 3: Select Exported Prompt Studio Project​

Choose the tool that will process your data.

  • Actions:
    1. Your previously exported Prompt Studio projects will be available for selection
    2. Use Change Prompt Studio project to select a different project
    3. Click Configure Settings to modify tool-specific settings

Tool Settings​

Clicking Configure Settings opens the Tool Settings panel. These settings apply to every run of the workflow, whether it is deployed as an API, ETL or Task pipeline.

SettingDescription
Challenger LLMLLM to use for LLMChallenge
Enable LLMChallengeEnables LLMChallenge
Enable highlightEnables highlight data and line-level confidence for extracted values
Enable SummarizedExtractionEnables SummarizedExtraction
Enable word confidence(Optional) Enables word-level confidence for extracted values. Requires Enable highlight
Enable SinglePass ExtractionEnables SinglePass Extraction
Enabling Highlight and Word Confidence​

The numbers below match the markers in the screenshot.

  1. Navigate to Workflows in the side navigation.
  2. Open your workflow.
  3. In Select Exported Prompt Studio project, click Configure Settings.
  4. Check Enable highlight to get highlight data and line-level confidence.
  5. (Optional) Check Enable word confidence to also get word-level confidence. This requires Enable highlight. Leave it unchecked if line-level confidence is enough.
  6. Click Save to apply your changes.
To getEnable
Line-level confidenceEnable highlight
Word-level confidenceEnable highlight and Enable word confidence

Workflow Tool Settings

Required for confidence-based HITL rules

HITL confidence rules and the Highlight Data API depend on Enable highlight (and preferably Enable word confidence) being turned on here. Highlight data is collected only when the exported project's LLM profile uses LLMWhisperer as its Text Extraction service. Without these, no confidence data is produced and HITL treats every field as 100% confident, so low-confidence rules such as confidence <= 80 never send documents to review. Other comparisons (for example, confidence >= 80) still evaluate against that default score of 100.

The Highlighting settings in Prompt Studio apply only within Prompt Studio and are not used by deployed workflows.

info

At this point, you can run and test the workflow, select Actions -> Run Workflow to ensure it's working as expected. This is optional but recommended.

Step 4: Deploy Workflow​

Deploy your workflow with the appropriate deployment type. After deployment, you'll see a confirmation message with your workflow identifier.

Deployment Types Overview​

Different deployment types are optimized for specific use cases:

Deployment TypeSourceDestinationPrimary Use Case
ETL PipelineFile SystemDatabaseAutomated data extraction and loading into databases
Task PipelineFile SystemFile SystemAutomated data extraction and loading into filesystem
API DeploymentAPIAPIOn-demand API-based processing
HITL (Human-in-the-Loop)File SystemDatabaseETL with human review and approval workflow

Monitoring and Debugging​

  • Check the Debug Panel at the bottom of the interface for a high level view of what's being executed.
  • For more detailed logs, check the Logs side navigation menu -> Workflow Sessions -> select your desired workflow session.