Human in the Loop Node

Pause workflow execution to request human approval, rejection, or editing of data items.

Purpose

The Human in the Loop node enables human oversight and intervention in automated workflows. When the workflow reaches a Human in the Loop node, it pauses execution requesting review of the input data. The workflow continues only after the human has approved or rejected all items.

Parameters

runMode

The Human in the Loop node can run in 2 different modes:

• Item: The node will process each item in an input array individually. Each item will be presented separately for review.
• Input: The node will process the entire input as a single item for review.

description (optional)

A text description explaining what needs to be reviewed. This provides context for the reviewer.

How It Works

1. Workflow Pauses: When the workflow reaches a Human in the Loop node, execution pauses and the workflow status changes to "human_in_the_loop".
2. Visual Indicators: The Human in the Loop node displays with an amber/yellow border and pulsing animation to indicate it's awaiting review.
3. Toast Notification: A "Human Action Required" toast message appears with an amber icon when viewing the workflow or run.
4. Human Review: Navigate to the Human in the Loop node and review each item in the review interface.
5. Decision Making: For each item, the reviewer can:
   • Approve: Accept the item as-is
   • Edit & Approve: Modify the item and then approve it (not available for images)
   • Reject: Reject the item (output will be undefined)
6. Workflow Continuation: After reviewing all items, click the "Continue Workflow" button to resume execution.

Input and Output

Input

The Human in the Loop node accepts any type of input:

• Arrays of items (processed individually in item mode)
• Single values (processed as one item in input mode)
• Images (rendered with visual preview) - supports multiple formats:
  • Data URLs: data:image/png;base64,iVBORw0...
  • Raw base64 strings: iVBORw0KGgoAAAANSUhEUg...
  • Public URLs: https://example.com/image.jpg
• Text and JSON (editable during review)
• Complex objects or primitive values

Output

The output is an array of review results:

[
  { input: "original value", decision: "approved", output: "original value" },
  {
    input: "edited value",
    decision: "accepted-with-edits",
    output: "edited value",
  },
  { input: "rejected value", decision: "rejected", output: undefined },
];

Each result contains:

• input: The original input value
• decision: One of "approved", "accepted-with-edits", or "rejected"
• output: The final value (original if approved, edited if accepted-with-edits, undefined if rejected)

Examples

Example 1: Review Generated Content

Generate AI content and have a human review it before publishing:

AI Node → Human in the Loop Node (item mode) → Publish Node

The Human in the Loop node will present each generated piece of content for approval or editing.

Example 2: Data Quality Check

Process data and have a human verify the results:

Transform Node → Human in the Loop Node (input mode) → Save to Database Node

The Human in the Loop node presents the entire transformed dataset for review before saving.

Example 3: Image Approval

Generate or process images and get human approval:

Image Generation Node → Human in the Loop Node (item mode) → Upload Node

Base64-encoded images (format: data:image/png;base64,... or data:image/jpeg;base64,...) are automatically detected and displayed visually in the review interface. Images can be approved or rejected, but cannot be edited. The approved images are passed through to downstream nodes in their original base64 format.

Workflow Behavior

Multiple Human in the Loop Nodes

If a workflow contains multiple Human in the Loop nodes:

• Each Human in the Loop node operates independently
• Nodes in parallel branches can be reviewed simultaneously
• Downstream nodes only execute after their upstream Human in the Loop node is complete
• The workflow status remains "human_in_the_loop" until all Human in the Loop nodes are complete

Workflow Status

The workflow status is displayed with color-coded badges:

• Human Action Required (amber badge): At least one Human in the Loop node is awaiting review
  • The Human in the Loop node displays with an amber border and pulsing animation
  • A toast notification appears saying "Human Action Required"
  • Visible in workflow runs list and individual run pages
• running (blue badge): All Human in the Loop nodes are complete and workflow is executing
• completed (green badge): All nodes including Human in the Loop nodes are finished

Visual Feedback

When a Human in the Loop node is awaiting review:

• Node Appearance: Amber/yellow border with a pulsing animation
• Toast Notification: "Human Action Required" message with amber icon appears in:
  • Workflow builder page (when running a test)
  • Individual workflow run page
• Status Badge: Shows "Human Action Required" with amber background in the runs list
• Clock Icon: A small clock icon appears on the node to indicate waiting status

Best Practices

1. Use Descriptive Text: Provide a clear description to help reviewers understand what they're approving
2. Item Mode for Lists: Use item mode when reviewing multiple independent items
3. Input Mode for Batches: Use input mode when the entire dataset needs to be reviewed as a whole
4. Consider Timing: Human in the Loop nodes pause workflows indefinitely until reviewed - use them strategically