Loop¶

The Loop component iterates over an array, executing a body of nodes once for each item. Results from each iteration are collected into an output array.

Ports¶

Inputs¶

Outputs¶

Configuration¶

The Loop component accepts the following configuration in extra_config:

Usage¶

1. Add a Loop node from the Node Palette (Logic category)
2. Connect an upstream node that produces an array
3. Add body nodes that should execute for each item
4. Connect the Loop to the first body node via a loop_body edge
5. Connect the last body node back to the Loop via a loop_return edge
6. Connect the Loop's output to downstream nodes

Special edge labels¶

Loop nodes use two special edge labels that bypass standard type-compatibility validation:

Canvas rendering¶

On the canvas, loop_return edges render with a distinctive visual style:

• Path: Routes below the nodes (right from source, down, left, up to target) with rounded 90-degree corners
• Stroke: Dashed line
• Label: "return"

This visual treatment makes loop structures easy to identify at a glance.

Data flow¶

For each item in the input array:

1. The Loop emits the current item to the body node via the loop_body edge
2. The body node(s) process the item
3. The last body node returns its result via the loop_return edge
4. The result is appended to the Loop's results array

The Loop also sets a _loop key in the state containing {"items": [...]} for internal orchestrator use.

Example¶

Process each search result individually with an agent:

flowchart LR
    T[Chat Trigger] --> S[Search Agent]
    S --> L[Loop]
    L -->|loop_body| P[Process Agent]
    P -.->|loop_return| L
    L --> R[Summary Agent]

Loop Extra Config:

{
  "source_node": "agent_search",
  "field": "results"
}

If the search agent returns {"results": ["item1", "item2", "item3"]}, the Process Agent runs three times -- once for each item. The Loop collects all three results into its results array, which the Summary Agent then receives.