Code¶

The Code component executes custom Python code as a step in the workflow graph. It provides direct access to the workflow state and upstream node outputs, making it suitable for data transformation, computation, and custom logic that does not require an LLM.

Ports¶

Inputs¶

Outputs¶

Configuration¶

The Code component accepts the following configuration in extra_config:

Usage¶

1. Add a Code node from the Node Palette (Logic category)
2. Connect upstream nodes to provide input data
3. Write your Python code in the Code Snippet field in the node details panel
4. Connect the Code node's output to downstream nodes

Available variables¶

Your code has access to the following local variables:

Output resolution¶

The Code component determines its output using this priority:

1. If you set the result variable, its string value becomes the output
2. If result is not set, anything written to stdout via print() becomes the output
3. If neither produces content, the output is empty

Using return¶

If your code contains a return statement, it is automatically wrapped in a function. The return value is assigned to result:

data = node_outputs.get("agent_abc123", {})
text = data.get("output", "")
return text.upper()

Jinja2 in code¶

The Code Snippet field supports Jinja2 syntax highlighting. You can use {{ }} template expressions that are resolved before execution:

name = "{{ trigger.text }}"
result = f"Hello, {name}!"

Example¶

A code node that transforms an agent's output into a structured summary:

import json

agent_output = node_outputs.get("agent_abc123", {})
text = agent_output.get("output", "")

lines = text.strip().split("\n")
result = json.dumps({
    "line_count": len(lines),
    "first_line": lines[0] if lines else "",
    "last_line": lines[-1] if lines else "",
    "total_chars": len(text),
})

flowchart LR
    T[Chat Trigger] --> A[Agent]
    A --> C[Code]
    C --> D[Downstream Node]
    M[AI Model] -.->|model| A