Skip to content
Conductor

Inputs and Outputs

Workflows accept inputs and produce outputs for reusability and composition.

Inputs

Define parameters your workflow accepts:

name: greet
inputs:
  name:
    type: string
    description: Person to greet
    required: true
  enthusiasm:
    type: number
    default: 5
steps:
  - id: greet
    llm:
      prompt: "Greet ${inputs.name} with enthusiasm level ${inputs.enthusiasm}"

Pass inputs when running:

Terminal window
conductor run greet.yaml -i name="Alice" -i enthusiasm=10

Input Types

  • string - Text values
  • number - Numeric values
  • boolean - true/false
  • object - Structured JSON data
  • array - List of values

Required vs Optional

Mark inputs as required or provide defaults:

inputs:
  required_param:
    type: string
    required: true
  optional_param:
    type: string
    default: "default value"

Outputs

Return data from workflows:

name: analyze
steps:
  - id: process
    llm:
      prompt: "Analyze this data"
outputs:
  result: ${steps.process.output}
  metadata:
    timestamp: ${steps.process.startTime}
    model: ${steps.process.model}

Access outputs when composing workflows or using them programmatically.

Output Formats

Specify how outputs should be formatted and validated using the format field:

outputs:
  - name: recipe
    type: string
    value: "{{.steps.generate.response}}"
    description: Generated recipe with formatting
    format: markdown


  - name: config
    type: string
    value: "{{.steps.extract.response}}"
    description: JSON configuration
    format: json


  - name: count
    type: string
    value: "{{.steps.calculate.response}}"
    description: Numeric count
    format: number


  - name: script
    type: string
    value: "{{.steps.generate_code.response}}"
    description: Python code with syntax highlighting
    format: code:python

Supported formats:

  • string - Plain text (default, no validation)
  • number - Validates numeric values (integers, floats, scientific notation)
  • markdown - Markdown text with CLI rendering and formatting
  • json - Validates JSON and pretty-prints with 2-space indentation
  • code - Code without syntax highlighting
  • code:<language> - Code with syntax highlighting (e.g., code:python, code:javascript, code:go)

Format validation:

Output format validation occurs after the output value expression is evaluated. If validation fails, the workflow fails with an OutputValidationError. Error messages are generic to avoid exposing sensitive data, but full details are logged with the run ID for authorized debugging.

CLI display formatting:

When outputs are displayed via conductor run or conductor runs show:

  • Formatting is applied only when stdout is an interactive TTY
  • Markdown renders with headers, lists, and emphasis formatting
  • JSON is pretty-printed with 2-space indentation
  • Code is syntax-highlighted when a language is specified
  • Piped output contains no ANSI codes for clean machine processing

Variable Syntax

Reference inputs, step outputs, and environment variables:

# Input reference
${inputs.paramName}


# Step output reference
${steps.stepId.output}


# Environment variable
${ENV_VAR_NAME}


# Nested object access
${steps.stepId.output.field.nested}


# Array element
${steps.stepId.output[0]}

Step Outputs

Each step produces an output field. For LLM steps, this is the generated text. For other actions, it’s the result of the operation:

steps:
  - id: read
    file:
      action: read
      path: data.txt
  - id: process
    llm:
      prompt: "Summarize: ${steps.read.output}"

Complex Outputs

Structure outputs with nested data:

outputs:
  summary: ${steps.analyze.output}
  details:
    processedAt: ${steps.analyze.timestamp}
    items: ${steps.collect.outputs}
    status: "complete"

Environment Variables

Reference environment variables for secrets and configuration:

steps:
  - id: api_call
    http:
      url: https://api.example.com
      headers:
        Authorization: "Bearer ${API_TOKEN}"

Set before running:

Terminal window
export API_TOKEN="your-token"
conductor run workflow.yaml