Loops

Conductor supports two loop patterns: foreach for iterating over lists, and type: loop for iterative refinement.

Iterative Refinement

Use type: loop to repeatedly execute steps until a condition is met:

steps:
  - id: refine
    type: loop
    max_iterations: 5
    until: "steps.validate.score >= 8"
    steps:
      - id: generate
        type: llm
        prompt: |
          {{if .loop.history}}
          Previous feedback: {{.loop.history.validate.feedback}}
          Improve based on this feedback.
          {{else}}
          Generate initial content.
          {{end}}

      - id: validate
        type: llm
        output_schema:
          type: object
          properties:
            score:
              type: number
            feedback:
              type: string
        prompt: "Rate this content 1-10 and provide feedback: {{.steps.generate.response}}"

Loop Context

Within a loop, access:

• {{.loop.iteration}} - Current iteration (0-based)
• {{.loop.history}} - Previous iteration outputs
• {{.loop.history.stepId.field}} - Specific field from previous iteration

Termination

Loops terminate when:

1. The until condition evaluates to true
2. max_iterations is reached (required, max 100)

Foreach

Iterate over a list with foreach:

steps:
  - id: greet_all
    foreach: "{{.inputs.names}}"
    steps:
      - id: greet
        type: llm
        prompt: "Say hello to {{.item}}"

Access the current item with {{.item}} and index with {{.index}}.

Static Lists

steps:
  - id: process
    foreach:
      items:
        - Alice
        - Bob
        - Carol
    steps:
      - id: greet
        type: llm
        prompt: "Greet {{.item}}"

Dynamic Lists

Iterate over step outputs:

steps:
  - id: fetch
    http.get: https://api.example.com/users

  - id: process
    foreach: "{{.steps.fetch.users}}"
    steps:
      - id: analyze
        type: llm
        prompt: "Analyze user: {{.item.name}}"

Parallel Foreach

Run iterations concurrently:

steps:
  - id: process
    foreach: "{{.inputs.items}}"
    parallel: true
    steps:
      - id: handle
        type: llm
        prompt: "Process {{.item}}"

Loop Results

Access all outputs from a foreach:

steps:
  - id: recipes
    foreach:
      items: [breakfast, lunch, dinner]
    steps:
      - id: generate
        type: llm
        prompt: "Generate a {{.item}} recipe"

  - id: combine
    type: llm
    prompt: "Create meal plan from: {{.steps.recipes.outputs}}"

{{.steps.loopId.outputs}} contains an array of all iteration outputs.