Skip to content

Invoke Other Flows

Reuse flows as composable building blocks by invoking them from other flows with input and output bindings.

QType YAML

flows:
  # Define reusable flow
  - type: Flow
    id: summarize_text
    variables:
      - id: input_text
        type: text
      - id: output_summary
        type: text
    inputs: [input_text]
    outputs: [output_summary]
    steps:
      - type: LLMInference
        id: summarizer
        model: my_model
        inputs: [input_text]
        outputs: [output_summary]

  # Main flow invokes the reusable flow
  - type: Flow
    id: main
    variables:
      - id: article
        type: text
      - id: summary
        type: text
    inputs: [article]
    outputs: [summary]
    steps:
      - type: InvokeFlow
        id: get_summary
        flow: summarize_text           # Reference to flow by ID
        input_bindings:
          input_text: article          # Map flow input to step variable
        output_bindings:
          output_summary: summary      # Map flow output to step variable

Explanation

  • InvokeFlow: Step type that executes another flow with variable mapping
  • flow: ID of the flow to invoke (must be defined in the application)
  • input_bindings: Maps flow input variables to the invoking step's variables (format: flow_input_name: step_variable_name)
  • output_bindings: Maps flow output variables to the invoking step's variables (format: flow_output_name: step_variable_name)
  • Reusability: Flows can be invoked multiple times with different bindings

Complete Example

id: text_analysis_pipeline
description: Demonstrates invoking reusable flows for text analysis

models:
  - type: Model
    id: nova_lite
    provider: aws-bedrock
    model_id: amazon.nova-lite-v1:0

flows:
  # Reusable flow for text summarization
  - type: Flow
    id: summarize_text
    description: Summarizes input text
    variables:
      - id: input_text
        type: text
      - id: output_summary
        type: text
    inputs:
      - input_text
    outputs:
      - output_summary
    steps:
      - type: LLMInference
        id: summarize_step
        model: nova_lite
        system_message: "You provide concise summaries."
        inputs: [input_text]
        outputs: [output_summary]

  # Reusable flow for sentiment analysis
  - type: Flow
    id: analyze_sentiment
    description: Analyzes sentiment of text
    variables:
      - id: input_for_sentiment
        type: text
      - id: output_sentiment
        type: text
    inputs:
      - input_for_sentiment
    outputs:
      - output_sentiment
    steps:
      - type: LLMInference
        id: sentiment_step
        model: nova_lite
        system_message: "Analyze sentiment. Respond with only: positive, negative, or neutral."
        inputs: [input_for_sentiment]
        outputs: [output_sentiment]

  # Main processing flow
  - type: Flow
    id: main
    description: Orchestrates text analysis using multiple flows
    variables:
      - id: article_text
        type: text
      - id: summary
        type: text
      - id: sentiment
        type: text
      - id: report
        type: text
    inputs:
      - article_text
    outputs:
      - report
    steps:
      # Invoke summarization flow
      - type: InvokeFlow
        id: get_summary
        flow: summarize_text
        input_bindings:
          input_text: article_text
        output_bindings:
          output_summary: summary

      # Invoke sentiment analysis flow
      - type: InvokeFlow
        id: get_sentiment
        flow: analyze_sentiment
        input_bindings:
          input_for_sentiment: article_text
        output_bindings:
          output_sentiment: sentiment

      # Combine results into a report
      - type: PromptTemplate
        id: create_report
        template: |
          Analysis Report
          ---------------
          Sentiment: {sentiment}
          Summary: {summary}
        inputs: [summary, sentiment]
        outputs: [report]

Run it: 

qtype run examples/data_processing/invoke_other_flows.qtype.yaml \
  --flow main \
  --input '{"article_text": "Your article text here..."}'

See Also