Transitions and Tool calling

This example demonstrates how to create a simple workflow with three transitions that perform sequential operations. We’ll build a data processing workflow that extracts, transforms, and validates data to illustrate the core workflow concepts.

Note: This example uses create_mock tools for demonstration purposes to keep the example simple and focused on core workflow concepts. In real-world use cases, you would replace these mock tools with actual tools that perform the specific processing operations your automation requires (such as API calls, data transformations, file operations, or LLM interactions).

Example: Data Processing Workflow

The workflow demonstrates the three core concepts through a simple data processing example:


workflows:
  - name: process_data
    type: stateMachine
    transitions:
      - name: extract_data           # Step 1: Extract raw data
        from: start
        to: data_extracted
        call:
          - tool: create_mock
            arguments:
              input: 'customer_records.csv'
              output: 'John Doe,john@example.com,Premium,2024-01-15'
            as: RAW_DATA
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: ${ RAW_DATA }
 
      - name: transform_data         # Step 2: Transform the data
        from: data_extracted
        to: data_transformed
        call:
          - tool: create_mock
            arguments:
              input: ${ RAW_DATA }
              output:
                name: "John Doe"
                email: "john@example.com"
                tier: "Premium"
                signup_date: "2024-01-15"
            as: STRUCTURED_DATA
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: |
                Transformed Entry Name: {{ STRUCTURED_DATA.name }}
 
      - name: validate_data          # Step 3: Validate the result
        from: data_transformed
        to: end
        call:
          - tool: create_mock
            arguments:
              input: ${ STRUCTURED_DATA }
              output: true
            as: VALIDATION_RESULT
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: 'Data processing complete: {{#if VALIDATION_RESULT}}Validation passed.{{else}}Validation failed.{{/if}}'

Understanding Transitions

Each transition represents a step in your business process and moves the workflow between states:

Workflow starts in the start state
extract_data transition moves to data_extracted state and exports RAW_DATA
transform_data transition moves to data_transformed state and exports STRUCTURED_DATA
validate_data transition moves to end state, completing the workflow

Transition 1: Extract Data


- name: extract_data
  from: start                    # Starting state
  to: data_extracted             # Target state after completion
  call:                         # Tools to execute during this step
    - tool: create_mock
      arguments:
        input: 'customer_records.csv'
        output: 'John Doe,john@example.com,Premium,2024-01-15'
      as: RAW_DATA   # Make result available to next transitions
 
    - tool: create_chat_message
      arguments:
        role: 'assistant'
        content: ${ RAW_DATA }    # Display the extracted data

Key Concepts:

State Progression: Moves workflow from start to data_extracted
Mock Tool: create_mock simulates data extraction for demonstration
Data Export: as: RAW_DATA stores the result for later use
Immediate Display: Chat message shows the extracted data in the interface

Transition 2: Transform Data


- name: transform_data
  from: data_extracted           # Must match the 'to' state from previous transition
  to: data_transformed
  call:
    - tool: create_mock
      arguments:
        input: ${ RAW_DATA }    # Access data from previous transition
        output:
          name: "John Doe"
          email: "john@example.com"
          tier: "Premium"
          signup_date: "2024-01-15"
      as: STRUCTURED_DATA
 
    - tool: create_chat_message
      arguments:
        role: 'assistant'
        content: |
          Transformed Entry Name: {{ STRUCTURED_DATA.name }}

Key Concepts:

Data Access: Use ${ RAW_DATA } to access exported variables from previous transitions

Transition 3: Validate Data


- name: validate_data
  from: data_transformed
  to: end                       # 'end' is a special state that completes the workflow
  call:
    - tool: create_mock
      arguments:
        input: ${ STRUCTURED_DATA }
        output: true            # Boolean output for validation result
      as: VALIDATION_RESULT
 
    - tool: create_chat_message
      arguments:
        role: 'assistant'
        content: 'Data processing complete: {{#if VALIDATION_RESULT}}Validation passed.{{else}}Validation failed.{{/if}}'

Key Concepts:

End State: Workflows complete when they reach the end state
Template Logic: Use {{#if VALIDATION_RESULT}} ... {{else}} ... {{/if}} syntax for conditional logic

Tool Calling

Tools perform the actual work during transitions and can be chained together:


call:
  - tool: create_mock           # First tool call
    arguments:
      input: ${ PREVIOUS_RESULT }
      output: 'Processed result'
    as: NEW_RESULT
 
  - tool: create_chat_message    # Second tool call in same transition
    arguments:
      role: 'assistant'
      content: ${ NEW_RESULT }       # Uses result from first tool call

Key Concepts:

Tool Chaining: Multiple tools can execute in sequence within one transition
Immediate Access: Later tool calls in the same transition can access as results
Chat Integration: create_chat_message displays results in the Studio interface

Data Flow between transitions

Variables flow between transitions using the as mechanism:

Data Flow Pattern:

Step 1: RAW_DATA contains the raw CSV string
Step 2: STRUCTURED_DATA contains the parsed object
Step 3: VALIDATION_RESULT contains the boolean validation result
Scope: Variables are available to all subsequent transitions in the same workflow

Running This Example

This example is already available in Loopstack Studio:

Navigate to the Studio interface
Switch to the examples workspace
Select “Example 3: Transitions and Tool Calling” from the available automations
Watch each transition execute in order, displaying results in the chat interface

Configuration Location: You can also view the complete configuration file at src/config/examples/basic/transitions-and-tool-calling-example.yaml

Complete Example:


include:
  - core/tools/create-mock.yaml
  - core/tools/create-chat-message.yaml
 
pipelines:
  - name: transitions_tool_example
    title: "Example 3: Transitions and Tool Calling"
    type: root
    workspace: examples
    sequence:
      - workflow: process_data
 
workflows:
  - name: process_data
    type: stateMachine
    transitions:
      - name: extract_data           # Step 1: Extract raw data
        from: start
        to: data_extracted
        call:
          - tool: create_mock
            arguments:
              input: 'customer_records.csv'
              output: 'John Doe,john@example.com,Premium,2024-01-15'
            as: RAW_DATA
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: ${ RAW_DATA }
 
      - name: transform_data         # Step 2: Transform the data
        from: data_extracted
        to: data_transformed
        call:
          - tool: create_mock
            arguments:
              input: ${ RAW_DATA }
              output:
                name: "John Doe"
                email: "john@example.com"
                tier: "Premium"
                signup_date: "2024-01-15"
            as: STRUCTURED_DATA
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: |
                Transformed Entry Name: {{ STRUCTURED_DATA.name }}
 
      - name: validate_data          # Step 3: Validate the result
        from: data_transformed
        to: end
        call:
          - tool: create_mock
            arguments:
              input: ${ STRUCTURED_DATA }
              output: true
            as: VALIDATION_RESULT
 
          - tool: create_chat_message
            arguments:
              role: 'assistant'
              content: 'Data processing complete: {{#if VALIDATION_RESULT}}Validation passed.{{else}}Validation failed.{{/if}}'