Skip to main content

​
YAML Workflow Configuration

Flo AI supports defining entire multi-agent workflows in YAML, making it easy to version control, share, and manage complex AI systems.

​
Basic Workflow Structure

basic-workflow.yaml
metadata:
  name: "content-analysis-workflow"
  version: "1.0.0"
  description: "Multi-agent content analysis pipeline"

arium:
  agents:
    - name: "analyzer"
      role: "Content Analyst"
      job: "Analyze the input content and extract key insights."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
        temperature: 0.3
    
    - name: "summarizer"
      role: "Content Summarizer"
      job: "Create a concise summary based on the analysis."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
        temperature: 0.2

  workflow:
    start: "analyzer"
    edges:
      - from: "analyzer"
        to: ["summarizer"]
    end: ["summarizer"]

​
Advanced Workflow Patterns

​
Conditional Routing

conditional-workflow.yaml
metadata:
  name: "support-routing-workflow"
  version: "1.0.0"

arium:
  agents:
    - name: "classifier"
      role: "Request Classifier"
      job: "Classify incoming requests by type and urgency."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
    
    - name: "technical_support"
      role: "Technical Support"
      job: "Handle technical issues and troubleshooting."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "billing_support"
      role: "Billing Support"
      job: "Handle billing and account questions."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "general_support"
      role: "General Support"
      job: "Handle general inquiries and questions."
      model:
        provider: "openai"
        name: "gpt-4o-mini"

  routers:
    - name: "support_router"
      type: "conditional"
      routing_logic: |
        def route_request(memory):
            last_message = str(memory.get()[-1]) if memory.get() else ""
            if "technical" in last_message.lower():
                return "technical_support"
            elif "billing" in last_message.lower():
                return "billing_support"
            else:
                return "general_support"

  workflow:
    start: "classifier"
    edges:
      - from: "classifier"
        to: ["technical_support", "billing_support", "general_support"]
        router: "support_router"
    end: ["technical_support", "billing_support", "general_support"]

​
LLM-Powered Routing

smart-routing-workflow.yaml
metadata:
  name: "smart-content-workflow"
  version: "1.0.0"

arium:
  agents:
    - name: "content_analyzer"
      role: "Content Analyzer"
      job: "Analyze content and determine the best processing approach."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
    
    - name: "technical_writer"
      role: "Technical Writer"
      job: "Create technical documentation and guides."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "creative_writer"
      role: "Creative Writer"
      job: "Create engaging creative content."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "marketing_writer"
      role: "Marketing Writer"
      job: "Create marketing copy and promotional content."
      model:
        provider: "openai"
        name: "gpt-4o"

  routers:
    - name: "content_router"
      type: "smart"
      routing_options:
        technical_writer: "Technical content, documentation, tutorials, code examples"
        creative_writer: "Creative writing, storytelling, fiction, poetry"
        marketing_writer: "Marketing copy, sales content, campaigns, advertisements"
      model:
        provider: "openai"
        name: "gpt-4o-mini"
        temperature: 0.1

  workflow:
    start: "content_analyzer"
    edges:
      - from: "content_analyzer"
        to: ["technical_writer", "creative_writer", "marketing_writer"]
        router: "content_router"
    end: ["technical_writer", "creative_writer", "marketing_writer"]

​
Reflection Patterns

​
A→B→A→C Pattern

reflection-workflow.yaml
metadata:
  name: "reflection-writing-workflow"
  version: "1.0.0"

arium:
  agents:
    - name: "writer"
      role: "Content Writer"
      job: "Write initial content based on requirements."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "critic"
      role: "Content Critic"
      job: "Review and critique the written content."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "finalizer"
      role: "Content Finalizer"
      job: "Create the final polished version."
      model:
        provider: "openai"
        name: "gpt-4o"

  routers:
    - name: "reflection_router"
      type: "reflection"
      flow_pattern: ["writer", "critic", "writer"]
      model:
        provider: "openai"
        name: "gpt-4o-mini"

  workflow:
    start: "writer"
    edges:
      - from: "writer"
        to: ["critic"]
      - from: "critic"
        to: ["writer"]
        router: "reflection_router"
      - from: "writer"
        to: ["finalizer"]
    end: ["finalizer"]

​
Plan-Execute Workflows

​
Cursor-Style Development

plan-execute-workflow.yaml
metadata:
  name: "development-workflow"
  version: "1.0.0"

arium:
  agents:
    - name: "planner"
      role: "Development Planner"
      job: "Create detailed execution plans for development tasks."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "developer"
      role: "Code Developer"
      job: "Implement features according to the plan."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "tester"
      role: "Code Tester"
      job: "Test implementations and validate functionality."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "reviewer"
      role: "Code Reviewer"
      job: "Review and approve completed work."
      model:
        provider: "openai"
        name: "gpt-4o"

  routers:
    - name: "plan_execute_router"
      type: "plan_execute"
      settings:
        planner_agent: "planner"
        executor_agent: "developer"
        reviewer_agent: "reviewer"
        max_iterations: 3
      model:
        provider: "openai"
        name: "gpt-4o-mini"

  workflow:
    start: "planner"
    edges:
      - from: "planner"
        to: ["developer"]
      - from: "developer"
        to: ["tester"]
      - from: "tester"
        to: ["reviewer"]
        router: "plan_execute_router"
    end: ["reviewer"]

​
Parallel Processing

​
Fan-out/Fan-in Pattern

parallel-workflow.yaml
metadata:
  name: "parallel-analysis-workflow"
  version: "1.0.0"

arium:
  agents:
    - name: "coordinator"
      role: "Analysis Coordinator"
      job: "Coordinate parallel analysis tasks."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
    
    - name: "sentiment_analyzer"
      role: "Sentiment Analyzer"
      job: "Analyze sentiment and emotional tone."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
    
    - name: "topic_extractor"
      role: "Topic Extractor"
      job: "Extract main topics and themes."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "keyword_analyzer"
      role: "Keyword Analyzer"
      job: "Extract and analyze keywords."
      model:
        provider: "openai"
        name: "gpt-4o-mini"
    
    - name: "synthesizer"
      role: "Analysis Synthesizer"
      job: "Combine all analysis results into a comprehensive report."
      model:
        provider: "openai"
        name: "gpt-4o"

  workflow:
    start: "coordinator"
    edges:
      - from: "coordinator"
        to: ["sentiment_analyzer", "topic_extractor", "keyword_analyzer"]
      - from: ["sentiment_analyzer", "topic_extractor", "keyword_analyzer"]
        to: ["synthesizer"]
    end: ["synthesizer"]

​
Complex Multi-Step Workflows

​
Research and Analysis Pipeline

research-workflow.yaml
metadata:
  name: "research-pipeline"
  version: "1.0.0"

arium:
  agents:
    - name: "researcher"
      role: "Research Agent"
      job: "Conduct research on the given topic."
      model:
        provider: "openai"
        name: "gpt-4o"
      tools:
        - name: "web_search"
          description: "Search the web for information"
        - name: "database_query"
          description: "Query internal databases"
    
    - name: "analyzer"
      role: "Data Analyst"
      job: "Analyze research data and identify patterns."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "synthesizer"
      role: "Content Synthesizer"
      job: "Synthesize findings into coherent insights."
      model:
        provider: "anthropic"
        name: "claude-3-5-sonnet-20240620"
    
    - name: "validator"
      role: "Content Validator"
      job: "Validate accuracy and completeness of findings."
      model:
        provider: "openai"
        name: "gpt-4o"
    
    - name: "presenter"
      role: "Presentation Creator"
      job: "Create final presentation of findings."
      model:
        provider: "openai"
        name: "gpt-4o"

  workflow:
    start: "researcher"
    edges:
      - from: "researcher"
        to: ["analyzer"]
      - from: "analyzer"
        to: ["synthesizer"]
      - from: "synthesizer"
        to: ["validator"]
      - from: "validator"
        to: ["presenter"]
    end: ["presenter"]

​
Workflow Configuration Options

​
Memory Configuration

memory-workflow.yaml
metadata:
  name: "memory-workflow"
  version: "1.0.0"

arium:
  memory:
    type: "message"
    max_messages: 10
    include_metadata: true
  
  agents:
    - name: "conversational_agent"
      role: "Conversational Agent"
      job: "Maintain context across multiple interactions."
      model:
        provider: "openai"
        name: "gpt-4o"
      memory:
        enabled: true
        max_context: 5

  workflow:
    start: "conversational_agent"
    end: ["conversational_agent"]

​
Error Handling

error-handling-workflow.yaml
metadata:
  name: "robust-workflow"
  version: "1.0.0"

arium:
  error_handling:
    max_retries: 3
    retry_delay: 1.0
    fallback_agent: "fallback_handler"
    timeout: 30
  
  agents:
    - name: "primary_agent"
      role: "Primary Processor"
      job: "Main processing agent."
      model:
        provider: "openai"
        name: "gpt-4o"
      retries: 2
      timeout: 20
    
    - name: "fallback_handler"
      role: "Fallback Handler"
      job: "Handle errors and provide fallback responses."
      model:
        provider: "openai"
        name: "gpt-4o-mini"

  workflow:
    start: "primary_agent"
    edges:
      - from: "primary_agent"
        to: ["fallback_handler"]
        on_error: true
    end: ["primary_agent", "fallback_handler"]

​
Loading and Executing YAML Workflows

​
Basic Loading

from flo_ai.arium import AriumBuilder

# Load workflow from file
workflow = AriumBuilder.from_yaml('workflow.yaml')

# Execute workflow
result = await workflow.build_and_run(["Input data here"])

​
Advanced Execution

# Load with custom configuration
workflow = AriumBuilder.from_yaml(
    'workflow.yaml',
    config_overrides={
        'agents.analyzer.model.temperature': 0.1,
        'agents.summarizer.model.temperature': 0.5
    }
)

# Execute with variables
result = await workflow.build_and_run(
    ["Input data"],
    variables={
        'user_id': '123',
        'priority': 'high'
    }
)

​
Workflow Validation

# Validate workflow before execution
try:
    workflow = AriumBuilder.from_yaml('workflow.yaml')
    workflow.validate()
    print("✅ Workflow is valid")
except Exception as e:
    print(f"❌ Workflow validation failed: {e}")

​
Best Practices

​
YAML Structure

  1. Use meaningful names: Choose descriptive agent and workflow names
  2. Version your workflows: Always include version numbers
  3. Document thoroughly: Add descriptions for all components
  4. Validate schemas: Use YAML schema validation tools

​
Performance Optimization

# Optimize for performance
arium:
  agents:
    - name: "fast_agent"
      model:
        provider: "openai"
        name: "gpt-4o-mini"  # Use faster model
        temperature: 0.1      # Lower temperature
        max_tokens: 500       # Limit response length
      cache_ttl: 3600         # Cache for 1 hour
      timeout: 10            # Short timeout

​
Security Considerations

# Secure workflow configuration
arium:
  agents:
    - name: "secure_agent"
      model:
        provider: "openai"
        name: "gpt-4o"
        temperature: 0.1     # Lower temperature for consistency
      max_tokens: 200        # Limit response length
      timeout: 5            # Short timeout
      retries: 1            # Limit retries
This YAML-based approach makes complex multi-agent workflows declarative, versionable, and easily shareable across teams!
⌘I