Understanding Routines
Learn how routines provide structured task execution and reusable workflows in Ax.
What is a Routine?
In Ax, based on OpenAI's Swarm framework, a Routine is a predefined sequence of steps that agents can follow to accomplish specific tasks. Routines provide structured task execution, reusable workflows, and consistent behavior patterns.
Structure
- •Clear I/O
Well-defined inputs and outputs
- •Execution Steps
Defined sequence of operations
- •Success Criteria
Clear completion conditions
Execution Flow
- •Input Validation
Verify and validate inputs
- •Step Execution
Process each defined step
- •Result Formatting
Format and return results
Types of Routines
1. Task Routines
async def get_city_weather(city: str):
"""Get current weather for a city using Weather API"""
# 1. Validate city input
# 2. Make API request to weather service
# 3. Parse weather data response
# 4. Return formatted weather data
return {
"city": city,
"temp": temp_celsius,
"conditions": conditions,
"humidity": humidity
}- • Specific task execution
- • Clear start and end
- • Defined outcomes
2. Decision Routines
- • Evaluation logic
- • Routing decisions
- • Conditional flows
3. Integration Routines
- • External system interaction
- • Data transformation
- • API handling
Best Practices
Modularity
- • Keep routines focused
- • Enable reusability
- • Maintain independence
Error Handling
- • Define failure modes
- • Implement recovery steps
- • Log important events
Documentation
- • Clear purpose
- • Input/output specifications
- • Usage examples