This guide explains how to create, test, and deploy compound flows that implement complex, multi-stage processing pipelines. Compound flows provide sophisticated capabilities for implementing business logic, custom processing stages, and integration with existing elemental flows. 🔄💡

​
Development Lifecycle

Creating a compound flow follows a systematic process:

1

Configuration: Define your compound flow using YAML

2

Testing: Validate the entire processing pipeline

3

Deployment: Make your flow available

4

Execution: Run your multi-stage flow in production

5

Updates: Iterate and improve

Let’s explore each step in detail.

Use OpenAI (gpt-4o) or Anthropic (claude-3.5-sonnet) models when building complex workflows to ensure optimal performance. Checkout Available Models

​
Step 1: YAML Configuration

​
Step 2: Testing

Test your compound flow to ensure all stages work correctly:

Python
from mira_sdk import MiraClient, CompoundFlow
from mira_sdk.exceptions import FlowError

client = MiraClient(config={"API_KEY": "YOUR_API_KEY"})     # Initialize Mira Client
flow = CompoundFlow(source="/path/to/compound_flow.yaml")           # Load flow configuration

test_input = {                                              # Prepare test inputs
    "prime_input_1": "test data",
    "prime_input_2": "test parameters"
}

try:
    response = client.flow.test(flow, test_input)           # Test entire pipeline
    print("Test response:", response)
except FlowError as e:
    print("Test failed:", str(e))                           # Handle test failure

​
Step 3: Deployment

Deploy your compound flow to make it available:

Python
from mira_sdk import MiraClient, CompoundFlow
from mira_sdk.exceptions import FlowError

flow = CompoundFlow(source="/path/to/compound_flow.yaml")           # Load flow configuration

try:
    client.flow.deploy(flow)                               # Deploy to platform
    print("Compound flow deployed successfully!")          # Success message
except FlowError as e:
    print(f"Deployment error: {str(e)}")                   # Handle deployment error

​
Step 4: Execution

After deployment, execute your compound flow:

Python
from mira_sdk import MiraClient
from mira_sdk.exceptions import FlowError

flow_name = "your-username/your-flow-name"                 # Flow identifier
input_data = {                                             # Execution inputs
    "prime_input_1": "production data",
    "prime_input_2": "production parameters"
}

try:
    result = client.flow.execute(flow_name, input_data)    # Execute workflow
    print("Execution result:", result)                     # Display result
except FlowError as e:
    print("Execution error:", str(e))                      # Handle execution error

​
Step 5: Updates

When you need to modify your compound flow:

Python
from mira_sdk import MiraClient
from mira_sdk.exceptions import FlowError

flow = client.flow.get("your-username/your-flow-name")     # Retrieve existing flow
flow.save("/path/to/compound_flow.yaml")                   # Save for editing

try:
    client.flow.deploy(flow)                               # Deploy updated version
    print("Updated flow deployed successfully!")           # Success message
except FlowError as e:
    print(f"Update error: {str(e)}")                       # Handle update error

# In case you forget to bump the flow version, it will get bumped by default every time you deploy the same flow.

By following these steps, you can create, test, deploy, and manage your own custom compound flows on the Mira Flows platform. 🌟🔧

Build Elemental FlowsImplement RAG Capabilities