az-cortex-sdk 0.2.0

Python SDK for Azure Cortex ML API

Azure Cortex SDK

A Python SDK for interacting with the Azure Cortex ML API, providing seamless integration for machine learning operations including environment management, model versioning, inference handling, and conversation management.

Features

Core ML Operations

• Environment Management: Create, update, and manage Azure ML environments with conda dependencies
• Environment Versioning: Register, archive, and set default environment versions
• Model Management: Upload, register, and manage ML models with metadata and tags
• Model Versioning: Register model versions from files or buffers with version control
• Endpoint Management: Create and manage online endpoints for model serving
• Deployment Operations: Deploy models with traffic allocation, scaling, and blue-green deployments
• Scoring Scripts: Manage Python scoring scripts for custom inference logic
• Secrets Management: Securely manage Azure Key Vault secrets for API keys and credentials

Inference & LLM

• Direct Inference: Perform real-time inference on deployed ML endpoints
• LLM Configuration: Configure OpenAI, Azure OpenAI, and other LLM providers
• LLM Inference: Perform chat completions with streaming support
• Message Management: Create and manage conversation messages with auto-inference
• Conversation Management: Handle multi-turn conversations with escalation and resolution
• Inference Tracking: List, filter, and annotate all inference requests

Developer Experience

• Authentication: OAuth2 client credentials flow with automatic token refresh
• Type Safety: Full type hints and Pydantic models for request/response validation
• Error Handling: Comprehensive exception handling with retry mechanisms
• File Operations: Support for large file uploads with streaming and buffer support
• Test Data Management: Built-in utilities for creating and cleaning up test data
• Pagination: Automatic pagination support for all list operations
• Filtering: Advanced filtering options for conversations, messages, and inferences

Installation

pip install az-cortex-sdk

Quick Start

from az_cortex_sdk import CortexClient

# Initialize client from environment variables
client = CortexClient.from_env()

# Or initialize with explicit credentials
client = CortexClient(
    base_url="https://api.az-dev.nearlyhuman.ai",
    client_id="your-client-id",
    client_secret="your-client-secret",
    tenant="demo"
)

# Check API health
health = client.health_check()
print(f"API Status: {health['status']}")

# List environments
environments = client.environments.list()
print(f"Found {len(environments.items)} environments")

# List models
models = client.models.list()
for model in models.items:
    print(f"Model: {model.name}")

# Perform LLM inference
response = client.llm.infer(
    config_name="my-llm-config",
    messages=[{"role": "user", "content": "Hello!"}]
)
print(f"Response: {response.content}")

Configuration

Set environment variables:

export CORTEX_API_BASE_URL="https://api.az-dev.nearlyhuman.ai"
export CORTEX_CLIENT_ID="your-client-id"
export CORTEX_CLIENT_SECRET="your-client-secret"
export CORTEX_TENANT="demo"

Documentation

Getting Started

• Getting Started Guide - Comprehensive setup and usage guide
• Examples - Practical usage examples

Core Guides

• API Guide - Endpoints, deployments, inference, messages, and conversations
• LLM Guide - LLM configuration, inference, and streaming
• Version Management - Environment and model version management
• Secrets Management - Azure Key Vault secrets management

Reference

• API Reference - Complete API documentation below

Examples

Environment & Model Management

# Create and manage environments
client.environments.create(
    name="my-env",
    image="mcr.microsoft.com/azureml/minimal-ubuntu20.04-py38-cpu-inference:latest",
    conda_file="path/to/conda.yml",
    description="Production environment"
)

# Register environment version
client.environment_versions.register(
    environment_name="my-env",
    version="1.0.0",
    image="mcr.microsoft.com/azureml/minimal-ubuntu20.04-py38-cpu-inference:latest"
)

# Upload and create model
client.models.upload_and_create(
    file_path="./model.pkl",
    name="my-model",
    description="My ML model",
    tags={"framework": "sklearn"}
)

# Register model version from file
client.model_versions.register_from_file(
    file_path="./model.pkl",
    model_name="my-model",
    version="1.0.0",
    description="Initial release"
)

Endpoint & Deployment

# Create endpoint
client.endpoints.create(name="my-endpoint")

# Create deployment
client.deployments.create(
    endpoint_name="my-endpoint",
    name="blue",
    model_name="my-model",
    model_version="1.0.0",
    environment_name="my-env",
    environment_version="1.0.0",
    instance_type="Standard_DS3_v2",
    scoring_script_name="score.py",
    instance_count=1,
    traffic_percentage=100
)

# Update traffic allocation (blue-green deployment)
client.deployments.update_traffic(
    endpoint_name="my-endpoint",
    traffic_allocation={"blue": 90, "green": 10}
)

# Perform inference on endpoint
result = client.endpoints.infer(
    name="my-endpoint",
    inputs={"data": [[1, 2, 3, 4]]}
)

LLM Configuration & Inference

# Create LLM configuration for Azure OpenAI
client.llm.create(
    config_name="gpt4-prod",
    provider="azure_openai",
    model="gpt-4",
    api_key="your-api-key",
    endpoint="https://your-resource.openai.azure.com",
    deployment_name="gpt-4-deployment",
    default_parameters={
        "temperature": 0.7,
        "max_tokens": 500
    }
)

# Perform LLM inference
response = client.llm.infer(
    config_name="gpt4-prod",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is machine learning?"}
    ]
)
print(response.content)

# Stream LLM inference
for event in client.llm.infer_stream("gpt4-prod", messages):
    if event['type'] == 'chunk':
        print(event['content'], end='', flush=True)
    elif event['type'] == 'done':
        print(f"\nTokens used: {event['usage']}")

Messages & Conversations

# Create message with auto-inference
message = client.messages.create(
    endpoint_name="my-endpoint",
    text="Hello, I need help!",
    email="[e-mail]",
    llm_config_name="gpt4-prod",
    auto_infer=True
)

# List conversations with filters
conversations = client.conversations.list(
    endpoint_name="my-endpoint",
    is_escalated=False,
    is_resolved=False
)

# Escalate conversation
client.conversations.escalate(
    id=conversation_id,
    escalation_routes=["[e-mail]"]
)

# Resolve conversation
client.conversations.resolve(id=conversation_id)

Inference Management

# Perform direct inference
result = client.inferences.infer(
    endpoint_name="my-endpoint",
    inputs={"messages": [{"role": "user", "content": "Hello!"}]},
    tenant_key="customer-123"
)

# List and filter inferences
inferences = client.inferences.list(
    endpoint_name="my-endpoint",
    has_annotation=False,
    tenant_key="customer-123"
)

# Update inference with annotation
client.inferences.update(
    id=inference_id,
    annotation={"quality": "good", "category": "support"}
)

Secrets & Scoring Scripts

# Manage Azure Key Vault secrets
client.secrets.create(name="api-key", value="secret-value")
client.secrets.update(name="api-key", value="new-value")
secrets = client.secrets.list()

# Create scoring script from file
client.scorings.create_from_file(
    name="score.py",
    file_path="./scoring/score.py"
)

# List scoring scripts
scripts = client.scorings.list()

Test Data Management

from az_cortex_sdk import TestDataManager

test_manager = TestDataManager(client)

# Create test messages (automatically marked as test-generated)
test_message = test_manager.create_test_message(
    endpoint_name="my-endpoint",
    text="Test message",
    email="[e-mail]"
)

# List only test data
test_conversations = client.conversations.list_test_conversations()
test_messages = client.messages.list_test_messages()
test_inferences = client.inferences.list_test_inferences()

# Clean up test data
cleanup_stats = test_manager.cleanup_all_test_data()
print(f"Cleaned up {cleanup_stats['total_deleted']} test items")

API Reference

Core Services

The SDK provides access to all Azure Cortex ML API endpoints through service classes:

Environment Management

client.environments - Manage Azure ML environments

Method	Description	Parameters
list(page, per_page)	List all environments	page (int), per_page (int)
get(name)	Get environment by name	name (str)
create(name, image, conda_file, description)	Create new environment	name, image, conda_file, description
update(name, image, conda_file, description)	Update environment	name, image, conda_file, description
delete(name)	Delete environment	name (str)

client.environment_versions - Manage environment versions

Method	Description	Parameters
list(environment_name, page, per_page)	List versions for environment	environment_name, page, per_page
register(environment_name, version, image, conda_file, description)	Register new version	environment_name, version, image, conda_file, description
archive(environment_name, version)	Archive a version	environment_name, version
set_default(environment_name, version)	Set default version	environment_name, version

Model Management

client.models - Manage Azure ML models

Method	Description	Parameters
list(page, per_page)	List all models	page (int), per_page (int)
get(name)	Get model by name	name (str)
create(name, description, tags)	Create new model	name, description, tags
upload_and_create(file_path, name, description, tags)	Upload file and create model	file_path, name, description, tags
upload_buffer_and_create(buffer, filename, name, description, tags)	Upload buffer and create model	buffer, filename, name, description, tags
delete(name)	Delete model	name (str)

client.model_versions - Manage model versions

Method	Description	Parameters
list(model_name, page, per_page)	List versions for model	model_name, page, per_page
register_from_file(file_path, model_name, version, description, tags)	Register version from file	file_path, model_name, version, description, tags
register_from_buffer(buffer, filename, model_name, version, description, tags)	Register version from buffer	buffer, filename, model_name, version, description, tags
archive(model_name, version)	Archive a version	model_name, version
set_default(model_name, version)	Set default version	model_name, version

Endpoint & Deployment Management

client.endpoints - Manage online endpoints

Method	Description	Parameters
list(page, per_page)	List all endpoints	page (int), per_page (int)
get(name)	Get endpoint by name	name (str)
create(name)	Create new endpoint	name (str)
update(name, identity)	Update endpoint identity	name, identity (EndpointIdentity)
delete(name)	Delete endpoint	name (str)
infer(name, inputs)	Perform inference on endpoint	name, inputs (dict)

client.deployments - Manage endpoint deployments

Method	Description	Parameters
list(endpoint_name, page, per_page)	List deployments for endpoint	endpoint_name, page, per_page
get(name, endpoint_name)	Get deployment by name	name, endpoint_name
create(endpoint_name, name, model_name, model_version, environment_name, environment_version, instance_type, scoring_script_name, instance_count, traffic_percentage)	Create deployment	See deployment parameters
delete(name, endpoint_name)	Delete deployment	name, endpoint_name
update_traffic(endpoint_name, traffic_allocation)	Update traffic allocation	endpoint_name, traffic_allocation (dict)

Inference & Messaging

client.inferences - Manage ML inferences

Method	Description	Parameters
list(endpoint_name, page, per_page, has_annotation, is_test_generated, ...)	List inferences with filters	Multiple filter parameters
get(id)	Get inference by ID	id (str)
update(id, annotation, escalation_routes)	Update inference	id, annotation, escalation_routes
infer(endpoint_name, inputs, tenant_key, ip_address, is_test_generated)	Perform direct inference	endpoint_name, inputs, optional params
list_test_inferences(...)	List only test-generated inferences	Same as list()

client.messages - Manage conversation messages

Method	Description	Parameters
list(endpoint_name, conversation_id, page, per_page, is_test_generated, ...)	List messages with filters	Multiple filter parameters
get(id)	Get message by ID	id (str)
create(endpoint_name, text, email, tenant_key, llm_config_name, conversation_id, ip_address, auto_infer, is_test_generated)	Create message	endpoint_name, text, email, optional params
update(id, annotation, escalation_routes)	Update message	id, annotation, escalation_routes
list_test_messages(...)	List only test-generated messages	Same as list()

client.conversations - Manage conversations

Method	Description	Parameters
list(endpoint_name, page, per_page, is_escalated, is_resolved, is_test_generated, ...)	List conversations with filters	Multiple filter parameters
get(id)	Get conversation by ID	id (str)
update(id, is_escalated, is_resolved, annotation, escalation_routes)	Update conversation	id, optional params
escalate(id, escalation_routes)	Escalate conversation	id, escalation_routes
resolve(id)	Resolve conversation	id (str)
list_test_conversations(...)	List only test-generated conversations	Same as list()

LLM Configuration & Inference

client.llm - Configure and use LLM providers

Method	Description	Parameters
list(page, per_page, is_archived)	List LLM configurations	page, per_page, is_archived
get(config_name)	Get LLM config by name	config_name (str)
create(config_name, provider, model, api_key, endpoint, deployment_name, default_parameters)	Create LLM config	See LLM parameters
update(config_name, provider, model, api_key, endpoint, deployment_name, default_parameters)	Update LLM config	Same as create()
archive(config_name)	Archive LLM config	config_name (str)
delete(config_name)	Delete LLM config	config_name (str)
infer(config_name, messages, parameters, is_test_generated)	Perform LLM inference	config_name, messages, optional params
infer_stream(config_name, messages, parameters, is_test_generated)	Stream LLM inference	Same as infer()

Secrets & Scoring Scripts

client.secrets - Manage Azure Key Vault secrets

Method	Description	Parameters
list(page, per_page)	List all secrets	page (int), per_page (int)
create(name, value)	Create new secret	name, value
update(name, value)	Update secret value	name, value
delete(name)	Delete secret	name (str)

client.scorings - Manage scoring scripts

Method	Description	Parameters
list(page, per_page)	List all scoring scripts	page (int), per_page (int)
get(name)	Get scoring script by name	name (str)
create(name, content)	Create scoring script	name, content
create_from_file(name, file_path)	Create from file	name, file_path

Test Data Management

The SDK includes built-in support for test data management:

• Automatic Filtering: By default, all list() methods hide test-generated data
• Test Data Creation: Use TestDataManager to create test data marked with is_test_generated=True
• Convenience Methods: Use list_test_*() methods to view only test data
• Cleanup Utilities: Built-in methods to clean up test data after testing

Class	Description	Key Methods
TestDataManager	Test data utilities	create_test_message(), cleanup_all_test_data(), count_test_data()

Test Data Filtering Options

All list methods support the is_test_generated parameter:

• None (default): Hide test-generated objects
• 'false': Explicitly hide test-generated objects
• 'true': Show only test-generated objects
• 'all': Show all objects regardless of test generation status

Development

# Install development dependencies
pip install -e .[dev]

# Run unit tests only (no live API calls)
make test

# Run unit tests with coverage
make test-cov

# Run integration tests (requires live API access)
make test-integration

# Run all tests
make test-all

License

MIT License - see LICENSE file for details.