kairoz 0.1.3

SDK for the Kairoz API

Kairoz Python SDK

The Kairoz SDK provides a simple interface to interact with the Kairoz API for managing and using text and chat prompts.

Installation

pip install kairoz

Quick Start

Python

from kairoz import Kairoz

# You can pass up to two arguments:
# Kairoz(api_key, base_url)
client = Kairoz(api_key="your-api-key")

Configuration

You can configure the SDK using the constructor or environment variables:

• api_key: Your Kairoz API Key (can use the KAIROZ_API_KEY environment variable)
• base_url: API base URL (optional, defaults to https://api.kairozai.com)

Example

# All arguments are optional except api_key
client = Kairoz(api_key="your-api-key", base_url="https://api.kairozai.com")

Or using environment variables:

export KAIROZ_API_KEY=your-api-key

Usage Guide

Getting a Prompt

prompt = client.get_prompt("prompt-name")
prompt_by_label = client.get_prompt("prompt-name", label="production")
prompt_by_version = client.get_prompt("prompt-name", version="2")

Note: If the prompt does not exist (404), the SDK raises an exception immediately and does not retry the request.

Creating a Prompt

Text Prompt

text_prompt = client.create_prompt(
    name="greeting",
    type="text",
    prompt="Hello {{name}}!",
    config={"temperature": 0.7},
    labels=["greeting"]
)

Chat Prompt

chat_prompt = client.create_prompt(
    name="chat-assistant",
    type="chat",
    prompt=[
        {"role": "system", "content": "You are a {{type}} assistant"},
        {"role": "user", "content": "Hello {{name}}!"},
        {"role": "assistant", "content": "How can I help you?"}
    ],
    config={"temperature": 0.8},
    labels=["chat", "assistant"]
)

Formatting Prompts

You can replace variables in prompts using the format method, which returns a new Prompt instance with the formatted content.

Text

from kairoz import Kairoz

kairoz = Kairoz("your-api-key")
prompt = kairoz.get_prompt("greeting")

formatted_prompt = prompt.format(name="John", service="Kairoz")
# formatted_prompt is a new Prompt instance with the formatted content
print(formatted_prompt.prompt)  # "Hello John! Welcome to Kairoz."

Chat

kairoz = Kairoz("your-api-key")
prompt = kairoz.get_prompt("chat-greeting")

formatted_prompt = prompt.format(type="friendly", name="John")
# formatted_prompt is a new Prompt instance with the formatted messages
print(formatted_prompt.prompt)
# [
#   {"role": "system", "content": "You are a friendly assistant"},
#   {"role": "user", "content": "Hello John!"},
#   {"role": "assistant", "content": "How can I help you?"}
# ]

Additional Methods

validate()

The validate method ensures that the prompt object is correctly structured and raises an exception if any validation fails.

Example:

from kairoz.prompt import Prompt

prompt = Prompt(
    name="example",
    type="text",
    prompt="Hello {{name}}!",
    config={"temperature": 0.7},
    labels=["example-label"]
)

try:
    prompt.validate()
    print("Prompt is valid")
except ValueError as error:
    print("Validation failed:", error)

to_dict()

The to_dict method converts a Prompt object into a plain dictionary that can be sent to the API.

Example:

prompt = Prompt(
    name="example",
    type="text",
    prompt="Hello {{name}}!",
    config={"temperature": 0.7},
    labels=["example-label"]
)

prompt_data = prompt.to_dict()
print(prompt_data)
# Output:
# {
#   "name": "example",
#   "type": "text",
#   "config": {"temperature": 0.7},
#   "messages": [{"role": "user", "content": "Hello {{name}}!"}],
#   "labels": ["example-label"]
# }

from_dict()

The from_dict method populates a Prompt object from a plain dictionary.

Example:

prompt_data = {
    "name": "example",
    "type": "text",
    "messages": [{"role": "user", "content": "Hello {{name}}!"}],
    "config": {"temperature": 0.7},
    "labels": ["example-label"]
}

prompt = Prompt.from_dict(prompt_data)
print(prompt)
# Output:
# <Prompt object with name="example", type="text", prompt="Hello {{name}}!">

Tracing

The tracing module allows you to monitor and log the execution flow of your application, capturing valuable information about the performance and behavior of your code. Here’s how to use it.

Initialization

To start using tracing, you must first initialize the tracker with your API key.

from kairoz.tracking import initialize_tracking

initialize_tracking(api_key="your-api-key")

Basic Usage

Tracing is implemented through context managers that allow you to wrap blocks of code to log traces, generations, and spans.

• Trace: Represents the complete execution flow of an operation.
• Generation: Used to log a generation from a language model (LLM).
• Span: A generic execution block that you can use to monitor any part of your code.

Complete Example

from kairoz.tracking import trace, initialize_tracking

initialize_tracking(api_key="your-api-key")

# Start a trace
with trace(
    name="example-trace",
    user_id="user-123",
    session_id="session-abc",
    metadata={"key": "value"},
    tags=["tag1", "tag2"]
) as t:
    # Log a generation within the trace
    with t.generation(
        name="example-generation",
        model="gpt-4",
        input_data={"prompt": "Hello world"},
        metadata={"provider": "openai"}
    ) as g:
        # Simulate an LLM call
        output = "This is a generated response."
        usage = {"input_tokens": 10, "output_tokens": 20}
        
        # Update the generation with the result
        g.update(output=output, usage=usage)

    # Log a span within the trace
    with t.span(
        name="example-span",
        input_data={"data": "some data"},
        metadata={"source": "internal"}
    ) as s:
        # Simulate an operation
        result = "Processed data"
        
        # Update the span with the result
        s.update(output=result)

Complete Examples

Create and Use a Text Prompt

from kairoz import Kairoz

client = Kairoz(api_key="your-api-key")

prompt = client.create_prompt(
    name="custom-greeting",
    type="text",
    prompt="Hello {{name}}! Welcome to {{service}}.",
    config={"temperature": 0.7},
    labels=["greetings"]
)

saved_prompt = client.get_prompt("custom-greeting")
formatted_prompt = saved_prompt.format(name="Mary", service="Kairoz")
print(formatted_prompt.prompt)  # "Hello Mary! Welcome to Kairoz."

Create and Use a Chat Prompt

from kairoz import Kairoz

client = Kairoz(api_key="your-api-key")

prompt = client.create_prompt(
    name="chat-assistant",
    type="chat",
    prompt=[
        {"role": "system", "content": "You are a {{type}} assistant"},
        {"role": "user", "content": "Hi! My name is {{name}}"},
        {"role": "assistant", "content": "How can I help you?"}
    ],
    config={"temperature": 0.8},
    labels=["chat", "assistant"]
)

saved_prompt = client.get_prompt("chat-assistant")
formatted_prompt = saved_prompt.format(type="friendly", name="Charles")
print(formatted_prompt.prompt)
# [
#   {"role": "system", "content": "You are a friendly assistant"},
#   {"role": "user", "content": "Hi! My name is Charles"},
#   {"role": "assistant", "content": "How can I help you?"}
# ]

Error Handling

• If the prompt does not exist (404), the SDK raises an exception and does not retry the request.
• If a text prompt is not a string, raises: "Text prompts must have a string prompt".
• If a chat prompt is not a list, raises: "Chat prompts must have a list of messages".
• If a message is missing role or content, raises: "Each message must have 'role' and 'content' fields".
• If a variable is missing during formatting, raises: "Variable '{name}' not found".

Additional Notes

• The SDK supports Python 3.7 and above.
• All main functions are synchronous and raise exceptions for errors.
• The SDK validates data before sending to the API and raises descriptive exceptions if something is wrong.

Using OpenAI with Provider Fallback

The Kairoz SDK also provides a seamless way to use OpenAI-compatible APIs with automatic fallback to other providers.

Basic Setup

from kairoz import KairozOpenAI, Providers

# Configure with primary and fallback providers
openai = KairozOpenAI(
    primary=Providers.openai("your-openai-api-key"),
    fallback=Providers.anthropic("your-anthropic-api-key"),
    enable_logging=True,
)

Chat Completions with Fallback

async def example_usage():
    try:
        # Chat completions with automatic fallback
        chat_response = await openai.chat.completions.create(
            model="gpt-4",
            model_fallback="claude-3-5-sonnet-20240620",
            messages=[{"role": "user", "content": "Hello, how are you?"}],
        )
        print("Chat response:", chat_response.choices[0].message.content)

    except Exception as error:
        print("Error:", error)

# Run the example
if __name__ == "__main__":
    import asyncio
    asyncio.run(example_usage())

Available Providers

The SDK supports multiple providers that can be used as primary or fallback:

• OpenAI: Providers.openai(api_key)
• Anthropic: Providers.anthropic(api_key)
• Groq: Providers.groq(api_key)
• Together.ai: Providers.together(api_key)
• Azure OpenAI: Providers.azure_openai(api_key, endpoint)
• LM Studio: Providers.lm_studio(api_key, base_url)

Advanced Configuration Examples

# OpenAI + Groq
openai_with_groq = KairozOpenAI(
    primary=Providers.openai(process.env.OPENAI_API_KEY),
    fallback=Providers.groq(process.env.GROQ_API_KEY),
    enable_logging=True,
)

# Azure OpenAI + OpenAI
azure_with_openai = KairozOpenAI(
    primary=Providers.azure_openai(process.env.AZURE_API_KEY, process.env.AZURE_ENDPOINT),
    fallback=Providers.openai(process.env.OPENAI_API_KEY),
    enable_logging=True,
)

# Local LM Studio + OpenAI
local_with_openai = KairozOpenAI(
    primary=Providers.lm_studio('dummy-key', 'http://localhost:1234/v1'),
    fallback=Providers.openai(process.env.OPENAI_API_KEY),
    enable_logging=True,
)