liev 0.1.2

Liev LLM Dispatcher client

Liev Client Library

A powerful Python client for interacting with the Liev AI API platform, providing seamless access to advanced language and multimodal AI capabilities.

---

Installation

pip install liev

Features

• Multiple AI Model Access: Connect to GPT-4o, Claude, and more through a unified API
• Multi-modal Capabilities: Text completion, image generation, and vision tasks
• Streaming Responses: Get real-time responses with token streaming
• Failover Mechanism: Automatic fallback to alternative models
• OpenAI-Compatible Interface: Use familiar OpenAI SDK patterns

Quick Start

from liev import Liev

# Initialize the client with your API token
liev = Liev(api_token='your-api-token')

# Simple text completion
response = liev.ask('What is the capital of Bulgaria?')
print(response)

# Generate an image
from PIL import Image
img_bytes = liev.ask('Generate a photo of a horse running', function='image')
if img_bytes:
    Image.open(img_bytes).save('horse.png')
Core Methods
Initialization

liev = Liev(
    api_token='your-api-token',  # Required
    base_url='https://another.liev.ai/'  # Optional, defaults to Liev api endpoint
)

Model Information

List available models

models = liev.get_llms()

Get models and their types

models_and_types = liev.get_llms_and_types()

Optionally filter by type

text_models = liev.get_llms_and_types(type='text')

Asking Questions

Basic text response with automatic prompt detection

response = liev.ask('How many moons does Saturn have?')

Explicitly specify function type and model

response = liev.ask(
    'Explain quantum computing',
    function='text',  # Options: 'text', 'code', 'image', 'embeddings', etc.
    llm_name='gpt4o',  # Specific model to use
    try_next_on_failure=True  # Automatic failover if specified model fails
)

Using the OpenAI-compatible format with messages

messages = [
    {"role": "system", "content": "You are a physics teacher."},
    {"role": "user", "content": "Why do objects fall?"}
]
response = liev.ask(messages=messages, function="text")

# Get results from all available models of a specific type
response = liev.ask('What is the meaning of life?', function='text', llm_name='all')
Streaming Responses

# Stream responses for real-time output
for chunk in liev.ask_stream('Tell the story of Mowgli'):
    print(chunk, end="")

Vision Capabilities

import base64
import requests

# Load an image and encode it to base64
image_url = "https://example.com/dog.jpg"
base64_image = base64.b64encode(requests.get(image_url).content).decode()

# Create messages with image and text
messages = [{
    "role": "user",
    "content": [
        {
            "type": "image",
            "source": {
                "type": "base64",
                "media_type": "image/jpg",
                "data": base64_image
            }
        },             
        {
            "type": "text",
            "text": "What is in this image?"
        }
    ]
}]

Send the request

response = liev.ask(messages=messages, function="text", llm_name='claude37')
print(response)

OpenAI-Compatible Interface

# Get an OpenAI-compatible client
openai = liev.get_openai_client()

# Use familiar OpenAI patterns
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "What is the capital of Bulgaria?"}
]

response = openai.chat.completions.create(
    model="gpt4o", 
    messages=messages,
    max_tokens=400,
    extra_headers={
        "x-liev-try-next-on-failure": "false",
        "x-liev-type-failover": "code"
    }
)


print(response.choices[0].message.content)

Usage Information and Metadata

You can access information about the last API call made:

liev.ask('What is the capital of France?')
call_info = liev.get_last_call_info()
print(f"Model used: {call_info['model']}")
print(f"Consumption info: {call_info['consumption_info']}")
print(f"Failover response: {call_info['is_failover_response']}")
print(f"Failed models: {call_info['response_failed_models']}")

Advanced Configuration

The ask method supports several parameters for fine-tuning your requests:

response = liev.ask(
    instruction="Explain the concept of quantum entanglement",
    input=None,  # Additional input data (instruction is ignored if this is provided)
    messages=None,  # List of message objects (instruction is ignored if this is provided)
    function=None,  # Specific function type to call (text, code, image, etc.)
    system_msg="You are a quantum physics expert",  # System message for context
    max_new_tokens=1000,  # Maximum tokens to generate
    temperature=0.1,  # Sampling temperature (0.0 = deterministic, higher = more random)
    timeout=120,  # Request timeout in seconds
    llm_name=None,  # Specific model to use
    try_next_on_failure=True  # Whether to try alternative models on failure
)

Error Handling

The library throws exceptions with detailed messages for various error conditions:

Authentication errors (401) Access denied (403) Endpoint not found (404) Connection errors Timeout errors JSON decode errors Other HTTP and request errors

Example:

try:
    response = liev.ask('What is quantum computing?')
    print(response)
except Exception as e:
    print(f"Error: {e}")

Limitations

The OpenAI-compatible interface is in alpha preview and doesn't have all functionalities Some models may have specific restrictions or requirements API rate limits may apply depending on your account tier