all-the-llms: A Universal LLM Interface

A unified interface for querying Large Language Models (LLMs) across multiple providers using LiteLLM and OpenRouter. This package provides intelligent model routing that automatically selects the best provider for each model request.

Installation

Install from PyPI:

pip install all-the-llms

The package is also available on PyPI.

What is LiteLLM?

LiteLLM is a Python library that provides a unified interface to call multiple LLM APIs with a consistent OpenAI-like API. It supports 100+ LLM providers including:

• OpenAI (GPT-4, GPT-3.5, etc.)
• Anthropic (Claude models)
• Azure OpenAI
• Google (Gemini, PaLM)
• OpenRouter (aggregator for multiple models)
• And many more...

LiteLLM handles provider-specific differences, retries, rate limiting, and error handling, allowing you to switch between providers with minimal code changes.

What is OpenRouter?

OpenRouter is a unified API that provides access to 100+ LLM models from various providers through a single interface. It's particularly useful when:

• You don't have direct API keys for specific providers
• You want to access models not available through your direct provider accounts
• You need a fallback option when your primary provider is unavailable
• You want to compare models across different providers

OpenRouter requires credits (free tier available) and routes requests to the appropriate provider on your behalf.

Architecture

This package uses a three-tier routing system:

1. Azure (highest priority): Direct Azure OpenAI deployments
2. Provider (medium priority): Direct API access (OpenAI, Anthropic, etc.)
3. OpenRouter (fallback): Unified API for models not available through other routes

The system prioritizes routes in this order: Azure → Provider → OpenRouter.

The routing is handled by a "routing judge" - an LLM that intelligently selects the best route based on:

• Model name matching (semantic and exact)
• Available API keys
• Model availability in each catalog

Quick Start

1. Install

pip install all-the-llms

The package will automatically install all required dependencies (litellm, openrouter, python-dotenv, pydantic).

2. Set Up Environment Variables

Create a .env file in your project root. At minimum, you need OPENROUTER_API_KEY:

Required

• OPENROUTER_API_KEY: Get your key from OpenRouter

Optional (Direct Provider Access)

Only use if you have free credits you prefer over OpenRouter. Otherwise, prefer OpenRouter.

• OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, or any {PROVIDER}_API_KEY

Optional (Azure)

• AZURE_API_KEY: Azure OpenAI API key
• AZURE_API_BASE: Endpoint URL
• AZURE_API_VERSION: API version
• AZURE_API_MODELS: Comma-separated list of deployed models (e.g., "gpt-5,gpt-4.1,gpt-4.1-mini")

3. Use

from all_the_llms import LLM

llm = LLM("gpt-4o")
response = llm.completion([{"role": "user", "content": "Hello!"}])
print(response.choices[0].message.content)

4. Example .env File

# OpenRouter (required for routing judge and fallback models)
OPENROUTER_API_KEY=...

# Direct provider access (optional - only use if you have free credits available 
# that you prefer over OpenRouter. Prefer OpenRouter otherwise.)
OPENAI_API_KEY=...
ANTHROPIC_API_KEY=...

# Azure (specified for Harvard Medical School)
AZURE_API_KEY=...
AZURE_API_BASE="https://azure-ai.hms.edu"
AZURE_API_VERSION="2024-10-21"
AZURE_API_MODELS="gpt-5,gpt-5-mini,gpt-4.1,gpt-4.1-mini,gpt-4.1-nano"

Usage

Basic Example

from all_the_llms import LLM

# Initialize an LLM - routing happens automatically
llm = LLM("gpt-4o")

# Make a completion request
response = llm.completion(
    messages=[{"role": "user", "content": "Hello, how are you?"}],
    temperature=0.7,
    max_tokens=100
)

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

Note: All LiteLLM features (streaming, tools, structured output, etc.) are supported through the completion() method's **kwargs. See the LiteLLM documentation for a complete list of available parameters.

Advanced Example with Pydantic Structured Output

from all_the_llms import LLM
from pydantic import BaseModel
from enum import Enum

# Define a Pydantic model for structured output
class CoffeeQuality(str, Enum):
    excellent = "excellent"
    terrible = "terrible"
    meh = "meh"

class CoffeeReview(BaseModel):
    quality: CoffeeQuality
    caffeine_level: int  # 1-10 scale
    complaints: list[str]
    verdict: str

llm = LLM("claude-sonnet-4.5")

# Make a completion request with structured output using response_format
response = llm.completion(
    messages=[{"role": "user", "content": "Review this coffee: 'It tastes like someone dissolved a tire in hot water and called it a day.'"}],
    response_format=CoffeeReview,
    temperature=0.3,
)

# Extract and validate the structured response
content = response.choices[0].message.content
review = CoffeeReview.model_validate_json(content)
print(f"Quality: {review.quality}")
print(f"Caffeine Level: {review.caffeine_level}/10")
print(f"Complaints: {', '.join(review.complaints)}")
print(f"Verdict: {review.verdict}")

Custom Routing Judge

By default, the system uses openrouter/openai/gpt-4o-mini as the routing judge (free as long as you have an OpenRouter API key). This default can be customized by passing a different model to the routing_judge parameter:

llm = LLM("gpt-5-2025-11-16", routing_judge="azure/gpt-4.1-mini")

Development

For development, you can install the package in editable mode:

git clone https://github.com/payalchandak/all-the-llms.git
cd all-the-llms
pip install -e .

This allows you to make changes to the source code that take effect immediately without reinstalling.

What the Code Does

LLM Class

The LLM class is a thin wrapper that:

1. Resolves the model: Takes a user-friendly model name (e.g., "gpt-5-2025-11-16") and resolves it to a concrete provider-specific model ID (e.g., "azure/gpt-5" or "openrouter/openai/gpt-3.5-turbo")

2. Tests the connection: On initialization, sends a test request to verify the model is accessible and working

3. Exposes a simple API: Provides a completion() method that wraps litellm.completion() with the resolved model

ModelRouter Class

The ModelRouter class handles intelligent routing:

1. Loads model catalogs:

   • Azure models from AZURE_API_MODELS environment variable
   • Provider models from LiteLLM's catalog (based on available API keys)
   • OpenRouter models by querying the OpenRouter API

2. Exact matching: First tries to find exact matches in the catalogs (prioritizing Azure → Provider → OpenRouter). Model names are normalized (lowercase, whitespace removed) for matching.

3. LLM-based routing: If no exact match, uses a "routing judge" LLM to decide which route to use

4. Model resolution: Uses the routing judge again to map the requested model name to a specific model in the selected route

5. Fallback handling: If the selected route has no available models, falls back to other routes

Example Behavior

When you initialize an LLM, you'll see output like this:

Routing model gpt-5-2025-11-16 to valid LLM...
Selected route azure because the requested model 'gpt-5-2025-11-16' semantically matches the azure model 'gpt-5'.
Resolved gpt-5-2025-11-16 to azure/gpt-5
Testing LLM at azure/gpt-5
Successfully recieved response from gpt-5-2025-08-07

Routing Examples

Azure Route (when model matches Azure deployment):

Routing model gpt-5-2025-11-16 to valid LLM...
Selected route azure because the requested model 'gpt-5-2025-11-16' semantically matches the azure model 'gpt-5'.
Resolved gpt-5-2025-11-16 to azure/gpt-5

Provider Route (when direct API key is available):

Routing model claude-sonnet-4.5 to valid LLM...
Selected route provider because the requested model 'claude-sonnet-4.5' matches a model available from the provider with a direct api key.
Resolved claude-sonnet-4.5 to claude-sonnet-4-5

OpenRouter Route (fallback when no direct access):

Routing model gpt-3.5-turbo to valid LLM...
Selected route openrouter because requested model 'gpt-3.5-turbo' does not match any available azure models and there are no applicable provider options.
Resolved gpt-3.5-turbo to openrouter/openai/gpt-3.5-turbo

Error Handling

The system validates each model on initialization. If a model fails, you'll see an error:

RuntimeError: Could not get a valid response from openrouter/deepseek/deepseek-r1-0528. 
litellm.APIError: APIError: OpenrouterException - {"error":{"message":"This request requires more credits, 
or fewer max_tokens. You requested up to 7168 tokens, but can only afford 4706..."}}

Common issues:

• Insufficient credits: OpenRouter account needs more credits
• Invalid API key: Check your environment variables
• Model unavailable: The requested model may not be available on the selected route
• Rate limiting: Provider may be rate limiting requests