autourgos-llmkit 1.1.2

Unified LLM provider interface for Google Gemini, Anthropic Claude, OpenAI, Grok, Azure OpenAI, Ollama, and Vertex AI.

Autourgos LLM Kit

The lightweight, unified interface for state-of-the-art LLMs.
Connect to OpenAI, Gemini, Claude, Grok, Azure, Vertex AI, and Ollama with a single, consistent API.

Installation · Quick Start · Providers · Documentation

---

💡Why Autourgos LLM Kit?

Building AI agents requires flexibility. Today you might use GPT-4o for reasoning, tomorrow Gemini 3 for long context, and DeepSeek R1 for cost-efficiency. Switching providers usually means rewriting client code, handling different error types, and managing disparate retry logic.

Autourgos LLM Kit solves this without the bloat. Unlike heavy frameworks (LangChain, LlamaIndex) that enforce specific cognitive architectures, this library focuses solely on the connectivity layer.

• Unified Interface: One invoke() signature for all providers.
• Zero Lock-in: Switch from OpenAI to Vertex AI to Ollama just by changing a config string.
• Production Hardened: Built-in exponential backoff, retries, and standardized exception handling.
• Type Safe: Fully typed for modern Python development.
• Lightweight: Modular dependencies—install only the SDKs you need.

---

📦 Installation

Install the core package (lightweight, no heavy dependencies):

pip install autourgos-llmkit

Install provider-specific dependencies as needed to keep your build slim:

# Individual providers
pip install autourgos-llmkit[google]      # For Gemini
pip install autourgos-llmkit[openai]      # For GPT, Grok, Azure
pip install autourgos-llmkit[anthropic]   # For Claude
pip install autourgos-llmkit[ollama]      # For Ollama

# Or install everything
pip install autourgos-llmkit[all]

Note: Vertex AI support uses the Python standard library and requires no extra pip packages, just the Google Cloud CLI.

---

🚀 Quick Start

1. The Unified init_llm Factory

The easiest way to get started is using the factory function. This allows you to drive your model selection purely via configuration.

import os
from autourgos.llmkit import init_llm

# 1. Setup API Keys (or use .env file)
os.environ["OPENAI_API_KEY"] = "sk-..."
os.environ["GOOGLE_API_KEY"] = "AIza..."

# 2. Initialize a provider
# Change 'provider' to 'google', 'anthropic', 'grok', 'ollama', etc.
llm = init_llm(
    provider="openai", 
    model="gpt-4o", 
    temperature=0.7
)

# 3. Generate text
response = llm.invoke("Explain the concept of 'Agentic Workflow' in one sentence.")
print(f"Response: {response}")

# 4. Stream text (Real-time)
print("Streaming: ", end="")
for chunk in llm.stream("List 3 benefits of Python."):
    print(chunk, end="", flush=True)
print()

---

🔌 Supported Providers

Provider	Key Class	Env Variable	Capabilities
OpenAI	OpenAILLM	OPENAI_API_KEY	Frontier models (gpt-4o, o1), Function Calling, JSON mode.
Google Gemini	GoogleLLM	GOOGLE_API_KEY	Massive context (2M+ tokens), Native Multimodal, gemini-1.5-pro.
Google Vision	GoogleVisionLLM	GOOGLE_API_KEY	Specialized image analysis and reasoning.
Anthropic	AnthropicLLM	ANTHROPIC_API_KEY	High reliability, claude-3-5-sonnet.
xAI Grok	GrokLLM	XAI_API_KEY	grok-3, grok-2. Access via OpenAI-compatible endpoint.
Azure OpenAI	AzureLLM	AZURE_OPENAI_API_KEY	Enterprise gpt-4o deployments.
Azure Foundry	AzureLLM	AZURE_OPENAI_API_KEY	Serverless MaaS: deepseek-r1, phi-4, llama-3.3.
Vertex AI	VertexAILLM	VERTEX_AI_ACCESS_TOKEN	Google Cloud infrastructure. Supports Gemini + Model Garden (Mistral, Llama).
Ollama	OllamaCloudLLM	OLLAMA_API_KEY	Local or Cloud open-source models (llama3, mistral).

---

🛠️ Advanced Usage

Multimodal (Vision) Requests

Process images effortlessly using the Vision-capable providers (Google, OpenAI, Anthropic).

from autourgos.llmkit.GoogleVision import GoogleVisionLLM

llm = GoogleVisionLLM(model="gemini-1.5-flash")

# Pass local file paths or PIL Image objects
response = llm.invoke(
    prompt="Extract all the text visible in this image.",
    images=["./receipt.jpg"]
)
print(response)

Direct Class Instantiation (Type-Safe)

For strict typing and IDE autocompletion, instantiate provider classes directly.

from autourgos.llmkit.Microsoft import AzureLLM

# Connect to a DeepSeek R1 endpoint on Azure Foundry
llm = AzureLLM(
    deployment_name="deepseek-r1",
    azure_endpoint="https://DeepSeek-R1-xyzw.eastus.models.ai.azure.com/",
    api_key="your-azure-key"
)

text = llm.invoke("Solve this complex logic puzzle.")

Standardized Error Handling

Don't catch 5 different RateLimitError exceptions. Autourgos unifies them.

from autourgos.llmkit.Openai import OpenAILLM, OpenAILLMAPIError, OpenAILLMAuthError

try:
    llm = OpenAILLM(model="gpt-4o", api_key="invalid-key")
    llm.invoke("Hello")
except OpenAILLMAuthError:
    print("Please check your API credentials.")
except OpenAILLMAPIError as e:
    print(f"Provider returned an error: {e}")

Vertex AI (No Pip Dependencies)

Vertex AI is unique because it often requires complex auth libraries. We implemented it using raw urllib and gcloud CLI integration, so you can run it in restricted environments without installing the heavy google-cloud-aiplatform SDK.

# Just authenticate via CLI
gcloud auth login

from autourgos.llmkit.Vertexai import VertexAILLM

# Automatically picks up credentials from gcloud
llm = VertexAILLM(
    model="google/gemini-2.0-flash",
    project_id="your-gcp-project-id",
    region="us-central1"
)

---

_{Built with ❤️ by the Autourgos Team}