llm-provider-configuration

name	llm-provider-configuration
description	Skill for configuring OpenAI Agents SDK to work with alternative LLM providers using base URL and API key

LLM Provider Configuration Skill

This skill provides guidance for configuring the OpenAI Agents SDK to work with alternative LLM providers (like Google Gemini, Anthropic Claude, etc.) using a custom base URL and API key.

Key Concepts

The OpenAI Agents SDK can work with alternative LLM providers that support the OpenAI API format by configuring a custom base URL and API key. This is done through one of three levels:

1. Agent Level (recommended)
2. Run Level
3. Global Level

Agent Level Configuration (Recommended)

For per-agent configuration, use the OpenAIChatCompletionsModel with a custom client:

import asyncio
from openai import AsyncOpenAI
from agents import Agent, OpenAIChatCompletionsModel, Runner, set_tracing_disabled

# Initialize client with custom base URL and API key
client = AsyncOpenAI(
    api_key=your_api_key,
    base_url="https://your-provider-base-url/",
)

set_tracing_disabled(disabled=True)

# Create agent with custom model configuration
agent = Agent(
    name="Assistant",
    instructions="You are a helpful assistant.",
    model=OpenAIChatCompletionsModel(model="model-name", openai_client=client),
)

Required Parameters

1. Base URL: The API endpoint for the LLM provider (e.g., "https://generativelanguage.googleapis.com/v1beta/openai/" for Gemini)
2. API Key: The authentication key for the LLM provider
3. Model Name: The specific model identifier (e.g., "gemini-2.0-flash")

Gemini-Specific Configuration

For Google Gemini integration:

• Base URL: https://generativelanguage.googleapis.com/v1beta/openai/
• API Key: Google API key with Gemini access
• Model: gemini-2.0-flash, gemini-2.5-flash, etc.

Common Provider URLs

• Google Gemini: https://generativelanguage.googleapis.com/v1beta/openai/
• Anthropic Claude (if OpenAI compatible): https://api.anthropic.com/v1/
• Other providers: Check documentation for OpenAI-compatible endpoints