Developers can now build and test AI agents using the Microsoft Agent Framework without incurring Azure costs by leveraging GitHub Models. The Microsoft Agent Framework, an open-source kit for Python and C#, unifies Semantic Kernel and AutoGen to create AI agents and complex agentic workflows. This tutorial demonstrates how to integrate GitHub Models—specifically the OpenAI gpt-5-mini model—as a free development alternative to Azure Foundry.

The process involves generating a GitHub Personal Access Token (PAT), configuring the OpenAI ChatClient to point to the 'https://models.github.ai/inference' endpoint, and implementing OpenTelemetry for observability. Key code examples show how to manage stateless agent threads, serialize conversation history to JSON for persistence, and resume multi-turn conversations. While GitHub Models offer a cost-effective testing environment, developers must be mindful of usage limits.

This approach significantly lowers the barrier to entry for prototyping LLM-based applications.