Let's Learn .NET: Let's Learn GitHub Copilot SDK
James Montemagno and Matt Soucoup walk through how the GitHub Copilot SDK for .NET lets you build agentic applications without implementing an agent loop yourself.
Overview
The session explains how the Copilot SDK differs from calling a model API directly: instead of building your own agent loop, you get programmatic access to Copilot’s agent runtime.
Key runtime capabilities called out:
- Conversation management
- Planning
- Tool orchestration
- Streaming events
- Session state
What the livestream covers
Core concepts
The presenters explore the main building blocks used when developing with the SDK:
- Clients
- Sessions
- Prompts
- The agent runtime
Building multi-turn, real-time experiences
The session covers how to:
- Create multi-turn conversations
- Stream responses in real time
Exposing tools and extending with MCP
The livestream discusses how to:
- Expose application functions as tools
- Connect external capabilities through MCP servers
Controlling agent behavior
The presenters describe mechanisms for controlling what the agent can do, including:
- Hooks
- Permission handlers
When to use Copilot SDK vs Copilot CLI
The session draws a clear line between the two:
- Use the Copilot CLI for developer tasks in the terminal.
- Use the Copilot SDK when you want to embed Copilot’s agentic capabilities inside your own application or service.
Prerequisites
No prior experience with the GitHub Copilot SDK is required.