co-pilot SDK enables apps to manage and communicate with the co-pilot CLI

The speaker demos a newly stealth-launched technical preview of the co-pilot SDK (github.com/github/co-pilot SDK) that embeds the GitHub co-pilot agent and the co-pilot CLI into custom applications. The SDK is multi-platform (Node, Python, Go, and C# for .NET) and uses JSON RPC to talk to a co-pilot CLI running in server mode, exposing full session management, model selection across providers (Cloud models, GPT, Gemini), tool calling, streaming results, and events (for example session.on and an idle event to mark completion). Concrete examples: an interactive C# demo that selects GPT-5-2, shows a unique session ID, pings the agent, runs code-review and performance demos, and streams deltas back; a console app and a Blazor server web app (both using the same settings) that form a podcast metadata generator which loads a transcript and produces titles, descriptions, YouTube chapters, and SRTs in real time via the CLI. The speaker reports building "all three of these applications under one hour" and that it took an app into production in "the 20, 30 minutes." Operational details preserved from the demo: the CLI asks to trust files in a folder, supports slash commands and delegation to coding agents, lists models with the --model flag, and supports plan mode, cloud delegation, custom agents and subagents. The SDK packaging examples include a NuGet package named github.co-pilot.SDK and simple usage patterns: create a Co-pilot client, start a session, listen for events, send prompts, stream results, and set results when idle. The speaker highlights Microsoft Foundry as an alternate way they build AI, but positions the SDK as a way to directly leverage a user's co-pilot subscription and CLI to build integrated workflows rapidly. The conclusion: try the SDK, update the CLI, and provide feedback to the team.