Implement context management strategies that keep agents coherent across long, multi-turn sessions. This course addresses two distinct challenges: managing conversation history within the model's context window as sessions grow long, and maintaining structured application state that persists meaningful information beyond what conversation history alone can capture. You'll learn the boundary between what the Software Development Kit (SDK) handles automatically (message history within a session) and what your application must manage (summarization, structured state, and cross-session persistence).

Implement real-time streaming to progressively render agent responses and tool execution events in a user interface. Agent responses that arrive as a single block after a multi-second wait feel unresponsive; the same response streamed token by token, with tool execution status surfaced as it happens, feels immediate and trustworthy. Learn to subscribe to Software Development Kit (SDK) event streams and handle different event types: text chunks, tool invocations, execution status, and errors. Build rendering logic that keeps users informed during long agent operations, including graceful handling of mid-stream failures and user cancellation.

Define custom agents with specialized system prompts, restricted tool access, and behavioral constraints that the Copilot runtime can delegate to during task execution. Learn to design agents as modular, reusable components, each with a scoped identity and capability set.

Build a multi-turn conversational agent that combines session lifecycle management, context persistence, streaming responses, and custom agent delegation into a cohesive application. This project integrates the session, context, streaming, and custom-agent skills developed across this long course into a portfolio-ready deliverable.