Back to courses
Copilot SDKPythonTypeScriptC#MCP

Build Always-On AI Agents with the Copilot SDK

Go from zero to production 24/7 agents on Azure. Deep dive into the GitHub Copilot SDK across Python, TypeScript, and C# — CopilotClient, JSON-RPC, custom tools with Zod, Pydantic, and POCO handlers. Connect MCP servers, build skills directories, and configure infinite sessions with auto-compaction. Set up providers for Azure AI Foundry, OpenAI, or Anthropic via BYOK. Design custom agents with restricted tool sets, multi-agent routing, and permission policies. Add audit logging with secret redaction, test rigorously, then ship to Azure Container Apps with OpenTelemetry and CI/CD.

Duration 5 weeks
Level Intermediate
lessons 20
Total Hours 50+
Instructor Rafael Kemish Microsoft Certified Trainer · Senior Technical Consultant

Curriculum

The CopilotClient — JSON-RPC architecture & streaming across all three languages
Built-in tools, tool control (availableTools/excludedTools) & the permission system
Providers & BYOK: Azure AI Foundry, OpenAI & Anthropic via ProviderConfig
System messages (append/replace/customize modes), reasoning effort & your first agent
Hooks deep dive: onPreToolUse, onPostToolUse, onSessionStart & onErrorOccurred
MCP servers: local, HTTP & SSE transports with tool filtering
Custom tools: defineTool with Zod (TS), Pydantic (Python) & POCO (C#) handlers
Sessions, infinite context (auto-compaction) & skills directories
Custom agents: CustomAgentConfig with restricted tool sets & per-agent MCP
Multi-agent routing: delegation, agent selection & session handoff
Error handling: onErrorOccurred hooks, retry/skip/abort strategies & graceful degradation
Event streaming: message deltas, tool execution tracking & health monitoring
Permission policies: building approval rules beyond approveAll
Audit logging & compliance: hooks for tool call recording & secret redaction
Testing agents: mocking LLMs, evaluating outputs & integration tests
Queue-driven & event-triggered agents on Azure Service Bus
Containerizing agents & deploying to Azure Container Apps
Secrets management & CI/CD with GitHub Actions
Observability with OpenTelemetry (TelemetryConfig), cost controls & security
Capstone: Deploy a 24/7 multi-agent system on Azure

What You'll Build

  • Custom tools with Zod (TS), Pydantic (Python) & C# handlers — with permission policies
  • A multi-agent system using customAgents with restricted tool sets and routing
  • Audit logging and compliance hooks with secret redaction
  • A tested, monitored production deployment on Azure Container Apps with OpenTelemetry

Prerequisites

  • Python, TypeScript, or C# proficiency
  • A GitHub Copilot subscription (or BYOK API key for OpenAI/Azure/Anthropic) and an Azure subscription
  • Basic familiarity with Docker and cloud deployments

Frequently Asked Questions

Yes. After purchase you'll receive a detailed invoice via email. For custom invoicing or purchase orders, reach out via the contact form on our Teams page.

Absolutely. We offer up to 20% off for teams of 2 or more, with custom pricing for 10+. Visit our Teams page and fill out the contact form for a quote.

Full refund up to 7 days before the first session, no questions asked. After session 1, reach out — we'll find a fair solution.

Yes. Upon completing all sessions, you'll receive a verifiable certificate of completion that you can share on LinkedIn or with your employer.

Yes. All live sessions are recorded and available for 6 months after the cohort ends. Watch them at your own pace to review material.

No problem. Recordings are available within 24 hours. You can also ask questions asynchronously in the private community channel.

Each course lists specific prerequisites on this page. In general, you need comfortable programming skills and basic familiarity with APIs and cloud services.

About Your Instructor

Rafael Kemish, Instructor

Rafael Kemish is a Microsoft Certified Trainer with 25 years in the Microsoft ecosystem and 23 active certifications. As a Senior Technical Consultant at Microsoft, he has trained over 5,000 professionals at organizations including NASA, Microsoft, Honda, and Citrix. He specializes in Azure AI, OpenAI integration, and enterprise platform engineering — teaching what actually works in real-world production deployments.