AI agents

Public samples that show how I build and explore multi-agent systems.

GitHubCopilotSDK

GitHub Copilot SDK - same session API in C#, TS, Python, and Go

GitHubCopilotSDK is a multi-language learning repo that runs the same Copilot SDK session flow in C# (.NET 10), TypeScript, Python, and Go - all on SDK 1.0.14. Each CLI creates a session (model auto, approve-all permissions), sends a simple prompt, and prints the assistant reply. The Go sample highlights CLI-path discovery because that SDK does not bundle a runtime; Python’s runner bootstraps a venv and avoids the Windows Store stub. Aimed at engineers comparing Copilot SDK ergonomics across languages for agent or CLI tooling.

artefact-driven-sdlc

Lean artefact-driven SDLC - agent cohorts + Next.js UI

artefact-driven-sdlc is a standalone .NET 10 + Next.js product for artefact-driven agentic SDLC. Specialised agent cohorts mature work along a deterministic graph - refined idea to story to module-spec to UI design to wireframe to initial assessment - using a shared producer/reviewer skill model. A Spectre.Console CLI and AAC.Api drive the same Azure AI Foundry agent path; a chat-style Next.js UI lets you pick a cohort, point at a feature folder, and run single or pipeline modes. Built as a focused app slice (Aspire portal and brownfield assessment paths intentionally out of scope).

AgentCohortHostingPlatformDemo

Orleans multi-silo agent-cohort hosting on Aspire

AgentCohortHostingPlatformDemo is an Aspire + Microsoft Orleans sample of an Agent Cohort hosting platform: artefact state lives in grains, not the HTTP layer. It runs a 3-replica silo with Azure Table clustering and Azure Blob grain persistence (Azurite locally), an Orleans client API over grain contracts, and a React UI to load artefacts, append prompts, and deactivate grains. Persistent state, grain lifecycle hooks, custom DTO serialization, and the Orleans Dashboard are wired in - aimed at engineers building distributed AI/agent platforms on .NET.

AfSessionExperiment

Multi-agent hand-off with MAF session StateBag

AfSessionExperiment is a Microsoft Agent Framework (MAF) console sample where Producer and Reviewer agents share work in one session. The Producer stores a test-plan artefact in session StateBag; the Reviewer gets that content injected as context before it runs, then stores its review the same way. It talks to Gemma hosted on Docker Desktop through an OpenAI-compatible endpoint - handy for engineers exploring multi-agent collaboration without stuffing everything into chat history.

Related videos

Artefact-driven SDLC with Microsoft Agent Framework: Part 1, Part 2, and Part 3 - or watch on the Videos page.