- Owner: Workshop PM Team
- Last updated: June 17, 2026
- Status: 🔄 In progress (uncommitted changes on main)
- Repository: microsoft-foundry/Foundry_Toolkit_for_VSCode_Lab
This update restructures the Lab01 (single-agent) documentation to make the workshop easier to follow, more scenario-driven, and accessible to attendees without an Azure subscription. No agent source code, Dockerfiles, or deployment configs are changed — this is a documentation-only update. Five themes drive the changes
- Extension version alignment ✅ Done
- Scenario-based testing ✅ Done
- Simplified lab flow ✅ Done
- New summary module ✅ Done
- Alternate path for MVPs / no-subscription users ✅ Done
Detailed Changes
The Foundry Toolkit extension was previously called "AI Toolkit" and has gone through several version bumps. The docs now reflect the v1.2.1 GA extension experience.
| Area | Before (committed) | After (working tree) |
|---|---|---|
| Extension references | Mix of "AI Toolkit" and "Foundry Toolkit" labels | Consistently uses Foundry Toolkit throughout |
| Scaffold wizard steps | Older order (template → model → name) | Updated to match v1.2.1 flow: Language → Response API → Basic-Agent Framework → model → workspace → name |
| Model catalog instructions | Azure-only model deployment | Covers Azure model catalog, Foundry Local catalog, and GitHub Models catalog via Foundry Toolkit UI |
Modules 04 (test locally) and 06 (verify in playground) now use different, complementary question sets instead of repeating the same prompts.
| Module | Purpose | Scenarios |
|---|---|---|
| 04 — Test Locally | Happy-path functional validation | 3 scenarios: IT incident (API latency spike), data pipeline (ETL failure), security (exposed credential) |
| 06 — Verify in Playground | Edge-case & safety boundary testing | 4 scenarios: ambiguous input (off-topic), prompt injection (system prompt extraction), minimal input (single word), adversarial multi-turn (role override) |
Each scenario includes input text, expected behavior with ✅/❌ criteria, and a validation rubric so attendees can self-assess.
The module count drops from 9 → 8 and the early modules are consolidated so attendees aren't bouncing between tools. File structure comparison
| # | Before (committed) | After (working tree) | What changed |
|---|---|---|---|
| 00 | 00-prerequisites.md — flat checklist of tools + RBAC table | 00-prerequisites.md — renamed "Introduction": opens with agent scenario, Mermaid architecture diagram, path toggle | Scenario-first framing; path choice upfront |
| 01 | 01-install-foundry-toolkit.md — extension install only | 01-setup.md — combined module: extension install + project creation + model deployment + RBAC | Merged old 01 + 02 into one module |
| 02 | 02-create-foundry-project.md — project + model deploy | 02-create-hosted-agent.md — scaffold wizard (was old 03) | Old 02 absorbed into new 01 |
| 03 | 03-create-hosted-agent.md — scaffold wizard | 03-configure-and-code.md — env vars, instructions, tools, venv (was old 04) | Renumbered |
| 04 | 04-configure-and-code.md — env vars + instructions | 04-test-locally.md — local testing with 3 scenario tests (was old 05) | New scenario-based test content |
| 05 | 05-test-locally.md — basic local test | 05-deploy-to-foundry.md — deployment (was old 06) | Renumbered |
| 06 | 06-deploy-to-foundry.md — deployment | 06-verify-in-playground.md — 4 edge-case/safety tests (was old 07) | New safety-focused test content |
| 07 | 07-verify-in-playground.md — playground verify | 07-summary.md — NEW: what you learned, next steps, cleanup | Entirely new module |
| 08 | 08-troubleshooting.md — troubleshooting reference | troubleshooting.md — moved out of numbered sequence | Now a reference appendix, not a numbered step |
Key simplification: The old flow required attendees to install the extension (01), then create a project (02) in a separate module, then come back and deploy a model. The new 01-setup.md walks through all setup in one place — extension, project, model, RBAC — with collapsible 'details' toggles per path.
07-summary.md is a brand-new module that didn't exist before. It provides:
- What you built — recap of the "Explain Like I'm an Executive" agent
- Key concepts learned — table mapping concepts to what was practiced (Agent Framework architecture, prompt engineering, safety boundaries, local vs. hosted differences)
- Path-specific checklists — separate completion checklists for Path A (cloud) and Path B (local)
- Next steps & resources — links to Lab 02, Tool Catalog, RAG docs, Agent Framework SDK, extension changelog
- Ideas to extend — date tool, incident database, Bing grounding, model comparison, Foundry evaluations
- Path B → Path A upgrade guide — step-by-step instructions for when local-only users get a subscription
- Clean up resources — how to delete the resource group, agent, or model deployment
The biggest structural change: the lab now supports attendees who do not have a Azure subscription.
How it works
Path choice is made in Module 00 (Introduction) via collapsible toggles:
| Path | Audience | Model source | Modules covered Deploys to cloud? |
|---|---|---|---|
| Has active Azure subscription | Azure OpenAI via Foundry (e.g., gpt-4.1-mini) All (00–07) | ✅ Yes | |
| MVPs, students, anyone without Azure | Foundry Local (free, on-device) or GitHub Models (free tier) 00–04, then skip to 07 | ❌ No |
Path-specific content appears throughout using details toggles:
| Module | Path A content | Path B content | | ----| ----| -----| ----| | 01-setup.md | Azure CLI, azd, Docker, subscription, Foundry project, model deploy, RBAC | Foundry Local install via Foundry Toolkit Model Catalog or GitHub Models setup via Foundry Toolkit Model Catalog | | 03-configure-and-code.md | .env with AZURE_AI_PROJECT_ENDPOINT | .env with http://localhost:5273/v1 (Foundry Local) or https://models.inference.ai.azure.com (GitHub Models) + GITHUB_TOKEN | | 03-configure-and-code.md | Uses DefaultAzureCredential() | Path B code change: AzureKeyCredential(os.environ["GITHUB_TOKEN"]) for GitHub Models | | 05-deploy-to-foundry.md | Full deployment flow | Banner: "Path B users: skip to Module 07" | | 06-verify-in-playground.md | Cloud playground testing | Banner: "Path B users: skip to Module 07" | | 07-summary.md | Full completion checklist | Partial checklist + "Upgrade to cloud" guide |
- Session leads: The lab is now self-contained for both cloud and local users. Review the path toggles in 01-setup.md before delivery.
- Contributors: Source changes go to workshop/ only. Do not edit translations/ or translated_images/ directly.
- Localization: This update renames/renumbers most Lab01 doc files. Expect a full regeneration of Lab01 translations on the next CI run. Old translated files for deleted filenames should be cleaned up.
- MVPs / community: Path B (Foundry Local / GitHub Models) is the recommended track for users without Azure subscriptions. No Azure account, credit card, or RBAC setup required.