Workshop setup

This guide prepares your lab workstation and creates the fixed Azure resources for Microsoft Foundry: LangGraph Hosted Agents Workshop. You do not need to choose SKUs, names, or networking settings; the provisioning script creates the standard lab environment and writes .env.

Classroom assumptions

Each attendee uses a demo Azure subscription where they are Owner. Resource names include your short alias, for example rg-lgws-ada.

0. What you need

Requirement	Recommended	Minimum	Why
RAM	16 GB	8 GB	VS Code, notebooks, SDKs, and optional Docker
Disk	20 GB free	12 GB free	Python environment, docs tooling, optional containers
Permissions	Local admin	Local admin	Tool installation
Azure	Demo subscription Owner	Demo subscription Owner	Provisioning and RBAC assignment

1. Install tools

Tool	Version	Windows winget	macOS brew	Used for
Git	Current	winget install Git.Git	brew install git	Clone the repo
Python	3.13	winget install Python.Python.3.13	brew install python@3.13	Notebooks and validation
uv	Current	winget install astral-sh.uv	brew install uv	Python dependency sync
Azure CLI	>= 2.70	winget install Microsoft.AzureCLI	brew install azure-cli	Azure login and deployments
Azure Developer CLI	>= 1.27.1	winget install Microsoft.Azd	brew install azure-developer-cli	Hosted-agent deployment
PowerShell	7.x	winget install Microsoft.PowerShell	brew install powershell	Windows-friendly scripts
VS Code	Current	winget install Microsoft.VisualStudioCode	brew install --cask visual-studio-code	Labs and notebooks
Docker Desktop	Optional	winget install Docker.DockerDesktop	brew install --cask docker	Optional container-mode deploys
Node.js	Optional LTS 22	winget install OpenJS.NodeJS.LTS	brew install node@22	Optional MCP/tool labs

VS Code extensions installed by the script:

• ms-python.python
• ms-toolsai.jupyter
• ms-windows-ai-studio.windows-ai-studio (Foundry Toolkit)
• ms-azuretools.vscode-bicep
• ms-azuretools.vscode-docker

WindowsmacOSLinux

Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
.\scripts\install-tools.ps1
# Optional extras:
.\scripts\install-tools.ps1 -IncludeDocker -IncludeNode

./scripts/install-tools.sh
# Optional extras:
./scripts/install-tools.sh --include-docker --include-node

./scripts/install-tools.sh

2. Sign in

az login
az account show -o table
azd config set auth.useAzCliAuth true
azd auth login --check-status
azd ext install azure.ai.agents

The key setting is azd config set auth.useAzCliAuth true; it makes azd reuse your Azure CLI login during the labs.

3. Get the code

git clone https://github.com/Osshaikh/foundry-langgraph-workshop.git
cd foundry-langgraph-workshop
uv sync --extra docs
uv run python -m ipykernel install --user --name foundry-langgraph-workshop --display-name "Foundry LangGraph Workshop"

4. Provision Azure

Pick a short lowercase alias, 3-8 characters, such as your initials. The script creates everything with fixed names and writes .env.

WindowsmacOS / Linux

.\scripts\provision.ps1 -Alias <alias>

./scripts/provision.sh --alias <alias>

Estimated time: 10-15 minutes.

What gets created

Resource	Name	SKU / settings	Used by labs
Resource group	rg-lgws-<alias>	Container for all resources	All labs
Foundry account	aif-lgws-<alias>	AI Services S0, public access, project management enabled	Hosted agents, inference, evaluation
Foundry project	proj-lgws	System-assigned identity	All hosted-agent labs
Chat deployment	gpt-5.4-mini	GlobalStandard, default capacity 250	Most inference labs
Reasoning deployment	gpt-5.4	GlobalStandard, default capacity 150	Reasoning labs
Embedding deployment	text-embedding-3-large	GlobalStandard, default capacity 150	RAG labs
Fine-tune base	gpt-4.1-mini	GlobalStandard, default capacity 100	Later fine-tuning lab
Azure AI Search	srch-lgws-<alias>	Basic, semantic ranker free, Entra or API key auth	RAG and indexing labs
Log Analytics	log-lgws-<alias>	PerGB2018, 30-day retention	Observability
Application Insights	appi-lgws-<alias>	Workspace-based	Tracing and evaluation
Container Registry	acrlgws<alias>	Standard, admin disabled	Optional container-mode deploys
RAI policy	lgws-strict	Blocking, low threshold filters	Guardrails lab

Quota note

If model quota is tight, the provisioning scripts automatically retry once with half-size model capacities. If the retry still fails, ask your instructor for a different demo subscription or a region with quota.

Rough cost guide

Approximate list prices (USD, pay-as-you-go) for a two-day class. Verify current prices in the Azure Pricing Calculator for your region.

Item	Approx. cost	Notes
Azure AI Search Basic	~$2.50 / day	Fixed while the service exists
Container Registry Standard	~$0.70 / day	Fixed; only used for optional container deploys
Log Analytics + App Insights	< $1 / day	Pay per GB ingested; workshop traces are small
Model tokens (all labs)	~$2–10 per attendee	Pay per token; M8 deep research and M9/M12 evals use the most
Hosted-agent sessions	< $1 per attendee	CPU + memory only while a session is active (0.5 vCPU / 1 GiB)
Fine-tuning (M14)	~$5–15 training + tokens	The lab uses a Developer Tier deployment: no hosting fee, auto-deleted after 24 h. (Standard/Global FT deployments bill hourly.)
Typical total	~$15–30 per attendee	Run teardown when the class ends

Delete when done

Search and ACR keep billing while they exist (as would a Standard/Global fine-tuned deployment). teardown removes everything.

5. Verify

WindowsmacOS / Linux

.\scripts\preflight.ps1

./scripts/preflight.sh

Then open docs/modules/00-preflight.ipynb in VS Code, select the Foundry LangGraph Workshop kernel and Run All. Every required row should be ✅.

Docker is optional

The labs deploy hosted agents with code deploy: Foundry builds your Python dependencies in the cloud. Docker Desktop is only needed if you choose container-mode deploys.

6. Clean up

WindowsmacOS / Linux

.\scripts\teardown.ps1 -Alias <alias>

./scripts/teardown.sh --alias <alias>

The teardown deletes rg-lgws-<alias> and purges the soft-deleted Foundry account so the name can be reused.

Common problems

az is not on PATH after install

Restart your terminal, then run az --version again.

Quota errors

Re-run provisioning; it retries with lower model capacities. If it still fails, use another demo subscription or ask your instructor.

403 or authorization errors

Role assignments can take about 5 minutes to propagate. Wait, then run the preflight script again.

Lab machine is an Azure VM or Arc-enabled server

DefaultAzureCredential may pick up the machine's managed identity instead of your az login. Set it to use developer sign-ins only:

WindowsmacOS / Linux

[Environment]::SetEnvironmentVariable("AZURE_TOKEN_CREDENTIALS", "dev", "User")

echo 'export AZURE_TOKEN_CREDENTIALS=dev' >> ~/.bashrc

Restart VS Code afterwards. Hosted agents in Foundry are not affected.

Windows execution policy

Run Set-ExecutionPolicy -Scope CurrentUser RemoteSigned, then retry the PowerShell script.