Home

---

title: Foundry Hosted Agents Workshop for Financial Services description: Bilingual hands-on workshop hosting a synthetic auto-insurance quote-preparation agent on Microsoft Foundry Hosted Agents.

Welcome to the foundry-hosted-agents-fsi wiki, adapted from the sibling foundry-hosted-agents repository.

Important

Every fixture, rulebook, and calculator output in this workshop is synthetic and non-binding. Nothing here represents an actual Desjardins product, rate, or policy.

Quick facts

• Repository: devopsabcs-engineering/foundry-hosted-agents-fsi
• azd environments: desjardins-quote-preparation-staging, desjardins-quote-preparation-poc
• Foundry account naming pattern: aif-${environmentName}
• Foundry project naming pattern: proj-${environmentName}
• Hosted agent name: quote-preparation-agent
• MCP tool servers: application-server (get_application), rulebook-server (get_rulebook)
• Web chat pilot app: apps/web-chat (deployed out of band to production -- see the Deployment Links table below, or run it locally against the deployed agent per the repository README's "Try the web chatbot" section)

Note

The "Deployment Links" section below is refreshed automatically by publish-test-trends.yml after each trusted main-branch run, from whatever azd/environment values were live in that run -- never fabricated. Azure Portal rows need Azure sign-in; the MCP rows are protocol endpoints that answer POST JSON-RPC rather than browser pages; the workshop site needs GitHub sign-in while this repository is private.

Deployment Links

Values reflect the current azd/environment configuration for each listed environment at run time, not proof that this run deployed or validated them. Rows for destinations that are not currently configured are omitted rather than invented.

Staging

Values reflect the current azd/environment configuration for the staging environment at run time, not proof that this run deployed or validated them. Rows for destinations that are not currently configured are omitted rather than invented.

Destination	Access and purpose
Try the staging web chatbot	Same-tenant pilot members only
Web app health (staging)	Public process health; not an agent-invocation test
Web app in Azure (staging)	Revisions, logs and metrics
Open the staging reviewer app	Same-tenant reviewers holding the Reviewer app role
Reviewer app health (staging)	Public process health; not an authorization test
Reviewer app in Azure (staging)	Revisions, logs and metrics
Resource group	Azure access required
Container registry	Image digests and remote builds
Foundry project (staging)	Azure access required
Responses API (staging)	Authenticated POST API, not a browser chat page
Application-server MCP (staging)	Synthetic MCP protocol endpoint, not a chat page
Rulebook-server MCP (staging)	Synthetic MCP protocol endpoint, not a chat page
Workshop site	Browsable lab content; GitHub sign-in required while the repository is private
Repository	Source and workflow history

Production

Values reflect the current azd/environment configuration for the production environment at run time, not proof that this run deployed or validated them. Rows for destinations that are not currently configured are omitted rather than invented.

Destination	Access and purpose
Try the production web chatbot	Same-tenant pilot members only
Web app health (production)	Public process health; not an agent-invocation test
Web app in Azure (production)	Revisions, logs and metrics
Open the production reviewer app	Same-tenant reviewers holding the Reviewer app role
Reviewer app health (production)	Public process health; not an authorization test
Reviewer app in Azure (production)	Revisions, logs and metrics
Resource group	Azure access required
Container registry	Image digests and remote builds
Foundry project (production)	Azure access required
Responses API (production)	Authenticated POST API, not a browser chat page
Application-server MCP (production)	Synthetic MCP protocol endpoint, not a chat page
Rulebook-server MCP (production)	Synthetic MCP protocol endpoint, not a chat page
Workshop site	Browsable lab content; GitHub sign-in required while the repository is private
Repository	Source and workflow history

Pages

• Workflows: what each GitHub Actions workflow does, gating status, and Azure identity/RBAC prerequisites.
• Architecture: the LangGraph agent topology, the Foundry services defined in azure.yaml, and the MCP tool servers.
• Operations: the release path, identity prerequisites, and how to inspect a release.
• Release Evidence: how to reproduce the deterministic evaluation gate and offline test evidence this repository actually produces.
• Test Trends: CI test and evaluation measurements with immutable source-run links, updated automatically after each trusted main-branch run.

Useful commands

Important

quote-preparation-agent is deployed to the proof-of-concept environment only (rg-desjardins-quote-preparation-poc), together with the application-server and rulebook-server MCP toolboxes -- these are the endpoints listed under Deployment Links above. Promotion beyond the proof of concept remains gated behind Gates G2 (platform/security), G3 (reproducible compatibility), and G6 (regulatory/privacy) sign-off. Treat every response as synthetic training output, never an insurance quote.

# Provision infrastructure
azd provision --no-prompt

# Deploy the hosted agent and MCP toolboxes
azd deploy --no-prompt

# Invoke the deployed agent with a synthetic case reference
azd ai agent invoke quote-preparation-agent "Prepare a quote for case CASE-SYN-001." --no-prompt

# Show the current agent version and identity details
azd ai agent show quote-preparation-agent --no-prompt

# Stream the hosted agent's own runtime logs
azd ai agent monitor quote-preparation-agent --no-prompt

To run the same case locally with no Azure dependency:

python src/quote-preparation-agent/main.py CASE-SYN-001