Hosted Agent — runs in Foundry Agent Service (isolated sandbox)
A Updated customer support agent that can search a knowledge base, create support tickets, track ticket status, and escalate complex issues to human agents when needed. It responds politely, provides step-by-step troubleshooting, and always confirms resolution before closing a case.
This agent uses the Agent Framework with ResponsesHostServer to run as a containerized hosted agent in Foundry Agent Service. The platform manages:
- Isolated VM sandbox per session
- Automatic scaling and lifecycle
- Session state persistence
- OpenTelemetry traces and monitoring
# 1. Install azd and the Foundry extension
azd ext install microsoft.foundry
azd auth login
# 2. Initialize from this project
azd ai agent init
# 3. Provision Azure resources
azd provision
# 4. Test locally
azd ai agent run --no-inspector
# 5. In another terminal, invoke
azd ai agent invoke --local "Hello, I need help"azd deployAfter deployment, the agent is available at:
- Playground: Foundry portal → Agents → Updated-Hosted-agent → Playground
- API:
{project_endpoint}/agents/Updated-Hosted-agent/endpoint/protocols/openai/responses
| Variable | Description |
|---|---|
FOUNDRY_PROJECT_ENDPOINT |
Foundry project endpoint URL |
AZURE_AI_MODEL_DEPLOYMENT_NAME |
Model deployment name (e.g., gpt-4o) |
This agent uses the Responses protocol (OpenAI-compatible). Any OpenAI SDK client can interact with it.
The included GitHub Actions workflow deploys on push to main using azd deploy.