Microsoft Foundry extension for Azure Developer CLI is generally available
The microsoft.foundry extension bundle for Azure Developer CLI (azd) hits version 1.0 and is generally available. Hosted agents are now created, provisioned, tested locally in Agent Inspector, evaluated and deployed from CI/CD, all driven by an azure.yaml definition that lives next to your source.
Key points
- The microsoft.foundry extension bundle is GA at 1.0; existing installs upgrade with azd ext upgrade microsoft.foundry.
- Agent and deployment settings live in azure.yaml with the code, so changes are reviewed in the same pull request.
- azd ai agent run starts the agent locally and opens Agent Inspector; model calls still use your Azure deployment.
- azd deploy uploads the source to Foundry for a remote build, so no local container build is required.
- The azure.ai.evaluations extension generates datasets and rubrics and runs quality evaluations via azd ai eval commands.
- azd pipeline config scaffolds a GitHub Actions or Azure DevOps workflow, while azd ai agent doctor and azd ai agent monitor --follow help with troubleshooting.
The microsoft.foundry extension for Azure Developer CLI (azd) has reached version 1.0 and is generally available. azd no longer just provisions resources and deploys application code: it now creates and tests hosted agents, which run your own agent code on Foundry-managed compute. The agent definition and the Azure resources it needs live in azure.yaml next to the source, so configuration changes get reviewed in the same pull request as prompt or tool edits.
The loop starts with azd ai agent init from the hosted-agent sample, then azd provision; azd ai agent run starts the agent locally and opens Agent Inspector for request and response inspection, while azd deploy uploads the source for a remote build. The azure.ai.evaluations extension adds generated datasets and rubrics for quality checks, and azd pipeline config scaffolds a GitHub Actions or Azure DevOps workflow. You need azd 1.37.0 or later, a current uv release and model quota in your region.
Real-world use
- A team standing up its first agent in a Foundry project and keeping prompt plus infrastructure changes in one reviewable pull request
- A CI pipeline that scores the agent against a fixed dataset and rubric before each release and blocks deployment when quality drops
- Shipping the same hosted agent definition to separate stages with their own azd environments, subscriptions and regions
- Tearing down a demo environment with azd down to keep costs under control
Why it matters
Agent development, deployment and quality measurement collapse into one CLI workflow, so teams can review prompt changes alongside infrastructure and replay the same commands in CI.