azd-ext-azure-ai-agents_1.0.0-beta.3

1.0.0-beta.3 (2026-07-03)

Features Added

• [#8852] Provision Foundry memory stores during azd deploy. Declare one or more memory stores under the agent service's memoryStores list in azure.yaml (with chatModel, embeddingModel, and optional extraction/retention options), and azd creates them in the Foundry project before deploying the agent. Provisioning is idempotent: existing stores are left unchanged, so deployments are safe to re-run. azd does not update an existing store; if a declared definition diverges from the live store, deploy warns which azure.yaml change(s) were not applied.
• [#8952] azd ai agent init now routes unified azure.yaml templates selected from the template picker through the Foundry adoption flow, so choosing one downloads the azure.yaml and its sibling files and scaffolds the project instead of failing while trying to git clone a file URL.

Bugs Fixed

• [#8941] Fix hosted agent deploys failing for users who lack Microsoft.Authorization/roleAssignments/write: the extension no longer assigns the redundant Azure AI User role to each per-agent managed identity after deploy, since Microsoft Foundry now grants that permission internally. Thanks @m5i-work for the contribution!
• [#8926] Fix --deploy-mode, --runtime, and --entry-point being silently ignored when azd ai agent init -m <azure.yaml> adopts a unified Foundry azure.yaml; the flags now apply code_configuration to the agent service, and an explicit --deploy-mode overrides a sample's pre-configured deploy mode.
• [#8933] Fix azd ai agent init -m <azure.yaml> returning early after scaffolding without running subscription selection, Foundry project setup, or model deployment verification, which left an environment that could not provision without manual configuration.