microsoft-foundry

Azure AI Foundry agent deployment involves creating prompt or hosted agents, deploying containers, starting runtimes, and managing versions. You can handle the entire workflow from project creation through deployment, invocation, evaluation, tracing, and troubleshooting using Foundry MCP tools and Azure CLI.

Evaluating and troubleshooting Foundry agents requires querying traces in Application Insights, correlating evaluation results, and inspecting runtime failures. You can build datasets from production traces and iteratively optimize prompts with regression checks to improve agent performance.

Yes, Azure AI Foundry capacity planning lets you find available quota across regions and handle PTU planning for Azure OpenAI models. You can deploy models, verify regional capacity, and manage quota allocations before committing to production deployments.

Azure CLI and Foundry MCP tools are required for core agent operations, while Docker or Azure Developer CLI provide optional support for production-grade container deployments. Application Insights queries are also needed for tracing and troubleshooting agent behavior.

Azure AI Foundry project management includes RBAC configuration to control access to agents, models, and project resources. You configure role-based permissions across Foundry projects to ensure proper authorization for deployment, invocation, and evaluation workflows.

Dataset curation from production traces involves querying Application Insights for agent invocation data, extracting successful and failed interactions, and building evaluation datasets. You then iterate on prompts using these datasets with regression checks to validate improvements.