[AI-Suggested] Content update request from Blogs signal: From Local Models to Agent Workflows: Building a Deep Research Solution with Microsoft Agent Framework on Microsoft Foundry Local | Semantic Kernel #5742
Description
Background
A recent product update has been released that may impact our documentation. This could include new features, changes to existing functionality, or other product enhancements.
Your Action Required: Please review the suggestions below and update the relevant documentation to ensure accuracy and completeness.
Signal Details:
- Title: From Local Models to Agent Workflows: Building a Deep Research Solution with Microsoft Agent Framework on Microsoft Foundry Local | Semantic Kernel
- Source URL: https://devblogs.microsoft.com/semantic-kernel/from-local-models-to-agent-workflows-building-a-deep-research-solution-with-microsoft-agent-framework-on-microsoft-foundry-local
Source Signal Summary
Combines Foundry Local with Microsoft Agent Framework to build local AI agent workflows with privacy, low latency, and cost control while offering Azure OpenAI-like development patterns. Introduces security evaluation using Azure AI Evaluation Red Team with multiple attack strategies (e.g., ROT13, Base64, Unicode confusables) across risk categories (violence, hate/unfairness, sexual, self-harm), producing JSON scorecards. Implements a Deep Research workflow in MAF with modular executors and a research agent using web search tools, controlled iterative looping, and final structured reporting with citations. Provides MAF DevUI for interactive debugging with visual workflow topology, step-by-step node inspection, real-time parameter injection, and consolidated logs at http://localhost:8093. Integrates .NET Aspire and OpenTelemetry for enterprise observability, reporting latency, throughput, tool usage, model inference time, and memory metrics; setup leverages Azure CLI and Azure Identity for authentication.
Suggestions
Note
Suggestions are generated by AI and they may not be entirely accurate or complete. Please check impacted files scope and suggestions details before making changes.
- docs/ai/dotnet-ai-ecosystem.md:
- In the 'Develop with local AI models' section, add Microsoft Foundry Local as a supported local option alongside Ollama. Note that Foundry Local enables privacy-centric, low-latency, and cost-controlled development while offering Azure OpenAI–like development patterns for agents and model usage.
- In the 'Microsoft Agent Framework' section, briefly mention the MAF Developer UI (DevUI) for interactive debugging. Describe that it provides visual workflow topology, step-by-step node inspection, real-time parameter injection, and consolidated logs to accelerate troubleshooting during agent development.
- In the 'Microsoft Agent Framework' and/or a short 'Observability' paragraph, note that MAF integrates with .NET Aspire and OpenTelemetry for enterprise-grade telemetry. Summarize that common metrics include latency, throughput, tool usage, model inference time, and memory utilization to support production monitoring.
- Under 'Microsoft Agent Framework' responsible AI features, add a sentence that the framework can be paired with Azure AI Evaluation Red Team for adversarial safety testing. Indicate that typical strategies (such as ROT13, Base64, and Unicode confusables) are evaluated across risk categories like violence, hate/unfairness, sexual content, and self-harm with JSON scorecards.
- In the 'Develop with local AI models' section, clarify that MAF supports local agent workflows while maintaining familiar Azure-style development patterns. Briefly note that developers can use Azure Identity/Azure CLI when connecting tools or resources that require authentication, while local model inference itself can run without cloud dependencies.
📚 To learn more about agentic content maintenance workflow, visit Agentic workflow for Learn content maintenance. Sign in using Learn Profile to access the content on this link.
💬 Share your feedback on the Learn Content Maintenance Agentic Workflow here.