Companies face a fundamental trade-off. They want to leverage the reasoning capabilities of modern cloud models, but are often restricted from processing sensitive data outside their own infrastructure due to compliance, privacy, or IP concerns.

Traditional approaches fail to resolve this. Local models often lack the quality needed for complex reasoning, while purely cloud-based solutions are not viable in many real-world scenarios.

In this session, I will present a hybrid architecture that addresses this challenge. A local agent processes and reduces sensitive raw data directly on-device using Foundry Local, transforming it into a non-sensitive, structured representation. Only this abstracted information is passed to Azure AI Foundry in the cloud, where the actual reasoning takes place.

Key Takeaways
• How to securely split processing between local and cloud-based reasoning
• How to integrate local models as specialized tools within a cloud agent
• End-to-end demo of a hybrid AI workflow with strict separation of sensitive and non-sensitive data

Target audience:
Developers, AI Engineers, and Architects

Session duration:
45 minutes

Products:
Foundry Local, Azure AI Foundry, Microsoft Agent Framework