I'm trying to understand the networking model for Microsoft Foundry. When an agent needs to access services running inside an Azure VNet, does Foundry use a private endpoint, a service endpoint, or another connectivity method? I'd also like to understand the difference between running an agent with VNet integration in my tenant and using Microsoft-hosted networking. Any practical guidance or architecture examples would be helpful.
3 Answers
There are two separate traffic directions to consider. For inbound access to the Foundry resource, a private endpoint can let clients in your VNet reach the resource privately. For outbound access from an agent to resources in your VNet, you need the Foundry agent’s VNet networking or another network-connected intermediary. Without that configuration, outbound traffic uses Microsoft-hosted networking and does not automatically have access to your private VNet services.
A private endpoint is mainly for private inbound connectivity from your network to an Azure platform service. It does not make arbitrary services in your VNet reachable from Foundry. If the agent needs to call private APIs or other internal services, consider the supported Foundry VNet integration model, or place an intermediary such as API Management or an MCP server in a VNet and have the agent call that endpoint.
Think of a private endpoint as a network interface in your VNet that provides private access to a PaaS resource. It is not an outbound route from Foundry into your VNet. A VM can reach Foundry through the Foundry resource’s private endpoint, assuming DNS and routing are configured correctly, but that is different from Foundry being able to initiate connections back to the VM or another private service.
I’m new to Foundry—what is the practical difference between running an agent with VNet integration and using Microsoft-hosted networking?