Bring-your-own models reach GA in Microsoft's Foundry Agent Service

Microsoft said on April 27 that Bring Your Own Model, or BYOM, is now generally available in Foundry Agent Service for prompt agents. (techcommunity.microsoft.com) Foundry Agent Service is Microsoft’s managed platform for building and running AI agents, including no-code prompt agents and code-based hosted agents. It handles hosting, scaling, identity, observability, and security around those agents. (learn.microsoft.com) An agent is software that uses a large language model, instructions, and tools to carry out multi-step tasks instead of only answering with text. In Foundry, those tools can include web search, file search, memory, custom functions, and modular communication protocol servers. (learn.microsoft.com) BYOM lets a company keep its model endpoint behind its own gateway and still use Foundry’s agent features on top. Microsoft said the gateways can be Azure API Management or a third-party AI model gateway. (techcommunity.microsoft.com) (learn.microsoft.com) Microsoft said BYOM works with any model that implements the OpenAI Chat Completions API, regardless of provider. Setup is two steps: create a model connection, then create a prompt agent that uses that connection. (techcommunity.microsoft.com) The connection options are built for enterprise controls that already sit in front of model traffic. Microsoft listed API key, managed identity with a configurable audience, and OAuth 2.0 client credentials as supported authentication methods. (techcommunity.microsoft.com) Microsoft also said the routing can match different gateway URL patterns, including Azure OpenAI-style deployment paths and OpenAI-style routes. One gateway connection can register multiple model deployments, each with its own display name in Foundry. (techcommunity.microsoft.com) The company’s documentation adds a warning alongside the launch: BYOM models are treated as non-Microsoft products, and customers are responsible for their own safety mitigations, licensing review, and data-handling compliance. Microsoft said customers must review whether data sent to third-party models could leave Azure compliance or geographic boundaries. (learn.microsoft.com) This release lands a little over a month after Microsoft declared the next-generation Foundry Agent Service generally available on March 16. That broader GA added a Responses API-based runtime, private networking, Entra role-based access control, tracing, and evaluation features for production deployments. (devblogs.microsoft.com) Microsoft has also been tightening the path from local development to managed deployment. On April 22, it said Foundry Toolkit for Visual Studio Code reached general availability and can deploy agents directly to Foundry Agent Service from the development environment. (devblogs.microsoft.com) The net effect is that Microsoft is separating the agent runtime from the model provider more cleanly. Companies can keep their own gateways, policies, and endpoint choices while using Foundry as the control plane for agents. (techcommunity.microsoft.com) (learn.microsoft.com) For Microsoft, the April 27 update turns BYOM from a feature pitch into a production option inside a service it already moved to GA in March. For customers, it means the governance decisions now sit closer to the model gateway than the chatbot demo. (techcommunity.microsoft.com) (devblogs.microsoft.com)