Microsoft Makes Its Multi-Model Copilot Harness Available Through Foundry

The company is offering the orchestration layer used across several of its agent experiences.

Microsoft has announced that its Copilot harness is available through Microsoft Foundry, positioning it as a managed foundation for building AI agents. The company outlined the move in an October 28 post by executives from Microsoft and GitHub.

A harness is the surrounding software that supplies an agent with instructions, context, tools, an execution environment and operating controls. Microsoft says the same foundation supports several of its own products and can work across multiple model families. The announcement also describes support for standards and extension mechanisms, including Model Context Protocol and agent skills.

Developers can use an SDK or a cloud-managed option, according to the post. Microsoft says it is responsible for the managed agent loop and sandbox, while users supply their chosen instructions and tools. These are product descriptions from the vendor; this article does not independently verify every performance claim or establish pricing and availability for a particular customer’s account.

Context: The model is only one component of an agent application. Systems also need to decide what information is available, which actions are permitted and how work is checked. Changing the surrounding software can therefore affect the behaviour of an application even when its underlying model stays the same.

For teams evaluating the offering, the relevant questions include integration, permissions, data handling and how failures are surfaced. Those details should be assessed against current documentation and the intended workload. The announcement offers another managed route to agent development, while the practical benefit remains dependent on how the harness is configured and evaluated for the tasks it will actually perform.