Microsoft opens the engine behind Copilot to businesses building their own AI agents
Microsoft is making the Copilot harness, the agent engine behind GitHub Copilot, Copilot Studio and VS Code, available through Foundry. Customers can build agents on more than 20 models. Pricing is not stated.
Microsoft is giving businesses access to the same agent engine that runs its own Copilot products. The Copilot harness, which controls how AI agents receive instructions, tools and models, is becoming available through the Microsoft Foundry developer platform. Customers can build their own agents across more than 20 AI models without developing and maintaining that layer themselves.
Microsoft announced the move in a blog post on Wednesday 28 October, published at 16:51 UTC. The post is written by Tina Schuchman, corporate vice president for Foundry, and Vladimir Fedorov, chief technology officer of GitHub.
The engine behind Copilot
Microsoft describes a harness as everything around the model itself: the layer that gives an agent the right instructions, context, execution environment and controls so it can do real work. The model supplies the intelligence; the harness supplies the orchestration.
According to Microsoft, the Copilot harness was built by GitHub's engineers, and the company has now adopted it across the whole group. It powers Copilot Code, Copilot Studio, the GitHub Copilot app, the command-line tool, VS Code and other products. Microsoft says it has been tested at the scale of millions of users and hundreds of real-world use cases.
The company writes that the runtime was recently ported to the Rust programming language, which it says has cut session overhead and time to first token. Microsoft calls the performance industry-leading, but the post does not publish figures or independent measurements to support that.
Many models, one control layer
Microsoft says the harness supports more than 20 frontier AI models, and customers can also bring their own models with their own key. Microsoft says it can route work automatically to the model, or models, best suited to each task. The point is that an organisation can switch models without rebuilding the rest of its solution.
The harness supports open standards such as MCP, the protocol that connects agents to tools and data sources, as well as agent skills and plugins.
Customers can download a software development kit (SDK) and run the harness themselves, or run it in the cloud, managed by Foundry. In that case Microsoft runs the agent loop and the sandbox, with governance and optimisation built in. The customer supplies instructions, skills, tools and model choice. Microsoft stresses that the platform remains open to other harnesses and frameworks.
What the announcement does not say
The blog post does not state pricing, which regions the service runs in, whether it is generally available or in preview, or exactly which models are included. Nor does it explain how automatic model routing works when data must not be sent to particular providers.
The announcement comes on the same day that Microsoft-owned GitHub launched Copilot Fusion at its GitHub Universe developer conference, a feature that chooses which models should handle a task.
What it means for leaders
Our assessment: many organisations now build their own agent solutions and have to manage model switches, tool access and security themselves. Microsoft is offering to take over that work, using its own products as proof. That could mean faster development and less maintenance, but it also makes Microsoft a more visible intermediary between the organisation and the models. Before choosing it, IT and procurement teams should get answers on pricing, where data is processed and what controls they keep when the harness picks the model.
Sources and media
Illustration: AI-generated with Grok Bot / hogby.ai
