Microsoft opens its Copilot harness to developers in Foundry, with billing from 1 December

What Microsoft announced

Microsoft is making its Copilot harness available to outside developers. In a Foundry blog post on 28 October 2026, the company introduced the Microsoft Copilot harness in Foundry Agent Service in public preview. A harness is the code around a model that gives an agent its instructions, context, tools, execution environment and controls. In Microsoft Foundry, Microsoft runs the agent loop and the sandbox, and the developer supplies the knowledge, procedures and tools.

In a companion post on the Official Microsoft Blog, Foundry CVP Tina Schuchman and GitHub CTO Vladimir Fedorov say GitHub engineers built the harness and that Microsoft has adopted it company-wide. They say it powers Copilot Code, Copilot Studio, the GitHub Copilot app, the CLI and VS Code, that its runtime was recently ported to Rust, and that it works with more than 20 frontier models plus bring your own key.

How it works

The preview has four building blocks. Reusable skills package instructions, scripts and resources into procedures. Connected tools cover supported Foundry tools and Model Context Protocol servers, with explicit authentication and approvals. Agents can work with files and code and delegate bounded tasks to built-in subagents. Agent definitions can live in source control next to application code, for example through the Azure Developer CLI.

Developers must choose the harness when they create an agent. Existing agents are unchanged, and switching harness after creation is not yet supported. Foundry manages the sandbox's CPU, memory and idle settings. Microsoft quotes Confluence Technologies, which says it had a solution running in under a week. The GitHub Copilot SDK, MIT-licensed on GitHub, lets teams run the same harness themselves.

Pricing and hosted agents

Harness runtime use is free until 1 December 2026. From that date, sandbox compute costs $0.0994 per vCPU-hour and memory $0.0118 per GiB-hour, and Microsoft says prices may vary by region and agreement. Model inference and separately billed tools are charged as usual in both periods.

Microsoft also made long-running resilience for hosted agents generally available. Hosted agents let teams bring their own framework, such as the Microsoft Agent Framework, while Foundry handles hosting. Foundry detects interruptions and restarts work with its original input, and the developer's code is responsible for saving progress and avoiding duplicate actions.

What it means and what to watch

The pitch is that teams stop building their own agent loop and model router. It puts Microsoft in direct competition with agent runtimes from other clouds and labs, and it comes from the GitHub team behind the multi-model orchestration in Copilot Fusion, announced the same day. Watch the general availability date, which models Foundry allows in the harness, and what real workloads cost once billing starts. More agent platform news is on the AI Agents hub.

Questions people ask

What is the Microsoft Copilot harness?

The agent runtime GitHub built for Copilot. It manages the agent loop, context, tools and sandbox. Microsoft says it powers Copilot Code, Copilot Studio, VS Code and the GitHub Copilot app.

How much does the Copilot harness in Foundry cost?

The runtime is free until 1 December 2026. After that, sandbox compute is $0.0994 per vCPU-hour and memory $0.0118 per GiB-hour, plus normal model and tool charges.

Can I switch an existing Foundry agent to the Copilot harness?

Not during the preview. Microsoft says you select the harness when creating the agent, and switching after creation is not supported yet.

Sources

1. Introducing the Microsoft Copilot harness in Foundry Agent Service (Microsoft Foundry Blog, 28 Oct 2026)
2. Introducing the Microsoft Copilot harness: Unlocking the potential of truly multi-model agentic work (The Official Microsoft Blog, 28 Oct 2026)
3. github/copilot-sdk repository (GitHub, 28 Oct 2026)