All AI news

Foundry previews the Copilot harness; runtime billing starts in December

Runtime is free only until December 1; regional limits and model/tool charges still apply.

Microsoft announced the Copilot harness on October 28. The Foundry-specific release identifies its managed agent loop and sandbox as a public preview, not general availability.

A new agent is required: existing agents do not automatically change, and switching harnesses after creation is not supported. The nine deployment regions exclude Hong Kong, Macau and Taiwan; that does not determine residents’ account eligibility.

Runtime billing starts December 1, 2026. Before then, only runtime usage is free; model inference, tools and connected services can still incur charges. The preview has no service-level agreement and is not recommended for production workloads.

This is a developer-platform/API capability, not a consumer web or mobile-app release. Local prices, plan eligibility and Traditional Chinese performance remain unconfirmed. Not tested.

https://devblogs.microsoft.com/foundry/microsoft-copilot-harness-in-foundry/
https://learn.microsoft.com/en-us/azure/foundry/agents/how-to/use-tools-skills-copilot-harness
https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/limits-quotas-regions

Usage conditions

Hong Kong
Hong Kong is not among the nine preview deployment regions. This does not establish residents’ eligibility to use other regions; local account access, pricing and Traditional Chinese support remain unconfirmed.
Macau
Macau is not among the nine preview deployment regions. This does not establish residents’ eligibility to use other regions; local account access, pricing and Traditional Chinese support remain unconfirmed.
Taiwan
Taiwan is not among the nine preview deployment regions. This does not establish residents’ eligibility to use other regions; local account access, pricing and Traditional Chinese support remain unconfirmed.

Official source: Introducing the Microsoft Copilot harness: Unlocking the potential of truly multi-model agentic work

Source published:

Source checked:

Published here: