Foundry IQ integration reaches general availability in Copilot Studio
Microsoft announces general availability for Foundry IQ in Copilot Studio, with private connectivity and requirements for access, testing and cost.
Microsoft announced general availability of the Foundry IQ integration with Copilot Studio on October 1. The release announcement includes private connectivity through Azure Private Link and Power Platform virtual network support, giving organizations a way to connect agents to enterprise knowledge without exposing the underlying search service to the public internet.
The change concerns the direct integration with Copilot Studio. Foundry IQ knowledge bases were already generally available, and connecting them to Copilot Studio had previously been described as a preview. Teams evaluating the release should check the status of each component they plan to use.
Reuse a knowledge base that already works
A Foundry IQ knowledge base combines data sources with retrieval and ranking settings. The connection guide describes reusing a base that another team has built and tuned, rather than setting up the same content again inside Copilot Studio.
The documented workflow requires an agent powered by the GitHub Copilot harness and access to an existing knowledge base. Each agent can have one Foundry IQ connection. Supported authentication choices include Microsoft Entra ID, a service principal, a client certificate or an API key.
Retrieval appears as a tool in the agent. Makers can inspect the response activity trace to see whether Foundry IQ ran and which items it returned. If the underlying results need improvement, Microsoft directs teams to adjust the sources, retrieval instructions or ranking in Foundry IQ.
Private connectivity still requires configuration
The announcement combines a private endpoint for the search service with delegated subnets for Power Platform traffic. Teams also need routing and private DNS to resolve the usual service address to the private endpoint. These are deployment requirements rather than settings that the integration automatically completes.
Microsoft’s Azure AI Search networking guide explains that Private Link carries traffic over Microsoft’s backbone network. Private endpoints require the Basic search tier or higher and are separately billable. Public network access must be disabled when the goal is to block connections through the public endpoint.
Test access and answers separately
A useful acceptance test checks more than whether the agent answers a question. Microsoft recommends confirming the retrieval trace, expected citations, different user permissions and the actual network route. A working connection should not be treated as evidence that every user sees only the intended documents.
For a support assistant, for example, test both an ordinary policy question and a restricted document that the same user should not receive. That gives the team evidence about answer quality and access enforcement. The results should remain distinct so a good answer cannot hide an overly broad permission.
The connection guide also notes that building, testing and running agents may consume Copilot Credits. Organizations should measure those costs alongside search and private connectivity charges before expanding a pilot.
ByteForward’s report on Agent Gateway tracing examines another part of operating enterprise agents. In this release, the practical milestone is bringing an existing knowledge layer into Copilot Studio under a supported integration and an explicitly configured network path.
Illustrative server racks photographed by Brett Sayles on Pexels, used under the Pexels License. The photograph is illustrative and does not document this release.
