An Azure AI Foundry agent integrated with Copilot Studio uses Azure AI Search as its knowledge grounding index. When a Copilot Studio user asks a question and the Foundry agent retrieves an answer, where does the search for relevant knowledge actually occur?

Azure AI Foundry agents maintain their own knowledge infrastructure, which typically includes an Azure AI Search (formerly Cognitive Search) index populated with domain-specific documents. When Copilot Studio routes a user query to the Foundry agent, the Foundry agent independently executes retrieval-augmented generation (RAG) against its Azure AI Search index, then generates a grounded response.

This separation of knowledge planes is intentional and architecturally important: the Foundry agent's specialized knowledge (fine-tuned search index, custom chunking strategies, semantic ranking) operates independently of Copilot Studio's knowledge sources. The two systems each manage their own grounding data.

Option A (Copilot Studio knowledge sources) — the parent Copilot Studio agent's knowledge sources are only searched when the parent handles the query directly. When it delegates to Foundry, Foundry uses its own knowledge.

Option C (Dataverse conversation history) — conversation history stored in Dataverse provides context for the current session but is not a searchable knowledge index for grounding answers.

Option D (Bing public internet) — by default, Foundry agents are grounded in their configured knowledge index, not the public internet. Bing integration would be an explicit, additional configuration.

Architecture exam point: Copilot Studio knowledge ≠ Foundry agent knowledge. Each layer maintains its own retrieval infrastructure.