AI-103 Practice Question 66: Custom Vector Store Function Tool
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 7 · question 6 of 10
A retailer already embeds and stores product-catalog vectors in its own Azure Cosmos DB account, where several other services query them with the database's native vector index. A new Foundry agent must ground its answers in the same embeddings. Company policy forbids creating a second copy of the embeddings in any Microsoft-managed store, and the retailer does not use Azure AI Search anywhere in its architecture.
How should the agent retrieve this data?
Show answer and explanation
Answer:
Function calling is designed for exactly this: "Define a function with its name, parameters, and description, and the agent's Foundry model can request your app to call it. Your app executes the function and returns the output." The retailer's own service, not Foundry, owns the retrieval logic; it queries the Cosmos DB vector index it already has and hands the results back as tool output. No embeddings ever leave the retailer's database, satisfying both constraints in the scenario.
- A is wrong because file search parses, chunks, and embeds files into its own managed vector store, a second, Microsoft-managed copy of the catalog, which the policy explicitly forbids. - C is wrong for the same reason from a different angle: it stands up a brand-new Azure AI Search index that duplicates data already embedded and served from Cosmos DB, and the retailer has no existing AI Search footprint to justify adding one. - D is wrong because code interpreter is a sandboxed execution environment for code the model writes against files you upload to it; it is not a connector to a live production database, and loading a "snapshot" is itself an unwanted copy of the data.
Reference: Use function calling with Microsoft Foundry agents