GCP Generative AI Leader Practice Question 87: Data Store Tool Type
Google Cloud Generative AI Leader · Test 9 · question 7 of 10
A team is configuring tools for a new agent. They want the agent to ground its answers in a collection of internal PDF policy documents by retrieving relevant passages from them, rather than calling a live API endpoint or triggering an action in another system. Which tool type should they configure for this need?
Show answer and explanation
Answer:
A data store is the tool type built for retrieval grounding: it indexes a collection of documents so Agent Search can find and cite relevant passages when the agent answers a question, which is exactly what the PDF policy library needs. Option D is the tool built for this job.
- A is wrong because an extension is defined to reach a live API endpoint, for fetching fresh data or performing an action; the requirement here is retrieval over static documents, not an API call. - B is wrong because function calling has the model choose a function and parameters for the application to run in code; it is for invoking custom logic or APIs, not for indexing a document collection. - C is wrong because a plugin automates a workflow inside a third-party application; it does not index documents for retrieval-based grounding.
Reference: Grounding with Agent Search