Implement generative AI and agentic solutions 2.2 Build agents by using Foundry
You attach the file search tool to an agent and upload product manuals to a vector store.
Study guide skill: Build agents that integrate retrieval, function-calling, and conversation memory
Question B1-025
You attach the file search tool to an agent and upload product manuals to a vector store. Which two tasks does the file search tool handle for you? (Choose two.)
Select 2 answers.
Answer: B, E
The file search tool automatically parses and chunks documents, generates and stores embeddings, runs hybrid (keyword and semantic) search across vector stores, and reranks the results to select relevant content. In the standard agent setup it uses the Azure AI Search and Blob Storage resources you connected, but the service still handles the whole ingestion. It doesn't fine-tune the model, translate content, or write the agent's instructions.
Deep dive
How it works
The file search tool grounds a Foundry agent in your own documents. Ingestion happens in the vector store: when you add a file, the process automatically parses, chunks, embeds and stores it in a vector database that supports both keyword and semantic search. The documented defaults are 800-token chunks with a 400-token overlap, the text-embedding-3-large embedding model (256 dimensions), and up to 20 chunks in context.
At query time, file search rewrites the query to optimize it for search, breaks complex queries into parallel searches, runs hybrid search (keyword plus semantic matching) across vector stores, and reranks the results to pick the most relevant content for the response.
A few limits and caveats are worth knowing. A vector store can hold up to 10,000 files, and you can attach at most one vector store to an agent and one to a conversation. With the standard agent setup, your files stay in your own storage and your Azure AI Search resource ingests them. File search also has charges on top of the token-based fees for model usage.
The other options, and when you'd use them
- Translating the manuals: Translation is the job of Azure Translator, a neural machine translation service. Use it, or document translation for whole files, when content has to change language.
- Writing the agent's instructions: You write the instructions. A prompt agent is defined through configuration, including its instructions, model and tools. File search only retrieves content to ground answers.
- Fine-tuning the agent's model: Fine-tuning changes the model's weights through additional training, and it doesn't replace retrieval for current information. Use it when you have high-quality training data and want, for example, a consistent tone, format or schema.
Exam tip
File search handles ingestion (parse, chunk, embed, store) and retrieval (query rewrite, hybrid search, rerank). If you need to change the content or the model, that's a different service.
References
Last reviewed: