File Search Tool

Retrieve knowledge from uploaded documents using hybrid vector and keyword search in Microsoft Foundry agents.

File Search Tool

File Search augments agents with knowledge from documents like product information, internal documentation, or user-provided files.

How It Works

The file search tool implements retrieval best practices:

1. Rewrites queries for optimal search
2. Breaks down complex queries into multiple parallel searches
3. Runs hybrid search (vector + keyword) across vector stores
4. Reranks results to select most relevant content
5. Injects into context for agent to use

Default settings:

• Chunk size: 800 tokens
• Chunk overlap: 400 tokens
• Embedding model: text-embedding-3-large (256 dimensions)
• Max chunks in context: 20

Quick Start

Python

from azure.ai.projects.models import FileSearchTool, FilePurpose

# Upload file
file = project.agents.files.upload_and_poll(
    file_path="product_catalog.pdf",
    purpose=FilePurpose.AGENTS
)

# Create file search tool
file_search = FileSearchTool(file_ids=[file.id])

agent = project.agents.create_agent(
    model="gpt-4o",
    name="product-expert",
    instructions="Answer questions using uploaded product files",
    tools=file_search.definitions,
    tool_resources=file_search.resources,
)

Vector Stores

Vector stores enable file search capabilities:

• Store up to 10,000 files per vector store
• Attach one vector store to agent
• Attach one vector store to thread
• Automatic parsing, chunking, and embedding

Maximum file size: 512 MB Maximum tokens per file: 5,000,000

Setup Dependencies

Basic Setup

• Microsoft-managed storage and search

Standard Setup

• Uses your Azure Blob Storage
• Uses your Azure AI Search resource
• Full data control

See Standard Setup for details.