Process Images with Azure Computer Vision

The Azure Computer Vision Tool connects a LlamaIndex agent to an Azure Computer Vision account, letting the agent perform computer vision tasks directly on image URLs. It exposes one function, process_image, which sends an image to Azure for classification of objects and tags, captioning, or OCR text extraction.

This is a thin wrapper: the agent is simply given the ability to point at any accessible image URL and ask Azure Computer Vision what is in it or what text it contains, without the developer building a separate vision-API integration for the agent to call.

Use it when a LlamaIndex agent needs to reason about the visual content of an image that is accessible via URL - describing a picture, listing detected tags or objects, or reading text out of a screenshot or scanned document. Do not use it for images that are not reachable by URL (local-only files must be hosted or uploaded somewhere accessible first), and it depends entirely on an Azure Computer Vision resource being provisioned - it is not a standalone or self-hosted vision model.

process_image is the single tool: send an image URL and get back classification of objects and tags, an image caption, or OCR-extracted text, depending on what the agent asks for in its query.

from llama_index.tools.azure_cv import AzureCVToolSpec
from llama_index.core.agent.workflow import FunctionAgent
from llama_index.llms.openai import OpenAI

tool_spec = AzureCVToolSpec(api_key="your-key", resource="your-resource")

agent = FunctionAgent(
    tools=tool_spec.to_tool_list(), llm=OpenAI(model="gpt-4.1")
)

Requires an Azure account with an API key and a Computer Vision resource set up in advance.

Developers building LlamaIndex agents that need to see and describe images - captioning, tagging, or reading text from a URL - without integrating a separate vision API by hand.