Image Input for Tools

This example demonstrates how tools can receive and process images automatically through Agno’s joint media access functionality. It shows initial image upload and analysis, DALL-E image generation within the same run, and cross-run media persistence.

Code

image_input_for_tool.py

from typing import Optional, Sequence

from agno.agent import Agent
from agno.db.postgres import PostgresDb
from agno.media import Image
from agno.models.openai import OpenAIChat
from agno.tools.dalle import DalleTools


def analyze_images(images: Optional[Sequence[Image]] = None) -> str:
    """
    Analyze all available images and provide detailed descriptions.

    Args:
        images: Images available to the tool (automatically injected)

    Returns:
        Analysis of all available images
    """
    if not images:
        return "No images available to analyze."

    print(f"--> analyze_images received {len(images)} images")

    analysis_results = []
    for i, image in enumerate(images):
        if image.url:
            analysis_results.append(
                f"Image {i + 1}: URL-based image at {image.url}"
            )
        elif image.content:
            analysis_results.append(
                f"Image {i + 1}: Content-based image ({len(image.content)} bytes)"
            )
        else:
            analysis_results.append(f"Image {i + 1}: Unknown image format")

    return f"Found {len(images)} images:\n" + "\n".join(analysis_results)


def count_images(images: Optional[Sequence[Image]] = None) -> str:
    """
    Count the number of available images.

    Args:
        images: Images available to the tool (automatically injected)

    Returns:
        Count of available images
    """
    if not images:
        return "0 images available"

    print(f"--> count_images received {len(images)} images")
    return f"{len(images)} images available"


def create_sample_image_content() -> bytes:
    """Create a simple image-like content for demonstration."""
    return b"FAKE_IMAGE_CONTENT_FOR_DEMO"


def main():
    # Create an agent with both DALL-E and image analysis functions
    agent = Agent(
        model=OpenAIChat(id="gpt-4o"),
        tools=[DalleTools(), analyze_images, count_images],
        name="Joint Media Test Agent",
        description="An agent that can generate and analyze images using joint media access.",
        debug_mode=True,
        add_history_to_context=True,
        send_media_to_model=False,
        db=PostgresDb(db_url="postgresql+psycopg://ai:ai@localhost:5532/ai"),
    )

    print("=== Joint Media Access Test ===\n")

    # Test 1: Initial image upload and analysis
    print("1. Testing initial image upload and analysis...")

    sample_image = Image(id="test_image_1", content=create_sample_image_content())

    response1 = agent.run(
        input="I've uploaded an image. Please count how many images are available and analyze them.",
        images=[sample_image],
    )

    print(f"Run 1 Response: {response1.content}")
    print(f"--> Run 1 Images in response: {len(response1.input.images or [])}")
    print("\n" + "=" * 50 + "\n")

    # Test 2: DALL-E generation + analysis in same run
    print("2. Testing DALL-E generation and immediate analysis...")

    response2 = agent.run(input="Generate an image of a cute cat.")

    print(f"Run 2 Response: {response2.content}")
    print(f"--> Run 2 Images in response: {len(response2.images or [])}")
    print("\n" + "=" * 50 + "\n")

    # Test 3: Cross-run media persistence
    print("3. Testing cross-run media persistence...")

    response3 = agent.run(
        input="Count how many images are available from all previous runs and analyze them."
    )

    print(f"Run 3 Response: {response3.content}")
    print("\n" + "=" * 50 + "\n")


if __name__ == "__main__":
    main()

Usage

1. Create a virtual environment

   Open the Terminal and create a python virtual environment.

   Mac

   python3 -m venv .venv
   source .venv/bin/activate
   
2. Export your OpenAI API key

   Mac/Linux

     export OPENAI_API_KEY="your_openai_api_key_here"
   
3. Set up PostgreSQL

   # Start PostgreSQL with Docker
   docker run -d \
     --name postgres-ai \
     -e POSTGRES_DB=ai \
     -e POSTGRES_USER=ai \
     -e POSTGRES_PASSWORD=ai \
     -p 5532:5432 \
     postgres:16
   
4. Install libraries

   pip install -U agno openai psycopg
   
5. Create a Python file

   Create a Python file and add the above code.

   touch image_input_for_tool.py
   
6. Run Agent

   Mac

   python image_input_for_tool.py
   
7. Find All Cookbooks

   Explore all the available cookbooks in the Agno repository. Click the link below to view the code on GitHub:

   Agno Cookbooks on GitHub