Image Transcribe Document Agent
Transcribe a document image with a Ministral 14B agent.
The source-fidelity code uses deprecated pixtral-12b-2409. Replace it with ministral-14b-2512 before running the example.
"""
This agent transcribes an old written document from an image.
"""
from agno.agent import Agent
from agno.media import Image
from agno.models.mistral.mistral import MistralChat
# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------
agent = Agent(
model=MistralChat(id="pixtral-12b-2409"),
markdown=True,
)
agent.print_response(
"Transcribe this document.",
images=[
Image(url="https://ciir.cs.umass.edu/irdemo/hw-demo/page_example.jpg"),
],
)
# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------
if __name__ == "__main__":
passRun the Example
Set up your virtual environment
uv venv --python 3.12
source .venv/bin/activateInstall dependencies
uv pip install -U agno mistralaiExport your Mistral API key
export MISTRAL_API_KEY="your_mistral_api_key_here"Update the model
When saving the code, replace pixtral-12b-2409 with ministral-14b-2512.
Run the example
Save the code above as image_transcribe_document_agent.py, then run:
python image_transcribe_document_agent.pyFull source: cookbook/90_models/mistral/image_transcribe_document_agent.py