Playground doesn't show tool results

[manoctch]

Hii,

I’ve created a model in Databricks that uses different tools. One of these tools generates a chart based on data read from Genie.

I’d like to visualize this chart (basically an image) in Playground, but so far I haven’t found a way to do it. At the moment, I can only see the base64 output instead of the actual chart.

Is there a way to show the images on Playground? 

Thanks

[AbhaySingh]

  The Issue:

  Databricks Playground typically has limited native support for rendering base64 images

  directly in the chat interface. This is a known limitation in many AI playground

  environments.

  Potential Workarounds:

  1. Use Markdown Image Syntax

  If your tool returns base64 data, try formatting it as a data URI with markdown:

  # In your tool's return value

  image_b64 = "your_base64_string_here"

  return f""

  2. Return a Public URL Instead

  Instead of returning base64, save the chart to a location accessible via HTTP:

  # Save to DBFS and return a displayable URL

  dbutils.fs.put(f"/tmp/chart_{uuid.uuid4()}.png", image_bytes)

  # Return the URL or use display() in notebooks

  3. Use DisplayHTML in Notebooks

  If testing in a Databricks notebook (not Playground), you can use:

  from IPython.display import HTML, display

  display(HTML(f'<img src="data&colon;image/png;base64,{image_b64}" />'))

  4. Structured Output with Content Type

  Ensure your tool returns properly structured output with content type hints:

  {

      "type": "image",

      "content": base64_string,

      "mime_type": "image/png"

  }

  5. File Path Reference

  Return a file path that can be accessed through Databricks:

  # Save to a volume or DBFS location

  return f"Image saved to: /Volumes/catalog/schema/volume/chart.png"

  Recommended Approach:

  The most reliable workaround is to:

  1. Save the generated chart to a Databricks volume or DBFS

  2. Return the file path in the tool response

  3. View the image separately in a notebook using display(Image(path))

  The Playground interface is primarily designed for text-based interactions, so rich

  media rendering may require using notebooks or custom applications instead.