GitHub Copilot AI Assistant cannot generate insights from DataFrame only code?

[neenbohratwork]

Hi Marimo community,

I’m trying to use Marimo AI Assistant powered by GitHub Copilot (BASE_URL = https://api.githubcopilot.com/, model = gpt-4o-mini) to analyze a Pandas DataFrame and get insight/explanation (not code), but I’m running into an issue.

Here’s an example with fake data:

import pandas as pd

# Sample fake data
df= pd.DataFrame({
    "category": ["A", "B", "C", "A", "B"],
    "value": [10, 20, 15, 12, 18],
    "count": [100, 150, 120, 110, 140]
})

# Convert to markdown
df_text = df.to_markdown(index=False)
df_text

When I try to ask the AI Assistant in Marimo something like:

Please summarize the trends and explain the insights from df_text

It responds:

To summarize the trends and explain the insights from the DataFrame represented by df_text, I would typically need to see the actual contents of df_text. However, since I can't directly access your DataFrame or its markdown representation, I can guide you on how to analyze it based on what might typically be included.
Here's how you can approach summarizing trends and insights:
Possible Insights and Trends to Analyze...

Even after converting to markdown and running the cell, it still does not read the actual data and only provides guidance, not concrete insight.

It seems the AI Assistant (GitHub Copilot) only generates code based on variables, and cannot provide direct explanations/insights from DataFrames.

Question:
Has anyone experienced this issue? Is there a way to make AI Assistant read the DataFrame and give text explanations or summaries without forcing it to generate Python code?

I know that if I want explanations or summaries, I can use Marimo UI Chat + AI completion, but I just wanted to try using the GitHub Copilot AI Assistant.
Since I’m making a report in Marimo, it would be really nice if I could ask and get answers directly next to the report.

Thanks in advance for any suggestions!

[mscolnick]

@neenbohratwork are you using the @ symbol to reference the dataframe? e.g. @df_text?