Can any autogen agent read pandas dataframe from the memory?

[Moha_Murali (mraguth)]

I have a need to work on panads datasets that have been loaded in my memory. Is that possible via autogen (any agent)?

[Bart Kolendowski (bkolendowski)]

Yes, but the question is do you want the agent to execute commands on the dataframe or do you merely want it to be aware of the contents of the dataframe?

[Moha_Murali (mraguth)]

I want to execute commands on the data frame by agents. Is that possible?

[Chi Wang (sonichi)]

Check section 3 of this example: https://github.com/microsoft/autogen/blob/main/notebook/agentchat_auto_feedback_from_code_execution.ipynb
for executing code in IPython.

[Tamish Mhatre (tamish-max)]

Yes — but only if you expose the dataframe to the agent in a way it can actually use.

In practice there are 3 good patterns:

1. Small dataframe / awareness only
   Pass schema + head() + dtypes + a few summary stats in the prompt.

2. Real dataframe analysis
   Give the agent Python/code-execution access in the same runtime where the dataframe already exists.

3. Safer / repeated use
   Wrap dataframe operations as tools (describe, filter, groupby, plot, etc.) instead of dumping the whole dataframe into the prompt.

So the short answer is: the agent cannot read your Python memory “by itself”, but it can work with an in-memory dataframe if you either:

• run code in that process/kernel, or
• expose the dataframe through tools/context.

For anything non-trivial, the Python-executor/tool route is usually much cleaner than serializing a large dataframe into chat history.