How to retrieve pandas dataframe agent-generated file?

Question

I have a pandas dataframe agent deployed in an Azure FastAPI app service.

            agent = create_pandas_dataframe_agent(
            llm,
            df,
            verbose=True,
            prefix=prefix,
            agent_type=AgentType.OPENAI_FUNCTIONS,
            agent_executor_kwargs={"memory": conversation_memory},
            return_intermediate_steps=True,
            allow_dangerous_code=True,
        )

The agent sometimes suggests generating a file from the data retrieved from the dataframe especially when the retrieved data is quite large.

How and where do I access the generated file to make it available for download?

I've implemented a workaround where, from the agent's intermediate steps, I extract the query that the agent generates from the the given prompt. I then execute that query on the original dataframe in order to retrieve the same data and make it available for download as a CSV file.

I suspect there's hopefully a more straightforward way of accessing any files that the agent generates than my current workaround.

Thanks.

Muhammad Mudassir · Accepted Answer · 2025-09-11 07:18:02Z

0

Here are two solid approaches, from the most robust to a more direct workaround.

Approach 1: The "Custom Tool" Method (Most Robust)

This is the most professional and scalable solution. Instead of letting the agent use its default, generic file-saving tool, you create your own custom tool and give it to the agent. This gives you complete control over the process.

Your custom tool will do two things:

Save the DataFrame to a CSV in a known, web-accessible directory on your server.
Return the publicly accessible URL for that file as its final output.

Here’s how that would look in a FastAPI context:

1. Create a custom tool for your agent:

import pandas as pd
import uuid
from langchain.tools import tool

# Assume you have a '/static/downloads' directory that FastAPI can serve files from.
DOWNLOAD_DIR = "/app/static/downloads/"
BASE_URL = "https://your-azure-app-service.com" # Or your server's base URL

@tool
def save_dataframe_and_get_link(df: pd.DataFrame, filename_prefix: str = "export") -> str:
    """
    Saves a pandas DataFrame to a CSV file in a web-accessible directory
    and returns a public download link. Use this tool whenever you need to
    provide a file for the user to download.
    """
    try:
        # Generate a unique filename to avoid conflicts
        unique_id = uuid.uuid4()
        filename = f"{filename_prefix}_{unique_id}.csv"
        full_path = f"{DOWNLOAD_DIR}{filename}"

        # Save the dataframe
        df.to_csv(full_path, index=False)

        # Generate the public URL
        download_url = f"{BASE_URL}/downloads/{filename}"

        print(f"DataFrame saved. Download link: {download_url}")
        return f"Successfully saved the data. The user can download it from this link: {download_url}"
    except Exception as e:
        return f"Error saving file: {str(e)}"

# When you initialize your agent, you pass this tool in the `tools` list.
# agent_executor = create_pandas_dataframe_agent(..., tools=[save_dataframe_and_get_link])

2. Update your FastAPI to serve these static files:

from fastapi import FastAPI
from fastapi.staticfiles import StaticFiles

app = FastAPI()

# This tells FastAPI to make the 'static' directory available to the public
app.mount("/static", StaticFiles(directory="static"), name="static")

# Your existing agent endpoint...
@app.post("/chat")
def handle_chat(...):
    # ... your agent runs and uses the custom tool ...
    result = agent.run(...)
    # The 'result' will now contain the download URL!
    return {"response": result}

I would strongly recommend Approach 1 for any production application. It gives you much more control and reliability

answered Sep 11 at 7:18

Muhammad Mudassir

12 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

crux Sep 12 at 7:06

Thank you @Muhammad Mudassir. Your answer pointed me in the right direction. However, when the dataframe agent invokes the custom tool, the following error is thrown: 2 validation errors for dataframe_to_csv_and_link df Input should be an instance of DataFrame [type=is_instance_of, input_value='top_5_clients.csv', input_type=str] For further information visit errors.pydantic.dev/2.11/v/is_instance_of filename Field required [type=missing, input_value={'df': 'top_5.csv'}, input_type=dict]

Muhammad Mudassir Sep 15 at 1:24

corrected and more robust approach: import pandas as pd import uuid from langchain.tools import tool from langchain_experimental.agents.agent_toolkits import create_pandas_dataframe_agent from langchain_openai import OpenAI # --- Assume these are defined in your environment --- # This is the base URL where your files will be accessible BASE_URL = "your-app-service.com" # This is the local directory your static files are served from DOWNLOAD

Collectives™ on Stack Overflow

How to retrieve pandas dataframe agent-generated file?

1 Answer 1

Approach 1: The "Custom Tool" Method (Most Robust)

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Approach 1: The "Custom Tool" Method (Most Robust)

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related