Tutorial 4: Downloading agent-generated files

Overview

This tutorial shows you how to use the h2oGPTe Python client to download files that agents generate during chat sessions. You can use this process to retrieve documents, images, code, or other files that agents create.

What you’ll learn

How to retrieve and download files generated by h2oGPTe agents
How to handle file download errors
How to manage local file storage

Prerequisites

Before you begin, make sure you have:

the h2oGPTe Python library installed

a global API key and endpoint URL

h2oGPTe v1.6.29 or later

Install the h2oGPTe library

Install the library using pip:

pip install h2ogpte

Get your API credentials

Requirements:

An API key.

The endpoint URL for your h2oGPTe instance. The address should be where the API key was created.

To create a global API key, see Create an API key.

Note

Make sure the agent tool you want to use is configured and available in your h2oGPTe environment. Some tools require additional setup. For more information, see Agents tool documentation .

Download agent-generated files

1. Create a client connection

from h2ogpte import H2OGPTE
import json

client = H2OGPTE(
    address='https://h2ogpte.genai.h2o.ai',
    api_key='sk-XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX',
)

Replace the address and api_key values with your actual endpoint and API key.

2. Start a chat session

chat_session_id = client.create_chat_session()

3. Send a prompt to generate a file

with client.connect(chat_session_id) as session:
    reply = session.query(
        message='Generate an image of a zebra in a tophat', # The prompt for the agent
        timeout=60,
        llm_args={"use_agent": True}  # Enable agent tools for this query
    )

4. Get the file references

import json

agent_files = client.list_chat_message_meta_part(reply.id, "agent_files").content
agent_files = json.loads(agent_files)

The agent_files variable contains a list of dictionaries that map document IDs to filenames. For example:

[
    {"document_id1": "image_filename.png"},
    {"document_id2": "output.pdf"}
]

5. Download the files

for f in agent_files:
 doc_id = list(f.keys())[0]
 doc_name = f[doc_id]

 print(f"Attempting to download: {doc_name} (ID: {doc_id})")

 try:
     client.download_document("./", doc_name, doc_id)
     print(f"SUCCESS: Downloaded {doc_name} to current folder.")
 except Exception as e:
     print(f"FAILED to download {doc_name}: {e}")

The download_document() method saves files to the specified directory. Make sure you have write permissions for the target directory.