AgentGen + Microsoft AutoGen

AutoGen agents can reason and collaborate, but their outputs are text. Integrate AgentGen as a registered function and any AutoGen agent can produce a shareable PDF or image — invoices, reports, certificates — and hand the download link back to the user.

Install

pip install pyautogen requests python-dotenv

Step 1 — Define the AgentGen functions

import os, requests

AGENTGEN_KEY = os.environ["AGENTGEN_API_KEY"]

def generate_pdf(html: str, page_size: str = "A4", page_size_source: str = "css") -> dict:
    """
    Render an HTML string to a PDF document using AgentGen.
    Returns a dict with 'url' (CDN download link) and 'tokens_used'.
    Cost: 2 tokens per page.
    """
    r = requests.post(
        "https://www.agent-gen.com/api/v1/generate/pdf",
        headers={"X-API-Key": AGENTGEN_KEY, "Content-Type": "application/json"},
        json={"html": html, "page_size_source": page_size_source, "format": page_size},
        timeout=30,
    )
    if r.status_code == 402:
        return {"error": "Insufficient tokens. Top up at agent-gen.com/pricing."}
    r.raise_for_status()
    return r.json()

def generate_image(html: str, viewport_width: int = 1200, selector: str | None = None) -> dict:
    """
    Render an HTML string to a PNG image using AgentGen.
    Returns a dict with 'url' (CDN link). Cost: 1 token.
    """
    r = requests.post(
        "https://www.agent-gen.com/api/v1/generate/image",
        headers={"X-API-Key": AGENTGEN_KEY, "Content-Type": "application/json"},
        json={"html": html, "viewport_width": viewport_width, "selector": selector, "format": "png"},
        timeout=30,
    )
    r.raise_for_status()
    return r.json()

Step 2 — Register with AutoGen agents

import autogen

config_list = [{"model": "gpt-4o", "api_key": os.environ["OPENAI_API_KEY"]}]

# The assistant agent that writes HTML and calls tools
assistant = autogen.AssistantAgent(
    name="DocumentAgent",
    system_message=(
        "You are a document generation specialist. When asked to create a PDF or image, "
        "write complete, styled HTML with inline CSS, then call generate_pdf or generate_image. "
        "Always share the returned URL with the user."
    ),
    llm_config={
        "config_list": config_list,
        "functions": [
            {
                "name": "generate_pdf",
                "description": "Render HTML to a PDF. Returns a CDN download URL. Cost: 2 tokens/page.",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "html": {"type": "string", "description": "Complete HTML to render"},
                        "page_size": {"type": "string", "enum": ["A4", "Letter"], "default": "A4"},
                        "page_size_source": {"type": "string", "enum": ["css", "format"], "default": "css"},
                    },
                    "required": ["html"],
                },
            },
            {
                "name": "generate_image",
                "description": "Render HTML to a PNG image. Returns a CDN URL. Cost: 1 token.",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "html": {"type": "string"},
                        "viewport_width": {"type": "integer", "default": 1200},
                        "selector": {"type": "string"},
                    },
                    "required": ["html"],
                },
            },
        ],
    },
)

# The user proxy that executes function calls
user_proxy = autogen.UserProxyAgent(
    name="UserProxy",
    human_input_mode="NEVER",
    max_consecutive_auto_reply=5,
    function_map={
        "generate_pdf": generate_pdf,
        "generate_image": generate_image,
    },
)

Step 3 — Run the agent

user_proxy.initiate_chat(
    assistant,
    message=(
        "Create a PDF invoice for Acme Corp. "
        "Services: API integration consulting — $4,500. "
        "Due: April 15, 2026."
    ),
)
# DocumentAgent writes HTML, calls generate_pdf, returns:
# → "Your invoice is ready: https://cdn.agent-gen.com/output/abc123.pdf"

Multi-agent workflow

AgentGen works equally well in multi-agent AutoGen setups. A DataAgent can fetch and summarise metrics while a DocumentAgent renders the final report PDF — with the function registered on the executor proxy shared between both:

groupchat = autogen.GroupChat(
    agents=[data_agent, document_agent, user_proxy],
    messages=[],
    max_round=10,
)
manager = autogen.GroupChatManager(groupchat=groupchat, llm_config={"config_list": config_list})

user_proxy.initiate_chat(
    manager,
    message="Analyse last week's sales data and generate a PDF summary report.",
)

Why AutoGen teams use this pattern