Deploy AI agents on RunPod serverless GPU infrastructure

"""RunPod handler for CrewAI blog generation."""

import runpod
from crewai import Agent, Task, Crew, LLM
from crewai.tools import tool

# Configure Ollama LLM — loaded once at container startup
llm = LLM(model="ollama/openhermes", base_url="http://localhost:11434")

@tool("Research Tool")
def fake_research(topic: str) -> str:
    """Pretends to search for information about a topic."""
    return f"Key facts about {topic}: adoption is growing, costs are falling, and expert interest is high."

blog_writer = Agent(
    role="Blog Writer",
    goal="Write engaging and informative blog posts on various topics",
    backstory="You are a professional blog writer known for well-researched, clear articles.",
    tools=[fake_research],
    verbose=True,
    llm=llm
)

def create_blog_post(topic):
    """Creates a blog post on the given topic using CrewAI."""
    blog_task = Task(
        description=f"""
        Write a blog post about {topic}.
        Your blog should have an attention-grabbing title, a brief introduction,
        3-4 main points supported by research, and a conclusion.
        Use the Research Tool to gather facts about {topic}.
        """,
        expected_output="A well-structured blog post of approximately 500 words",
        agent=blog_writer
    )
    crew = Crew(agents=[blog_writer], tasks=[blog_task], verbose=True, llm=llm)
    result = crew.kickoff()
    return result.raw

def handler(job):
    """Handler function that will be used to process jobs."""
    job_input = job["input"]
    topic = job_input.get("topic", "technology")

    try:
        blog_post = create_blog_post(topic)
        return {"status": "success", "blog_post": blog_post}
    except Exception as e:
        return {"status": "error", "message": str(e)}


runpod.serverless.start({"handler": handler})