app.py

import os
from fastapi import FastAPI
from pydantic import BaseModel
from huggingface_hub import InferenceClient
from smolagents import BasicAgent
from crewai_tools import BaseTool

# Use your token from HF secrets
HF_TOKEN = os.environ.get("HF_TOKEN")  # Must be set in HF Space secrets

# Define a wrapper that conforms to SmolAgent’s LLM expectations
class HuggingFaceInferenceWrapper:
    def __init__(self, client: InferenceClient):
        self.client = client

    def __call__(self, prompt: str, **kwargs):
        # Use text_generation endpoint
        response = self.client.text_generation(prompt=prompt, max_new_tokens=512)
        return response

# Initialize InferenceClient
client = InferenceClient(
    model="Qwen/Qwen1.5-1.8B-Chat",  # Or another model you have access to
    token=HUGGINGFACE_API_KEY
)

# Wrap the client
llm = HuggingFaceInferenceWrapper(client)

# Define a dummy tool
class DuckDuckGoTool(BaseTool):
    name: str = "DuckDuckGo Search"
    description: str = "Use this tool to search for real-time facts and current events."

    def _run(self, query: str) -> str:
        return f"Search results for '{query}' (this is a placeholder)."

# Create the agent
agent = BasicAgent(
    role="Helpful AI",
    goal="Answer tough questions using search and reasoning",
    backstory="An expert at finding information and answering questions.",
    tools=[DuckDuckGoTool()],
    llm=llm
)

# Define FastAPI app
app = FastAPI()

class QuestionInput(BaseModel):
    question: str

@app.post("/ask")
def ask_question(payload: QuestionInput):
    result = agent.run(payload.question)
    return {"answer": result}