app.py

import gradio as gr
import os
import pandas as pd
import datasets

from smolagents import CodeAgent, OpenAIServerModel
from tools import DuckDuckGoSearchTool, WeatherInfoTool, HubStatsTool, NewsSearchTool
from retriever import load_guest_dataset

# Constants
SAMPLE_FILE = "sample_guests.csv"

# Generate sample dataset if not already present
def generate_sample_guest_csv():
    if not os.path.exists(SAMPLE_FILE):
        guest_dataset = datasets.load_dataset("agents-course/unit3-invitees", split="train")
        df = pd.DataFrame(guest_dataset)
        df.to_csv(SAMPLE_FILE, index=False)

generate_sample_guest_csv()

# Set up model
model = OpenAIServerModel(model_id="gpt-4o")

# Initialize tools
search_tool = DuckDuckGoSearchTool()
weather_info_tool = WeatherInfoTool()
hub_stats_tool = HubStatsTool()
news_tool = NewsSearchTool(api_key=os.getenv("CONTEXTUALWEB_API_KEY"))

# Dynamically create agent with selected guest file
def build_agent(file_path=None):
    guest_info_tool = load_guest_dataset(file_path=file_path)
    return CodeAgent(
        tools=[guest_info_tool, weather_info_tool, hub_stats_tool, search_tool, news_tool],
        model=model,
        add_base_tools=True,
        planning_interval=3
    )

# Agent instance placeholder
agent_instance = None

# Gradio UI
with gr.Blocks() as demo:
    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("## Agent interface")
            gr.Markdown("This web UI allows you to interact with a `smolagents` agent that can use tools and execute steps to complete tasks.")
            gr.File(value=SAMPLE_FILE, label="📥 Download sample_guests.csv", interactive=False)
            guest_file = gr.File(label="📤 Upload guest CSV/JSON", type="filepath", file_types=[".csv", ".json"])
            prompt = gr.Textbox(label="Your request")
            submit = gr.Button("Submit")

            example_prompts = [
                ["List guests"],
                ["Give some examples of conversation starters based on each guest's interests?"],
                ["What's the weather like in Amsterdam tonight? Will it be suitable for our fireworks display?"],
                ["One of our guests is from Qwen. What can you tell me about their most popular model?"],
                ["I need to speak with Dr. Nikola Tesla about recent advancements in wireless energy. Can you help me prepare for this conversation?"],
            ]

            gr.Examples(
                examples=example_prompts,
                inputs=[prompt],
                label="💡 Example Prompts sample_guests.csv",
            )
            gr.Markdown("Powered by **smolagents**")

        with gr.Column(scale=3):
            output = gr.Chatbot(label="Agent", type="messages")

    def run_query(prompt, file):
        global agent_instance
        agent_instance = build_agent(file_path=file)
        result = agent_instance.run(prompt)

        # Handle different result types to convert to string for chatbot output
        if isinstance(result, dict):
            result = "\n\n".join(f"**{k}**: {v}" for k, v in result.items())
        elif isinstance(result, list):
            if all(isinstance(item, dict) and "name" in item and "starter" in item for item in result):
                result = "\n\n".join(f"{item['name']}: {item['starter']}" for item in result)
            else:
                result = str(result)
        else:
            result = str(result)

        # Return as list of message dicts for Gradio chatbot type="messages"
        return [
            {"role": "user", "content": prompt},
            {"role": "assistant", "content": result}
        ]

    submit.click(fn=run_query, inputs=[prompt, guest_file], outputs=output)

if __name__ == "__main__":
    demo.launch()