app.py

# app.py
import os
from typing import Dict, TypedDict, Annotated, Sequence
import gradio as gr
from langgraph.graph import END, StateGraph
from langchain_core.messages import AIMessage
import google.generativeai as genai
from langsmith import Client
from langchain_core.tracers.context import tracing_v2_enabled

# Configuration - use environment variables for secrets
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
LANGSMITH_API_KEY = os.getenv("LANGSMITH_API_KEY")

# Setting up APIs
if GEMINI_API_KEY:
    genai.configure(api_key=GEMINI_API_KEY)

if LANGSMITH_API_KEY:
    os.environ["LANGCHAIN_TRACING_V2"] = "true"
    os.environ["LANGCHAIN_API_KEY"] = LANGSMITH_API_KEY
    os.environ["LANGCHAIN_PROJECT"] = "multi_agent"
    os.environ["LANGCHAIN_ENDPOINT"] = "https://api.smith.langchain.com"
    client = Client()

# I use my free tier model
gemini_model = genai.GenerativeModel('gemini-1.5-flash')

class AgentState(TypedDict):
    messages: Annotated[Sequence[AIMessage], lambda x, y: x + y]
    user_query: str

def research_agent(state: AgentState) -> Dict:
    user_query = state['user_query']
    prompt = f"""You are a research assistant. Conduct research on: {user_query}
    Provide:
    - 3 key findings
    - 2 reliable sources
    - Potential applications
    Keep response concise (1-2 paragraphs)."""
    
    try:
        response = gemini_model.generate_content(prompt)
        return {"messages": [AIMessage(content=f"RESEARCH REPORT:\n{response.text}")]}
    except Exception as e:
        return {"messages": [AIMessage(content=f"Research failed: {str(e)}")]}

def writing_agent(state: AgentState) -> Dict:
    messages = state['messages']
    research_report = next(m.content for m in messages if "RESEARCH REPORT" in m.content)
    prompt = f"""Write a short article (3 paragraphs max) about:
    Topic: {state['user_query']}
    Research: {research_report}
    Include:
    - Introduction
    - Key points
    - Conclusion"""
    
    try:
        response = gemini_model.generate_content(prompt)
        return {"messages": [AIMessage(content=f"ARTICLE DRAFT:\n{response.text}")]}
    except Exception as e:
        return {"messages": [AIMessage(content=f"Writing failed: {str(e)}")]}

def editing_agent(state: AgentState) -> Dict:
    draft = next(m.content for m in state['messages'] if "ARTICLE DRAFT" in m.content)
    prompt = f"""Improve this draft (keep under 300 words):
    {draft}
    Make it:
    - More engaging
    - Better structured
    - Easier to read"""
    
    try:
        response = gemini_model.generate_content(prompt)
        return {"messages": [AIMessage(content=f"FINAL OUTPUT:\n{response.text}")]}
    except Exception as e:
        return {"messages": [AIMessage(content=f"Editing failed: {str(e)}")]}

def create_workflow():
    workflow = StateGraph(AgentState)
    workflow.add_node("researcher", research_agent)
    workflow.add_node("writer", writing_agent)
    workflow.add_node("editor", editing_agent)
    workflow.add_edge("researcher", "writer")
    workflow.add_edge("writer", "editor")
    workflow.add_edge("editor", END)
    workflow.set_entry_point("researcher")
    return workflow.compile()

app = create_workflow()

def process_input(user_input):
    try:
        result = app.invoke({"messages": [], "user_query": user_input})
        final_output = next(m.content for m in result['messages'] if "FINAL OUTPUT" in m.content)
        return final_output.split("FINAL OUTPUT:")[1].strip()
    except Exception as e:
        return f"Error: {str(e)}"

demo = gr.Interface(
    fn=process_input,
    inputs=gr.Textbox(label="Your question", placeholder="Ask me anything..."),
    outputs=gr.Textbox(label="Generated Article", lines=10),
    title="🧠 AI Research Assistant",
    description="A multi-agent system that researches topics and writes articles using Gemini 1.5 Flash",
    examples=[
        ["Explain quantum computing simply"],
        ["What are the health benefits of meditation?"],
        ["How do solar panels work?"]
    ],
    allow_flagging="never"
)

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