app.py

import os
import gradio as gr
import json
import logging
import time
from typing import List, Dict, Generator
from dotenv import load_dotenv

# Try to use truststore for corporate network compatibility (local only)
if not os.getenv("SPACE_ID"):
    try:
        import truststore
        truststore.inject_into_ssl()
        print("💡 Truststore injected (Corporate SSL mode)")
    except ImportError:
        pass

# Import our agents from the src directory
from src.clarifier import Clarifier
from src.planner import Planner
from src.splitter import Splitter
from src.coordinator import Coordinator
from src.reviewer import Reviewer

# Configuration and Secrets
load_dotenv()

def get_secret(key):
    val = os.getenv(key)
    return val.strip() if val is not None else ""

HF_KEY = get_secret("HF_KEY")
TAVILY_API_KEY = get_secret("TAVILY_API_KEY")

# Logging setup
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# --- MODELS ---
# Using the most stable and powerful mid-sized model for all tasks
STABLE_MODEL = 'Qwen/Qwen2.5-7B-Instruct'

CLARIFIER_MODEL = STABLE_MODEL
PLANNER_MODEL = STABLE_MODEL
SPLITTER_MODEL = STABLE_MODEL
COORDINATOR_MODEL = STABLE_MODEL
SUBAGENT_MODEL = STABLE_MODEL
REVIEWER_MODEL = STABLE_MODEL

# --- GRADIO THEME ---
theme = gr.themes.Soft(
    primary_hue="blue",
    secondary_hue="slate",
    neutral_hue="slate",
    font=[gr.themes.GoogleFont('Inter'), 'sans-serif'],
)

def start_clarification(topic, hf_key, state):
    if not topic:
        return gr.update(), "### ⚠️ Warning\nPlease enter a topic.", state, gr.update()
    
    clean_key = hf_key.strip() if hf_key else ""
    if not clean_key:
        return gr.update(), "### ⚠️ Warning\nPlease provide a valid Hugging Face Token.", state, gr.update()
    
    state["initial_topic"] = topic
    clarifier = Clarifier(model_name=CLARIFIER_MODEL, hf_key=clean_key)
    try:
        suggestions = clarifier.get_suggestions(topic)
        state["suggestions"] = suggestions
        
        if not suggestions:
            return gr.update(), "### ❌ Error\nNo suggestions received. This could be due to model traffic. Please try again in 5 seconds.", state, gr.update()

        suggestion_md = "### 💡 Refine Your Topic\n\nChoose one of the suggested directions or enter a custom one below:\n\n"
        for i, s in enumerate(suggestions):
            suggestion_md += f"**Option {i+1}: {s['title']}**\n{s['description']}\n\n"
        
        return gr.update(visible=True), suggestion_md, state, gr.update(visible=False)
    except Exception as e:
        error_msg = str(e)
        if "401" in error_msg:
            error_msg = "401 Unauthorized: Your Hugging Face Token is invalid for the Inference API. Ensure it is a 'Write' token or has 'Inference' scope enabled."
        return gr.update(), f"### ❌ Error\n{error_msg}", state, gr.update()

def select_suggestion(index, custom_topic, state):
    if custom_topic and custom_topic.strip():
        state["final_topic"] = custom_topic
    elif index is not None and 0 <= int(index)-1 < len(state.get("suggestions", [])):
        sug = state["suggestions"][int(index)-1]
        state["final_topic"] = f"{sug['title']}: {sug['description']}"
    else:
        return gr.update(), "### ⚠️ Warning\nPlease select an option or enter a custom topic.", state, gr.update()
    
    return gr.update(visible=True), f"### 🎯 Target Topic\n**{state['final_topic']}**", state, gr.update(visible=False)

def generate_strategy(hf_key, state):
    clean_key = hf_key.strip() if hf_key else ""
    if not clean_key:
        return "### ⚠️ Warning\nHF Key missing.", state, gr.update()
    
    planner = Planner(model_name=PLANNER_MODEL, hf_key=clean_key)
    try:
        plan = planner.plan(state["final_topic"])
        state["research_plan"] = plan
        return plan, state, gr.update(visible=True)
    except Exception as e:
        return f"### ❌ Error\n{str(e)}", state, gr.update()

def decompose_tasks(hf_key, state):
    clean_key = hf_key.strip() if hf_key else ""
    if not clean_key:
        return "### ⚠️ Warning\nHF Key missing.", state, gr.update()

    splitter = Splitter(model_name=SPLITTER_MODEL, hf_key=clean_key)
    try:
        subtasks = splitter.split(state["research_plan"])
        state["subtasks"] = subtasks
        
        tasks_md = "### 📋 Generated Subtasks\n\n"
        for task in subtasks:
            tasks_md += f"- **{task['title']}** (ID: `{task['id']}`)\n"
            
        return tasks_md, state, gr.update(visible=True)
    except Exception as e:
        return f"### ❌ Error\n{str(e)}", state, gr.update()

def run_research(hf_key, tavily_key, state):
    from smolagents import CodeAgent, tool, InferenceClientModel
    from src.prompts import SUBAGENT_DIRECTION, COORDINATOR_DIRECTION
    from tavily import TavilyClient
    
    clean_hf = hf_key.strip() if hf_key else ""
    clean_tavily = tavily_key.strip() if tavily_key else ""

    if not clean_hf:
        yield "### ❌ Error\nHF Token missing.", state, gr.update(), ""
        return
    if not clean_tavily:
        yield "### ❌ Error\nTavily API Key missing.", state, gr.update(), ""
        return

    tavily_client = TavilyClient(api_key=clean_tavily)

    @tool
    def web_search(query: str) -> str:
        """
        Search the web for real-time information using Tavily.
        Args:
            query: The search query to look up.
        """
        try:
            response = tavily_client.search(query=query, search_depth="advanced", max_results=5)
            results = response.get("results", [])
            formatted = [f"Title: {r['title']}\nURL: {r['url']}\nContent: {r['content']}\n" for r in results]
            return "\n---\n".join(formatted) if formatted else "No results."
        except Exception as e:
            return f"Search failed: {e}"

    coordinator_model = InferenceClientModel(model_id=COORDINATOR_MODEL, api_key=clean_hf)
    subagent_model = InferenceClientModel(model_id=SUBAGENT_MODEL, api_key=clean_hf)
    
    current_findings = []
    log_content = "### 🔍 Agentic Research Progress\n\n"
    
    subtasks = state.get("subtasks", [])
    if not subtasks:
        yield "### ❌ Error\nNo subtasks found.", state, gr.update(), ""
        return

    for i, task in enumerate(subtasks):
        t_id = task['id']
        t_title = task['title']
        t_desc = task['description']
        
        log_content += f"**Agent {i+1} working on:** {t_title}...\n"
        yield log_content, state, gr.update(), ""
        
        subagent = CodeAgent(
            tools=[web_search],
            model=subagent_model,
            add_base_tools=False,
            max_steps=2
        )
        prompt = SUBAGENT_DIRECTION.format(
            user_query=state.get("final_topic", ""),
            research_plan=state.get("research_plan", ""),
            subtask_id=t_id,
            subtask_title=t_title,
            subtask_description=t_desc
        )
        
        try:
            finding = subagent.run(prompt)
            current_findings.append(f"FINDINGS FOR TASK {t_id}: {t_title}\n\n{finding}")
            log_content += f"✅ {t_title} complete!\n\n"
            yield log_content, state, gr.update(), ""
        except Exception as e:
            log_content += f"❌ {t_title} failed: {e}\n\n"
            yield log_content, state, gr.update(), ""

    log_content += "### ✨ Synthesis: Generating Final Report...\n"
    yield log_content, state, gr.update(), ""
    
    sys_prompt = COORDINATOR_DIRECTION.format(
        user_query=state.get("final_topic", ""),
        research_plan=state.get("research_plan", ""),
        subtasks_json=json.dumps(subtasks, indent=2)
    )
    user_prompt = f"Synthesize these findings:\n\n" + "\n\n".join(current_findings)
    
    try:
        response = coordinator_model(messages=[
            {"role": "system", "content": sys_prompt},
            {"role": "user", "content": user_prompt}
        ])
        final_report = response.content
        if "<think>" in final_report and "</think>" in final_report:
            final_report = final_report.split("</think>")[-1].strip()
        
        log_content += "### 🖋️ Review: Polishing and Finalizing...\n"
        yield log_content, state, gr.update(), ""
        
        reviewer = Reviewer(model_name=REVIEWER_MODEL, hf_key=clean_hf)
        polished_report = reviewer.review(final_report)
        state["final_report"] = polished_report
        
        os.makedirs("temp_outputs", exist_ok=True)
        ts = int(time.time())
        pdf_name = f"research_{ts}.pdf"
        pdf_path = os.path.join("temp_outputs", pdf_name)
        if reviewer.generate_pdf(polished_report, pdf_path):
            state["pdf_path"] = pdf_path
            
        md_name = f"research_{ts}.md"
        md_path = os.path.join("temp_outputs", md_name)
        with open(md_path, "w", encoding="utf-8") as f:
            f.write(polished_report)
        state["md_path"] = md_path
            
        yield log_content + "✅ Research mission accomplished!", state, gr.update(visible=True), polished_report
    except Exception as e:
        yield log_content + f"❌ Synthesis failed: {e}", state, gr.update(), ""

def reset_all():
    return (
        gr.update(value="", visible=True), 
        gr.update(visible=False), 
        gr.update(visible=False), 
        gr.update(visible=False), 
        gr.update(visible=False), 
        {}, 
        "### 🧬 Status\nReady to research.", 
        ""
    )

with gr.Blocks(theme=theme, title="Deep Research Agent") as demo:
    state = gr.State({})
    
    gr.Markdown("# 🧬 Deep Research Agent")
    gr.Markdown("### The ultimate AI research pipeline that browses the web for you.")
    
    with gr.Row():
        with gr.Column(scale=1, variant="panel"):
            gr.Markdown("## ⚙️ Configuration")
            hf_key_input = gr.Textbox(label="Hugging Face Token", type="password", value=HF_KEY)
            tavily_key_input = gr.Textbox(label="Tavily API Key", type="password", value=TAVILY_API_KEY)
            reset_btn = gr.Button("🔄 New Research", variant="secondary")
            gr.Markdown("---")
            log_output = gr.Markdown("### 🧬 Status\nReady to research.")

        with gr.Column(scale=4):
            # STEP 1: Introduction
            with gr.Column(visible=True) as step1_col:
                gr.Markdown("## 1️⃣ What are you researching?")
                topic_input = gr.Textbox(label="Enter a broad topic or research question:", placeholder="e.g., Target market for sustainable polymers in Europe")
                start_btn = gr.Button("Clarify Topic ➡️", variant="primary")

            # STEP 2: Refinement
            with gr.Column(visible=False) as step2_col:
                gr.Markdown("## 2️⃣ Refine Your Topic")
                suggestion_display = gr.Markdown()
                with gr.Row():
                    opt_index = gr.Dropdown(choices=["1", "2", "3"], label="Select Option (Optional)")
                    custom_topic_input = gr.Textbox(label="Or type a custom refined topic:")
                refine_btn = gr.Button("Set Strategic Topic 🎯", variant="primary")

            # STEP 3: Planning & Splitting
            with gr.Column(visible=False) as step3_col:
                gr.Markdown("## 3️⃣ Strategy & Task Splitting")
                target_display = gr.Markdown()
                strat_btn = gr.Button("Generate Strategy 📋", variant="primary")
                strategy_display = gr.Markdown()
                
                split_btn = gr.Button("Decompose into Subtasks 🧩", variant="primary", visible=False)
                tasks_display = gr.Markdown()
                
                execute_btn = gr.Button("🚀 Launch Research Agents", variant="primary", visible=False)

            # STEP 4: Execution
            with gr.Column(visible=False) as step4_col:
                gr.Markdown("## 4️⃣ Agentic Research in Progress")
                gr.Markdown("Monitoring agent fleet... check the status panel on the left.")

            # STEP 5: Results
            with gr.Column(visible=False) as step5_col:
                gr.Markdown("## 🏁 Final Research Report")
                final_md_display = gr.Markdown()
                with gr.Row():
                    download_md = gr.File(label="Download Markdown")
                    download_pdf = gr.File(label="Download PDF")
                new_research_btn = gr.Button("Start Over", variant="primary")

    # --- Callbacks ---
    
    start_btn.click(
        start_clarification, 
        inputs=[topic_input, hf_key_input, state], 
        outputs=[step2_col, suggestion_display, state, step1_col]
    )
    
    refine_btn.click(
        select_suggestion,
        inputs=[opt_index, custom_topic_input, state],
        outputs=[step3_col, target_display, state, step2_col]
    )
    
    strat_btn.click(
        generate_strategy,
        inputs=[hf_key_input, state],
        outputs=[strategy_display, state, split_btn]
    )
    
    split_btn.click(
        decompose_tasks,
        inputs=[hf_key_input, state],
        outputs=[tasks_display, state, execute_btn]
    )
    
    execute_btn.click(
        lambda: gr.update(visible=True), outputs=step4_col
    ).then(
        run_research,
        inputs=[hf_key_input, tavily_key_input, state],
        outputs=[log_output, state, step5_col, final_md_display]
    ).then(
        lambda s: (s.get("md_path"), s.get("pdf_path")),
        inputs=[state],
        outputs=[download_md, download_pdf]
    )

    reset_btn.click(
        reset_all,
        outputs=[topic_input, step2_col, step3_col, step4_col, step5_col, state, log_output, final_md_display]
    )
    
    new_research_btn.click(
        reset_all,
        outputs=[topic_input, step2_col, step3_col, step4_col, step5_col, state, log_output, final_md_display]
    )

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