app.py

"""
MAKER Agent - Chat Interface
=============================
Reliable AI Agent with Web Search & File Upload
Based on: https://arxiv.org/abs/2511.09030
"""

import gradio as gr
import asyncio
import json
import re
import base64
from collections import Counter
from dataclasses import dataclass, field
from typing import Any, Callable, Optional
from pathlib import Path

# ============================================================================
# MAKER Core (Embedded)
# ============================================================================

@dataclass
class VotingConfig:
    k: int = 3
    max_samples: int = 30
    temperature_first: float = 0.0
    temperature_rest: float = 0.1
    parallel_samples: int = 3

@dataclass
class RedFlagConfig:
    max_response_chars: int = 3000
    min_response_length: int = 5
    banned_patterns: list = field(default_factory=lambda: [])


class LLMClient:
    """Universal LLM client."""
    
    def __init__(self, provider: str, api_key: str, model: str = None):
        self.provider = provider.lower()
        self.api_key = api_key
        self.model = model
        self._client = None
        self._setup_client()
    
    def _setup_client(self):
        if self.provider == "openai":
            from openai import AsyncOpenAI
            self._client = AsyncOpenAI(api_key=self.api_key)
            self.model = self.model or "gpt-4o-mini"
        elif self.provider == "anthropic":
            from anthropic import AsyncAnthropic
            self._client = AsyncAnthropic(api_key=self.api_key)
            self.model = self.model or "claude-sonnet-4-20250514"
        elif self.provider == "groq":
            from openai import AsyncOpenAI
            self._client = AsyncOpenAI(api_key=self.api_key, base_url="https://api.groq.com/openai/v1")
            self.model = self.model or "llama-3.3-70b-versatile"
        elif self.provider == "together":
            from openai import AsyncOpenAI
            self._client = AsyncOpenAI(api_key=self.api_key, base_url="https://api.together.xyz/v1")
            self.model = self.model or "meta-llama/Llama-3.3-70B-Instruct-Turbo"
        elif self.provider == "openrouter":
            from openai import AsyncOpenAI
            self._client = AsyncOpenAI(api_key=self.api_key, base_url="https://openrouter.ai/api/v1")
            self.model = self.model or "openai/gpt-4o-mini"
    
    async def generate(self, prompt: str, temperature: float = 0.0, max_tokens: int = 2000) -> str:
        if self.provider == "anthropic":
            r = await self._client.messages.create(
                model=self.model, max_tokens=max_tokens,
                messages=[{"role": "user", "content": prompt}]
            )
            return r.content[0].text
        else:
            r = await self._client.chat.completions.create(
                model=self.model,
                messages=[{"role": "user", "content": prompt}],
                temperature=temperature, max_tokens=max_tokens
            )
            return r.choices[0].message.content


class WebSearch:
    """Web search using DuckDuckGo (free)."""
    
    @staticmethod
    async def search(query: str, num_results: int = 5) -> list:
        try:
            from duckduckgo_search import DDGS
            results = []
            with DDGS() as ddgs:
                for r in ddgs.text(query, max_results=num_results):
                    results.append({
                        "title": r.get("title", ""),
                        "url": r.get("href", ""),
                        "snippet": r.get("body", "")
                    })
            return results
        except Exception as e:
            return [{"title": "Error", "url": "", "snippet": str(e)}]


class FileHandler:
    """Handle file uploads."""
    
    @staticmethod
    async def load_file(file_path: str) -> dict:
        path = Path(file_path)
        ext = path.suffix.lower()
        
        try:
            if ext in {'.txt', '.md', '.json', '.py', '.js', '.html', '.css', '.csv'}:
                content = path.read_text(encoding='utf-8', errors='replace')
                return {"type": "text", "name": path.name, "content": content[:50000]}
            
            elif ext == '.pdf':
                try:
                    import pymupdf
                    doc = pymupdf.open(str(path))
                    text = "\n\n".join([page.get_text() for page in doc])
                    doc.close()
                    return {"type": "pdf", "name": path.name, "content": text[:50000]}
                except ImportError:
                    return {"type": "error", "name": path.name, "content": "PDF requires: pip install pymupdf"}
            
            elif ext == '.docx':
                try:
                    from docx import Document
                    doc = Document(str(path))
                    text = "\n\n".join([p.text for p in doc.paragraphs])
                    return {"type": "docx", "name": path.name, "content": text[:50000]}
                except ImportError:
                    return {"type": "error", "name": path.name, "content": "DOCX requires: pip install python-docx"}
            
            elif ext in {'.png', '.jpg', '.jpeg', '.gif', '.webp'}:
                content = path.read_bytes()
                b64 = base64.b64encode(content).decode('utf-8')
                return {"type": "image", "name": path.name, "base64": b64}
            
            else:
                content = path.read_text(encoding='utf-8', errors='replace')
                return {"type": "text", "name": path.name, "content": content[:50000]}
        
        except Exception as e:
            return {"type": "error", "name": path.name, "content": str(e)}


class MAKERAgent:
    """MAKER Framework Agent."""
    
    def __init__(self, llm: LLMClient, voting: VotingConfig = None, red_flags: RedFlagConfig = None):
        self.llm = llm
        self.voting = voting or VotingConfig()
        self.red_flags = red_flags or RedFlagConfig()
        self.stats = {"samples": 0, "red_flags": 0, "tool_calls": 0}
    
    def _check_red_flags(self, response: str) -> bool:
        if len(response) > self.red_flags.max_response_chars:
            return True
        if len(response) < self.red_flags.min_response_length:
            return True
        for pattern in self.red_flags.banned_patterns:
            if re.search(pattern, response, re.IGNORECASE):
                return True
        return False
    
    def _normalize_response(self, response: str) -> str:
        """Normalize response for voting comparison."""
        return response.strip().lower()
    
    async def execute(self, prompt: str, use_search: bool = False,
                      file_context: str = None, progress_callback: Callable = None) -> dict:
        
        # Build the full prompt
        full_prompt = "You are a helpful assistant. Respond naturally and conversationally.\n\n"
        
        if file_context:
            full_prompt += f"The user has provided the following files for context:\n{file_context}\n\n"
        
        full_prompt += f"User: {prompt}\n\nAssistant:"
        
        # Handle web search if enabled
        search_results = None
        if use_search:
            if progress_callback:
                progress_callback(0.1, "Searching the web...")
            
            search_results = await WebSearch.search(prompt)
            self.stats["tool_calls"] += 1
            
            if search_results and search_results[0].get("title") != "Error":
                search_text = "\n".join([f"- {r['title']}: {r['snippet']}" for r in search_results[:5]])
                full_prompt = f"You are a helpful assistant with access to web search results.\n\n"
                if file_context:
                    full_prompt += f"Files provided:\n{file_context}\n\n"
                full_prompt += f"Web search results for '{prompt}':\n{search_text}\n\n"
                full_prompt += f"User question: {prompt}\n\nProvide a helpful response based on the search results. Assistant:"
        
        if progress_callback:
            progress_callback(0.2, "Getting response...")
        
        # Voting loop
        votes: Counter = Counter()
        responses_map = {}
        samples, flagged = 0, 0
        
        # First sample at temperature 0
        response = await self.llm.generate(full_prompt, temperature=0.0)
        samples += 1
        self.stats["samples"] += 1
        
        if not self._check_red_flags(response):
            key = self._normalize_response(response)
            votes[key] += 1
            responses_map[key] = response
        else:
            flagged += 1
            self.stats["red_flags"] += 1
        
        # Continue voting until we have a winner
        round_num = 1
        while samples < self.voting.max_samples:
            if votes:
                top = votes.most_common(2)
                top_count = top[0][1]
                second_count = top[1][1] if len(top) > 1 else 0
                if top_count - second_count >= self.voting.k:
                    break
            
            round_num += 1
            if progress_callback:
                progress_callback(0.2 + 0.7 * (samples / self.voting.max_samples), f"Voting round {round_num}...")
            
            for _ in range(self.voting.parallel_samples):
                if samples >= self.voting.max_samples:
                    break
                
                response = await self.llm.generate(full_prompt, temperature=self.voting.temperature_rest)
                samples += 1
                self.stats["samples"] += 1
                
                if not self._check_red_flags(response):
                    key = self._normalize_response(response)
                    votes[key] += 1
                    if key not in responses_map:
                        responses_map[key] = response
                else:
                    flagged += 1
                    self.stats["red_flags"] += 1
        
        if progress_callback:
            progress_callback(1.0, "Done!")
        
        if votes:
            top_key, top_count = votes.most_common(1)[0]
            return {
                "success": True,
                "response": responses_map[top_key],
                "votes": top_count,
                "total_samples": samples,
                "red_flagged": flagged,
                "search_results": search_results
            }
        
        return {
            "success": False,
            "response": "I couldn't generate a reliable response. Please try again.",
            "votes": 0,
            "total_samples": samples,
            "red_flagged": flagged,
            "search_results": search_results
        }


# ============================================================================
# Global State
# ============================================================================

current_agent = None
loaded_files = {}

# ============================================================================
# Functions
# ============================================================================

def setup_agent(provider, api_key, model, k_votes):
    global current_agent
    if not api_key:
        return "❌ Please enter your API key", gr.update(interactive=False)
    try:
        llm = LLMClient(provider, api_key, model if model else None)
        current_agent = MAKERAgent(llm, VotingConfig(k=k_votes))
        return f"✅ Connected to {provider} ({llm.model})", gr.update(interactive=True)
    except Exception as e:
        return f"❌ Error: {e}", gr.update(interactive=False)


def process_files(files):
    global loaded_files
    loaded_files = {}
    
    if not files:
        return "No files attached"
    
    names = []
    for f in files:
        info = asyncio.run(FileHandler.load_file(f.name))
        loaded_files[info['name']] = info
        names.append(info['name'])
    
    return f"📎 {', '.join(names)}"


async def chat_async(message, history, use_search, files, progress=gr.Progress()):
    global current_agent, loaded_files
    
    if not current_agent:
        return history + [[message, "⚠️ Please set up your API key first in the Settings tab."]]
    
    # Process any new files
    if files:
        for f in files:
            info = await FileHandler.load_file(f.name)
            loaded_files[info['name']] = info
    
    # Build file context
    file_context = None
    if loaded_files:
        parts = []
        for name, info in loaded_files.items():
            if info["type"] != "image" and info["type"] != "error":
                parts.append(f"=== {name} ===\n{info.get('content', '')[:10000]}")
        if parts:
            file_context = "\n\n".join(parts)
    
    def update_progress(pct, msg):
        progress(pct, desc=msg)
    
    try:
        result = await current_agent.execute(
            message, 
            use_search=use_search,
            file_context=file_context,
            progress_callback=update_progress
        )
        
        response = result["response"]
        
        # Add subtle stats footer
        stats = f"\n\n---\n*{result['votes']} votes, {result['total_samples']} samples*"
        
        return history + [[message, response + stats]]
    
    except Exception as e:
        return history + [[message, f"❌ Error: {str(e)}"]]


def chat(message, history, use_search, files):
    return asyncio.run(chat_async(message, history, use_search, files))


def clear_chat():
    global loaded_files
    loaded_files = {}
    return [], None, "No files attached"


# ============================================================================
# UI
# ============================================================================

with gr.Blocks(title="MAKER Agent") as demo:
    
    # Header
    gr.HTML("""
        <div style="text-align: center; padding: 20px 0 10px 0;">
            <h1 style="font-size: 2rem; margin: 0;">🔧 MAKER Agent</h1>
            <p style="color: #666; margin: 5px 0;">Reliable AI with Voting • <a href="https://arxiv.org/abs/2511.09030" target="_blank">Paper</a></p>
        </div>
    """)
    
    with gr.Tabs():
        
        # Chat Tab
        with gr.Tab("💬 Chat"):
            
            chatbot = gr.Chatbot(
                height=450,
            )
            
            with gr.Row():
                with gr.Column(scale=12):
                    msg = gr.Textbox(
                        placeholder="Ask anything...",
                        show_label=False,
                        lines=2,
                    )
                
                with gr.Column(scale=1, min_width=80):
                    send_btn = gr.Button("Send", variant="primary", interactive=False)
            
            with gr.Row():
                with gr.Column(scale=4):
                    file_upload = gr.File(
                        label="",
                        file_count="multiple",
                        file_types=[".pdf", ".docx", ".txt", ".md", ".json", ".csv"],
                        show_label=False,
                    )
                
                with gr.Column(scale=2):
                    file_status = gr.Markdown("No files attached")
                
                with gr.Column(scale=2):
                    use_search = gr.Checkbox(
                        label="🔍 Web Search",
                        value=False,
                        info="Search DuckDuckGo"
                    )
                
                with gr.Column(scale=1):
                    clear_btn = gr.Button("🗑️ Clear")
            
            # Event handlers
            file_upload.change(process_files, file_upload, file_status)
            
            msg.submit(chat, [msg, chatbot, use_search, file_upload], chatbot).then(
                lambda: "", None, msg
            )
            send_btn.click(chat, [msg, chatbot, use_search, file_upload], chatbot).then(
                lambda: "", None, msg
            )
            clear_btn.click(clear_chat, None, [chatbot, file_upload, file_status])
        
        # Settings Tab
        with gr.Tab("⚙️ Settings"):
            
            gr.Markdown("### Connect to an LLM Provider")
            
            with gr.Row():
                with gr.Column():
                    provider = gr.Dropdown(
                        ["groq", "openai", "anthropic", "together", "openrouter"],
                        value="groq",
                        label="Provider",
                        info="Groq is free & fast!"
                    )
                    api_key = gr.Textbox(
                        label="API Key",
                        type="password",
                        placeholder="Paste your API key here..."
                    )
                    model = gr.Textbox(
                        label="Model (optional)",
                        placeholder="Leave blank for default"
                    )
                
                with gr.Column():
                    k_votes = gr.Slider(
                        1, 7, value=3, step=1,
                        label="Reliability (K votes)",
                        info="Higher = more reliable, slower"
                    )
                    
                    gr.Markdown("""
                    ### Get API Keys
                    
                    **Groq** (recommended - free & fast):
                    [console.groq.com](https://console.groq.com)
                    
                    **OpenAI**: [platform.openai.com/api-keys](https://platform.openai.com/api-keys)
                    
                    **Anthropic**: [console.anthropic.com](https://console.anthropic.com)
                    """)
            
            connect_btn = gr.Button("🔌 Connect", variant="primary")
            status = gr.Markdown("👆 Enter your API key and click Connect")
            
            connect_btn.click(
                setup_agent,
                [provider, api_key, model, k_votes],
                [status, send_btn]
            )
        
        # About Tab
        with gr.Tab("ℹ️ About"):
            gr.Markdown("""
## How MAKER Works

This agent uses the **MAKER Framework** to achieve reliable AI responses:

1. **Multiple Samples** - Generates several responses for each question
2. **Voting** - Responses "vote" and the winner needs K votes ahead
3. **Red-Flagging** - Suspicious outputs are automatically discarded

### Why This Matters

Instead of hoping the AI gets it right, MAKER uses statistics to ensure reliability. The paper achieved **1 million steps with zero errors** using this approach.

### Features

- 🔍 **Web Search** - Free DuckDuckGo search (no API key needed)
- 📎 **File Upload** - PDF, DOCX, TXT, MD, JSON, CSV
- ⚡ **Multiple Providers** - Groq, OpenAI, Anthropic, and more

### Links

- 📄 [Research Paper](https://arxiv.org/abs/2511.09030)
- 🎥 [Video Explanation](https://youtube.com/watch?v=TJ-vWGCosdQ)
""")
    
    # Footer
    gr.HTML("""
        <div style="text-align: center; color: #888; padding: 15px; font-size: 0.85rem;">
            MAKER Framework • <a href="https://arxiv.org/abs/2511.09030" style="color: #888;">arxiv.org/abs/2511.09030</a>
        </div>
    """)

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