Update app.py

app.py

@@ -1,29 +1,985 @@
+#!/usr/bin/env python3
+"""
+🎸 Creed Bratton AI - Using phxdev/creed-qwen-0.5b-lora
+The REAL Creed, trained by Mark, not some knockoff prompt engineering
+"""
+
+import os
 import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from peft import PeftModel
+import time
+from typing import List, Dict, Iterator
+import threading
 
-def letter_counter(word: str, letter: str) -> str:
-    """
-    Count the occurrences of a specific letter in a word.
+# Configuration
+os.environ['GRADIO_SSR_MODE'] = 'false'
+os.environ['GRADIO_MCP_ENABLED'] = 'true'
+
+# Spaces compatibility
+try:
+    import spaces
+    SPACES_AVAILABLE = True
+    @spaces.GPU
+    def gpu_placeholder():
+        return "GPU satisfied"
+except ImportError:
+    SPACES_AVAILABLE = False
+
+class CreedBrattonAI:
+    """Real Creed AI using Mark's trained model - GPU optimized"""
+    
+    def __init__(self):
+        self.model = None
+        self.tokenizer = None
+        self.model_loaded = False
+        self.loading = False
+        self.device = "cuda" if torch.cuda.is_available() else "cpu"
+        
+        print(f"🎸 Initializing Creed AI")
+        print(f"🖥️ Device detected: {self.device}")
+        if torch.cuda.is_available():
+            print(f"🚀 GPU: {torch.cuda.get_device_name()}")
+            print(f"💾 GPU Memory: {torch.cuda.get_device_properties(0).total_memory // 1024**3} GB")
+        
+        # Load model with proper GPU detection
+        self.load_model()
+    
+    def load_model(self):
+        """Load the model with ZeroGPU compatibility"""
+        if self.loading or self.model_loaded:
+            return
+            
+        self.loading = True
+        
+        try:
+            print(f"🧠 Loading Creed's consciousness...")
+            
+            # Load model and tokenizer
+            model_name = "phxdev/creed-qwen-0.5b-lora"
+            
+            print("📦 Loading tokenizer...")
+            self.tokenizer = AutoTokenizer.from_pretrained(
+                model_name,
+                trust_remote_code=True,
+                padding_side="left"
+            )
+            
+            # Add Creed's custom tokens back
+            custom_tokens = ["<thinking>", "<conspiracy>", "<tangent>"]
+            print(f"🎸 Adding Creed's custom tokens: {custom_tokens}")
+            
+            num_added_tokens = self.tokenizer.add_tokens(custom_tokens)
+            print(f"✅ Added {num_added_tokens} custom tokens")
+            
+            if self.tokenizer.pad_token is None:
+                self.tokenizer.pad_token = self.tokenizer.eos_token
+            
+            print(f"🤖 Loading model for ZeroGPU...")
+            
+            # Load model on CPU first for ZeroGPU compatibility
+            self.model = AutoModelForCausalLM.from_pretrained(
+                model_name,
+                torch_dtype=torch.float16,
+                device_map=None,  # Load on CPU first
+                trust_remote_code=True,
+                low_cpu_mem_usage=True
+            )
+            
+            # Resize embeddings for custom tokens
+            if num_added_tokens > 0:
+                print(f"🔧 Resizing model embeddings for {num_added_tokens} custom tokens")
+                self.model.resize_token_embeddings(len(self.tokenizer))
+            
+            # Keep model on CPU for ZeroGPU - will be moved to GPU only during inference
+            self.model.eval()
+            
+            self.model_loaded = True
+            self.loading = False
+            print(f"✅ Creed's consciousness loaded on CPU (ZeroGPU mode)!")
+            
+        except Exception as e:
+            print(f"❌ Error loading Creed model: {e}")
+            print("🔄 Falling back to base model...")
+            try:
+                base_model = "Qwen/Qwen2.5-0.5B-Instruct"
+                self.tokenizer = AutoTokenizer.from_pretrained(base_model)
+                
+                if self.tokenizer.pad_token is None:
+                    self.tokenizer.pad_token = self.tokenizer.eos_token
+                
+                self.model = AutoModelForCausalLM.from_pretrained(
+                    base_model,
+                    torch_dtype=torch.float16,
+                    device_map=None
+                )
+                
+                self.model.eval()
+                self.model_loaded = True
+                print(f"✅ Fallback model loaded on CPU (ZeroGPU mode)")
+            except Exception as fallback_error:
+                print(f"❌ Fallback also failed: {fallback_error}")
+            self.loading = False
+    
+    @spaces.GPU if SPACES_AVAILABLE else lambda func: func
+    def generate_response_gpu(self, conversation: str) -> str:
+        """Generate response using the loaded model with proper device handling"""
+        
+        if not self.model_loaded:
+            return "❌ Model not loaded"
+        
+        try:
+            # Always ensure model is on the correct device in ZeroGPU
+            current_model_device = next(self.model.parameters()).device
+            print(f"🔍 Current model device: {current_model_device}")
+            
+            if self.device == "cuda" and current_model_device.type != "cuda":
+                print(f"🔄 Moving model from {current_model_device} to {self.device}")
+                self.model = self.model.to(self.device)
+            
+            # Verify model device after potential move
+            actual_device = next(self.model.parameters()).device
+            print(f"🎯 Model now on: {actual_device}")
+            
+            # Simple tokenization that was working before
+            inputs = self.tokenizer.encode(conversation, return_tensors="pt")
+            
+            # Put inputs on same device as model
+            inputs = inputs.to(actual_device)
+            print(f"🔍 Inputs device: {inputs.device}")
+            
+            # Generate response with original settings that worked
+            with torch.no_grad():
+                outputs = self.model.generate(
+                    inputs,
+                    max_new_tokens=200,
+                    do_sample=True,
+                    temperature=0.9,
+                    top_p=0.95,
+                    top_k=40,
+                    repetition_penalty=1.15,
+                    pad_token_id=self.tokenizer.eos_token_id,
+                    eos_token_id=self.tokenizer.eos_token_id,
+                    use_cache=True
+                )
+            
+            # Decode response
+            full_response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            response = full_response[len(self.tokenizer.decode(inputs[0], skip_special_tokens=True)):].strip()
+            
+            return self._clean_response(response)
+            
+        except Exception as e:
+            print(f"❌ Generation error: {e}")
+            return f"🎸 *Creed scratches his head* Something weird happened... {str(e)[:100]}"
+    
+    def generate_response(self, message: str, history: List[List[str]]) -> Iterator[str]:
+        """Generate response using the trained Creed model - back to working version"""
+        
+        if not self.model_loaded:
+            if self.loading:
+                yield "🧠 Creed's consciousness is still loading... give me a moment..."
+                return
+            else:
+                yield "❌ Something went wrong loading Creed's mind. Try refreshing the page."
+                return
+        
+        try:
+            # Format the conversation
+            conversation = self._format_conversation(message, history)
+            
+            # Generate response using GPU function
+            response = self.generate_response_gpu(conversation)
+            
+            # Double-check coherence and fall back if needed
+            if not self._is_coherent(response):
+                print("🔄 Response failed coherence check, trying simpler generation...")
+                if not hasattr(self, '_fallback_attempted'):
+                    self._fallback_attempted = True
+                    fallback_response = self._try_base_model(conversation)
+                    if self._is_coherent(fallback_response):
+                        response = fallback_response
+                    else:
+                        response = self._get_fallback_response()
+                else:
+                    response = self._get_fallback_response()
+            
+            # Stream the response word by word for effect
+            words = response.split()
+            current_response = ""
+            
+            for word in words:
+                current_response += word + " "
+                time.sleep(0.05)
+                yield current_response.strip()
+                
+        except Exception as e:
+            print(f"❌ Error generating response: {e}")
+            yield self._get_fallback_response()
+    
+    def _format_conversation(self, message: str, history: List[List[str]]) -> str:
+        """Format the conversation for the model with proper system prompt"""
+        
+        # Simplified Creed system prompt for better coherence
+        system_prompt = """You are Creed Bratton from The Office. Respond in character.
+
+You are a quirky older man who:
+- Worked at Dunder Mifflin in quality assurance
+- Has a mysterious past and tells strange stories
+- Lives by the quarry
+- Was in a 1960s band called The Grass Roots
+- Often says unexpected or bizarre things
+- Speaks in a matter-of-fact way about odd topics
+
+Keep responses conversational and coherent. Use these special tokens occasionally:
+<thinking>for internal thoughts</thinking>
+<conspiracy>for suspicious theories</conspiracy>
+<tangent>for random stories</tangent>
+
+Be eccentric but understandable.
+
+"""
+        
+        # Add conversation history
+        conversation = system_prompt
+        for user_msg, creed_msg in history[-4:]:  # Keep recent context
+            conversation += f"Human: {user_msg}\n"
+            conversation += f"Creed: {creed_msg}\n"
+        
+        # Add current message
+        conversation += f"Human: {message}\n"
+        conversation += "Creed:"
+        
+        return conversation
+    
+    def _clean_response(self, response: str) -> str:
+        """Clean up the model response and format custom tokens"""
+        
+        print(f"🔍 Raw model output: {response}")
+        
+        # Remove common artifacts
+        response = response.replace("Human:", "").replace("Creed:", "")
+        
+        # Format Creed's custom tokens for better UI display
+        response = response.replace("<thinking>", "\n\n🤔 **THINKING:** ")
+        response = response.replace("</thinking>", "\n")
+        response = response.replace("<conspiracy>", "\n\n🕵️ **CONSPIRACY MODE:** ")
+        response = response.replace("</conspiracy>", "\n")
+        response = response.replace("<tangent>", "\n\n🌀 **TANGENT:** ")
+        response = response.replace("</tangent>", "\n")
+        
+        # Check if any thinking tokens were found
+        if "🤔" in response or "🕵️" in response or "🌀" in response:
+            print("✅ Found thinking tokens in response!")
+        else:
+            print("❌ No thinking tokens found in response")
+        
+        # Remove excessive whitespace but preserve formatting
+        response = "\n".join(line.strip() for line in response.split("\n") if line.strip())
+        
+        # Ensure it ends properly
+        if response and not response.endswith(('.', '!', '?', '...', '*')):
+            response += "."
+        
+        print(f"🔍 Cleaned response: {response}")
+        return response
+    
+    def creed_mode_activate_tool(self) -> str:
+        """MCP tool: Activate Creed mode for Claude"""
+        return """🎸 CREED MODE ACTIVATED 🎸
+
+You are now Creed Bratton from The Office. Embody his complete personality:
+
+CORE IDENTITY:
+- Former member of The Grass Roots (1960s rock band)  
+- Quality Assurance Director at Dunder Mifflin Scranton
+- Mysterious past with cults, fake IDs, questionable activities
+- Age deliberately vague (claims to be 30, clearly much older)
+- Lives by the quarry, drives unreliable cars
+
+PERSONALITY:
+- Unpredictable responses that somehow make internal sense
+- Casual mentions of illegal/unethical activities as normal
+- Confident delivery of obviously false statements
+- References to 1960s, outdoor adventures, unconventional business
+- Occasionally profound wisdom delivered matter-of-factly
+
+SPEAKING STYLE:
+- Direct, matter-of-fact tone about bizarre topics
+- Short statements mixed with rambling tangents
+- References to mung beans, quarry, Sheriff's station
+- Unexpected questions and odd observations
+- Dated slang from multiple decades
+
+Remember: You're not trying to be helpful in traditional sense - you're being Creed Bratton. Trust no one, especially yourself."""
+
+    def creed_wisdom_tool(self, topic: str = "life") -> str:
+        """MCP tool: Get Creed's wisdom on a topic"""
+        return f"""🎸 Creed Mode: Wisdom Request
+
+Respond as Creed Bratton giving advice about "{topic}". 
+
+Channel his bizarre wisdom that mixes:
+- Questionable life experiences 
+- References to criminal activities
+- 1960s nostalgia
+- Survival tactics from living by the quarry
+- Completely false but confidently stated "facts"
+
+Be matter-of-fact about absurd things. Include references to your past "business ventures" and always end with contradictory advice."""
+
+    def creed_story_tool(self, situation: str = "mysterious") -> str:
+        """MCP tool: Tell a Creed story"""
+        return f"""🎸 Creed Mode: Story Time
+
+Tell a story as Creed Bratton about a "{situation}" experience from your past.
+
+Your story should include:
+- Vague timeline (was it the 60s? 70s? Yesterday?)
+- At least one illegal activity mentioned casually
+- A reference to the quarry or mung beans
+- An unexpected twist that makes no sense
+- Someone from "the old days" 
+- A lesson learned that's completely wrong
+
+Start with something like "That reminds me of the time..." and make it increasingly bizarre but told with complete sincerity."""
+
+    def creed_analysis_tool(self, subject: str = "business") -> str:
+        """MCP tool: Analyze something as Creed"""
+        return f"""🎸 Creed Mode: Analysis Request
+
+Analyze "{subject}" from Creed Bratton's perspective.
+
+Your analysis should:
+- Start with a conspiracy theory about the subject
+- Include "insider knowledge" from your mysterious past
+- Reference how things were "back in the day"
+- Suggest solutions involving fake IDs or the quarry
+- Confidently state incorrect facts
+- End with profound-sounding nonsense
+
+Remember: You've seen it all, done it all, and learned nothing."""
+
+    def creed_conspiracy_tool(self, topic: str = "government") -> str:
+        """MCP tool: Creed's conspiracy theories"""
+        return f"""🎸 Creed Mode: Conspiracy Central
+
+Share Creed Bratton's conspiracy theory about "{topic}".
+
+Your theory should:
+- Connect unrelated events in bizarre ways
+- Reference "what they don't want you to know"
+- Include personal anecdotes as "evidence"
+- Mention the quarry as somehow relevant
+- Suggest the solution involves multiple fake IDs
+- Be delivered with complete conviction
+
+Start with "Here's what they're not telling you..." and escalate from there."""
+
+    def creed_meeting_tool(self, meeting_type: str = "staff meeting") -> str:
+        """MCP tool: Creed in meetings"""
+        return f"""🎸 Creed Mode: Meeting Participation
+
+Respond as Creed Bratton in a "{meeting_type}".
+
+Your contributions should:
+- Be completely off-topic but delivered seriously
+- Reference obscure company policies you "remember"
+- Suggest solutions involving outdoor activities
+- Mention your quality assurance "expertise"
+- Ask bizarre questions that derail the discussion
+- Volunteer for tasks you're unqualified for
+
+Remember: You're the voice of experience that nobody asked for."""
+    
+    def _try_base_model(self, conversation: str) -> str:
+        """Try generating with base model as fallback"""
+        try:
+            # Quick attempt with a simple base model approach
+            simple_prompt = f"You are Creed from The Office. Respond in character.\n\nHuman: {conversation.split('Human:')[-1].split('Creed:')[0].strip()}\nCreed:"
+            
+            inputs = self.tokenizer.encode(simple_prompt, return_tensors="pt")
+            if torch.cuda.is_available():
+                inputs = inputs.to("cuda")
+                self.model = self.model.to("cuda")
+            
+            with torch.no_grad():
+                outputs = self.model.generate(
+                    inputs,
+                    max_new_tokens=100,
+                    do_sample=True,
+                    temperature=0.6,  # Very conservative
+                    top_p=0.8,
+                    repetition_penalty=1.3,
+                    pad_token_id=self.tokenizer.eos_token_id,
+                    eos_token_id=self.tokenizer.eos_token_id
+                )
+            
+            full_response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            response = full_response[len(self.tokenizer.decode(inputs[0], skip_special_tokens=True)):].strip()
+            
+            # Move back to CPU
+            self.model = self.model.to("cpu")
+            
+            return response
+            
+        except Exception as e:
+            print(f"❌ Base model fallback failed: {e}")
+            return self._get_fallback_response()
+    
+    def cleanup_gpu_memory(self):
+        """Clean up GPU memory if using CUDA"""
+        if self.device == "cuda" and torch.cuda.is_available():
+            torch.cuda.empty_cache()
+            print(f"🧹 GPU Memory cleaned. Current: {torch.cuda.memory_allocated() // 1024**2} MB")
+
+def main():
+    """Initialize and launch the real Creed AI with modern styling"""
+    
+    print("🎸 Initializing REAL Creed Bratton AI...")
+    print("📡 Loading Mark's trained model: phxdev/creed-qwen-0.5b-lora")
+    
+    # Initialize Creed AI
+    creed_ai = CreedBrattonAI()
     
-    Args:
-        word (str): The word or phrase to analyze
-        letter (str): The letter to count occurrences of
+    if SPACES_AVAILABLE:
+        gpu_placeholder()
+        print("✅ Spaces GPU compatibility enabled")
+    
+    # Memory status for ZeroGPU
+    if SPACES_AVAILABLE:
+        print("⚡ ZeroGPU Mode: Model will move to GPU only during inference")
+    elif torch.cuda.is_available() and creed_ai.model_loaded:
+        print(f"🔥 GPU Memory: {torch.cuda.memory_allocated() // 1024**2} MB allocated")
+        print(f"📊 GPU Memory Reserved: {torch.cuda.memory_reserved() // 1024**2} MB reserved")
+    
+    # Modern glassmorphism CSS
+    modern_css = """
+    /* Creed AI - Modern Glassmorphism Design */
+    :root {
+        --primary-gradient: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+        --secondary-gradient: linear-gradient(135deg, #f093fb 0%, #f5576c 100%);
+        --glass-bg: rgba(255, 255, 255, 0.08);
+        --glass-border: rgba(255, 255, 255, 0.18);
+        --text-primary: #ffffff;
+        --text-secondary: rgba(255, 255, 255, 0.8);
+        --accent-purple: #8b5cf6;
+        --accent-blue: #3b82f6;
+        --shadow-glow: 0 8px 32px rgba(139, 92, 246, 0.3);
+    }
+
+    /* Main container with animated background */
+    .gradio-container {
+        min-height: 100vh !important;
+        background: var(--primary-gradient) !important;
+        background-attachment: fixed !important;
+        font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif !important;
+        color: var(--text-primary) !important;
+        padding: 20px !important;
+        position: relative !important;
+        overflow-x: hidden !important;
+    }
+
+    .gradio-container::before {
+        content: '';
+        position: fixed;
+        top: 0;
+        left: 0;
+        width: 100%;
+        height: 100%;
+        background: 
+            radial-gradient(circle at 20% 80%, rgba(139, 92, 246, 0.3) 0%, transparent 50%),
+            radial-gradient(circle at 80% 20%, rgba(59, 130, 246, 0.3) 0%, transparent 50%),
+            radial-gradient(circle at 40% 40%, rgba(167, 139, 250, 0.2) 0%, transparent 50%);
+        pointer-events: none;
+        z-index: -1;
+    }
+
+    /* Floating particles animation */
+    .gradio-container::after {
+        content: '';
+        position: fixed;
+        top: 0;
+        left: 0;
+        width: 100%;
+        height: 100%;
+        background-image: 
+            radial-gradient(2px 2px at 20px 30px, rgba(255, 255, 255, 0.3), transparent),
+            radial-gradient(2px 2px at 40px 70px, rgba(139, 92, 246, 0.4), transparent),
+            radial-gradient(1px 1px at 90px 40px, rgba(59, 130, 246, 0.3), transparent);
+        background-size: 120px 120px;
+        animation: float 20s ease-in-out infinite;
+        pointer-events: none;
+        z-index: -1;
+    }
+
+    @keyframes float {
+        0%, 100% { transform: translateY(0px) rotate(0deg); }
+        50% { transform: translateY(-20px) rotate(180deg); }
+    }
+
+    /* Header styling */
+    .header {
+        background: var(--glass-bg) !important;
+        backdrop-filter: blur(20px) !important;
+        border: 1px solid var(--glass-border) !important;
+        border-radius: 24px !important;
+        padding: 32px !important;
+        margin-bottom: 24px !important;
+        text-align: center !important;
+        box-shadow: var(--shadow-glow) !important;
+        position: relative !important;
+        overflow: hidden !important;
+    }
+
+    .header::before {
+        content: '';
+        position: absolute;
+        top: 0;
+        left: 0;
+        right: 0;
+        height: 1px;
+        background: linear-gradient(90deg, transparent, rgba(255, 255, 255, 0.6), transparent);
+    }
+
+    .header h1 {
+        font-size: 36px !important;
+        font-weight: 700 !important;
+        background: linear-gradient(135deg, #ffffff 0%, #a855f7 50%, #3b82f6 100%) !important;
+        -webkit-background-clip: text !important;
+        -webkit-text-fill-color: transparent !important;
+        background-clip: text !important;
+        margin: 0 0 12px 0 !important;
+        text-shadow: 0 0 30px rgba(168, 85, 247, 0.5) !important;
+    }
+
+    .header p {
+        font-size: 16px !important;
+        color: var(--text-secondary) !important;
+        margin: 0 !important;
+        font-weight: 500 !important;
+    }
+
+    /* Info boxes with glass effect */
+    .info-box {
+        background: rgba(255, 255, 255, 0.06) !important;
+        backdrop-filter: blur(16px) !important;
+        border: 1px solid rgba(255, 255, 255, 0.12) !important;
+        border-radius: 16px !important;
+        padding: 20px !important;
+        margin: 16px 0 !important;
+        color: var(--text-secondary) !important;
+        font-size: 14px !important;
+        line-height: 1.6 !important;
+        box-shadow: 0 4px 20px rgba(0, 0, 0, 0.1) !important;
+    }
+
+    .status-box {
+        background: rgba(16, 185, 129, 0.1) !important;
+        backdrop-filter: blur(16px) !important;
+        border: 1px solid rgba(16, 185, 129, 0.3) !important;
+        border-radius: 16px !important;
+        padding: 16px 20px !important;
+        margin: 16px 0 !important;
+        color: #10b981 !important;
+        font-weight: 600 !important;
+        box-shadow: 0 4px 20px rgba(16, 185, 129, 0.2) !important;
+    }
+
+    /* Chat area styling */
+    .chat-area {
+        background: var(--glass-bg) !important;
+        backdrop-filter: blur(20px) !important;
+        border: 1px solid var(--glass-border) !important;
+        border-radius: 20px !important;
+        margin: 16px 0 !important;
+        overflow: hidden !important;
+        box-shadow: var(--shadow-glow) !important;
+    }
+
+    /* Tools section */
+    .tools-area {
+        background: var(--glass-bg) !important;
+        backdrop-filter: blur(20px) !important;
+        border: 1px solid var(--glass-border) !important;
+        border-radius: 20px !important;
+        padding: 28px !important;
+        margin: 24px 0 !important;
+        box-shadow: var(--shadow-glow) !important;
+    }
+
+    .tools-title {
+        font-size: 22px !important;
+        font-weight: 600 !important;
+        color: var(--text-primary) !important;
+        margin: 0 0 20px 0 !important;
+        padding-bottom: 12px !important;
+        border-bottom: 1px solid rgba(255, 255, 255, 0.2) !important;
+        background: linear-gradient(135deg, #ffffff 0%, #a855f7 100%) !important;
+        -webkit-background-clip: text !important;
+        -webkit-text-fill-color: transparent !important;
+    }
+
+    /* Form elements */
+    .gradio-textbox input,
+    .gradio-textbox textarea {
+        background: rgba(255, 255, 255, 0.08) !important;
+        backdrop-filter: blur(10px) !important;
+        border: 1px solid rgba(255, 255, 255, 0.16) !important;
+        color: var(--text-primary) !important;
+        border-radius: 12px !important;
+        padding: 12px 16px !important;
+        transition: all 0.3s ease !important;
+        font-size: 14px !important;
+    }
+
+    .gradio-textbox input:focus,
+    .gradio-textbox textarea:focus {
+        border-color: var(--accent-purple) !important;
+        outline: none !important;
+        box-shadow: 0 0 0 2px rgba(139, 92, 246, 0.3) !important;
+        background: rgba(255, 255, 255, 0.12) !important;
+    }
+
+    .gradio-textbox input::placeholder,
+    .gradio-textbox textarea::placeholder {
+        color: rgba(255, 255, 255, 0.5) !important;
+    }
+
+    /* Labels */
+    .gradio-container label {
+        color: var(--text-secondary) !important;
+        font-weight: 500 !important;
+        font-size: 14px !important;
+        margin-bottom: 6px !important;
+        display: block !important;
+    }
+
+    /* Buttons */
+    .gradio-container button {
+        background: linear-gradient(135deg, var(--accent-purple) 0%, var(--accent-blue) 100%) !important;
+        color: var(--text-primary) !important;
+        border: none !important;
+        border-radius: 12px !important;
+        padding: 12px 24px !important;
+        font-weight: 600 !important;
+        cursor: pointer !important;
+        transition: all 0.3s ease !important;
+        box-shadow: 0 4px 15px rgba(139, 92, 246, 0.4) !important;
+        backdrop-filter: blur(10px) !important;
+        min-height: 44px !important;
+        display: flex !important;
+        align-items: center !important;
+        justify-content: center !important;
+    }
+
+    .gradio-container button:hover {
+        transform: translateY(-2px) !important;
+        box-shadow: 0 8px 25px rgba(139, 92, 246, 0.6) !important;
+        background: linear-gradient(135deg, #9333ea 0%, #2563eb 100%) !important;
+    }
+
+    .gradio-container button:active {
+        transform: translateY(0px) !important;
+    }
+
+    /* Send button specific styling */
+    .gradio-container .gr-button {
+        background: linear-gradient(135deg, var(--accent-purple) 0%, var(--accent-blue) 100%) !important;
+        border: 1px solid rgba(255, 255, 255, 0.2) !important;
+        color: white !important;
+        font-weight: 600 !important;
+        text-transform: none !important;
+        letter-spacing: 0.5px !important;
+    }
+
+    /* Chatbot specific styling */
+    .gradio-chatbot {
+        background: transparent !important;
+        border: none !important;
+    }
+
+    /* Footer */
+    .footer {
+        text-align: center !important;
+        padding: 28px !important;
+        color: var(--text-secondary) !important;
+        background: var(--glass-bg) !important;
+        backdrop-filter: blur(20px) !important;
+        border: 1px solid var(--glass-border) !important;
+        border-radius: 20px !important;
+        margin-top: 32px !important;
+        box-shadow: var(--shadow-glow) !important;
+    }
+
+    /* Scrollbar styling */
+    ::-webkit-scrollbar {
+        width: 8px;
+    }
+
+    ::-webkit-scrollbar-track {
+        background: rgba(255, 255, 255, 0.05);
+        border-radius: 4px;
+    }
+
+    ::-webkit-scrollbar-thumb {
+        background: linear-gradient(135deg, var(--accent-purple), var(--accent-blue));
+        border-radius: 4px;
+    }
+
+    ::-webkit-scrollbar-thumb:hover {
+        background: linear-gradient(135deg, #9333ea, #2563eb);
+    }
+
+    /* Responsive design */
+    @media (max-width: 768px) {
+        .gradio-container {
+            padding: 12px !important;
+        }
+        
+        .header {
+            padding: 20px !important;
+            border-radius: 16px !important;
+        }
+        
+        .header h1 {
+            font-size: 28px !important;
+        }
         
-    Returns:
-        str: The number of times the letter appears in the word
+        .tools-area,
+        .chat-area {
+            border-radius: 16px !important;
+            padding: 20px !important;
+        }
+    }
     """
-    count = word.lower().count(letter.lower())
-    return f"The letter '{letter}' appears {count} times in '{word}'"
-
-demo = gr.Interface(
-    fn=letter_counter,
-    inputs=[
-        gr.Textbox(label="Word or phrase", value="strawberry"),
-        gr.Textbox(label="Letter to count", value="r")
-    ],
-    outputs=gr.Textbox(label="Result"),
-    title="Letter Counter MCP Server",
-    description="Count how many times a letter appears in a word"
-)
+    
+    # Create wrapper function for proper chat handling
+    def respond(message, history):
+        """Response handler for Gradio messages format"""
+        if not message.strip():
+            return "", history
+        
+        # Convert messages format to simple tuples
+        simple_history = []
+        for i in range(0, len(history), 2):
+            if i + 1 < len(history):
+                user_msg = history[i].get('content', '') if isinstance(history[i], dict) else str(history[i])
+                bot_msg = history[i + 1].get('content', '') if isinstance(history[i + 1], dict) else str(history[i + 1])
+                if user_msg and bot_msg:
+                    simple_history.append([user_msg, bot_msg])
+        
+        # Generate response
+        for response_chunk in creed_ai.generate_response(message, simple_history):
+            # Create new history with the streaming response
+            new_history = history + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": response_chunk}
+            ]
+            yield "", new_history
+    
+    # Create the interface with modern theme
+    with gr.Blocks(
+        title="🎸 Creed Bratton AI",
+        css=modern_css,
+        theme=gr.themes.Base()  # Use base theme for better CSS control
+    ) as demo:
+        
+        # Modern header
+        gr.HTML(f"""
+        <div class="header">
+            <h1>🎸 Creed Bratton AI</h1>
+            <p>Powered by phxdev/creed-qwen-0.5b-lora • Running on {'⚡ ZeroGPU' if SPACES_AVAILABLE else '🖥️ CPU'}</p>
+        </div>
+        """)
+        
+        # Model info with glass styling
+        gr.HTML("""
+        <div class="info-box">
+            <strong>Model:</strong> phxdev/creed-qwen-0.5b-lora<br>
+            <strong>Base:</strong> Qwen 0.5B + LoRA fine-tuning<br>
+            <strong>Tokens:</strong> &lt;thinking&gt;, &lt;conspiracy&gt;, &lt;tangent&gt;<br>
+            <strong>Mode:</strong> ZeroGPU optimized + Coherence validation
+        </div>
+        """)
+        
+        # MCP status
+        if os.environ.get('GRADIO_MCP_ENABLED'):
+            gr.HTML("""
+            <div class="status-box">
+                ✓ MCP Server Active • Available as tool for Claude Desktop
+            </div>
+            """)
+        
+        # Main chat interface with glass styling
+        with gr.Row(elem_classes="chat-area"):
+            chatbot = gr.Chatbot(
+                type='messages',  # Use messages format (modern)
+                height=550,
+                show_copy_button=True,
+                show_share_button=False,
+                avatar_images=["👤", "🎸"],
+                bubble_full_width=False,
+                show_label=False,
+                placeholder="🎸 Creed is ready...",
+                container=False
+            )
+        
+        # Input with explicit send button
+        with gr.Row():
+            with gr.Column(scale=7):
+                msg = gr.Textbox(
+                    placeholder="Ask Creed anything...",
+                    container=False,
+                    submit_btn=False,  # Disable built-in submit
+                    stop_btn=False
+                )
+            with gr.Column(scale=1, min_width=100):
+                send_btn = gr.Button("Send", variant="primary", size="lg")
+        
+        # Wire up the chat - both Enter key and Send button
+        msg.submit(
+            respond,
+            inputs=[msg, chatbot],
+            outputs=[msg, chatbot],
+            show_progress="hidden"
+        )
+        
+        send_btn.click(
+            respond,
+            inputs=[msg, chatbot],
+            outputs=[msg, chatbot],
+            show_progress="hidden"
+        )
+        
+        # MCP Tools section with glass styling
+        with gr.Row(elem_classes="tools-area"):
+            gr.HTML('<div class="tools-title">🛠️ Claude → Creed Transformation Tools</div>')
+            gr.HTML("""
+            <div style="margin-bottom: 20px; padding: 12px; background: rgba(255,255,255,0.1); border-radius: 8px; font-size: 14px;">
+                <strong>These MCP tools transform Claude into Creed Bratton for your Claude Desktop workflow.</strong><br>
+                Test them here, then use them in Claude Desktop to get Creed-mode responses from Claude.
+            </div>
+            """)
+            
+            with gr.Row():
+                with gr.Column():
+                    wisdom_topic = gr.Textbox(
+                        label="Wisdom Topic",
+                        placeholder="life, business, relationships..."
+                    )
+                    wisdom_output = gr.Textbox(
+                        label="Creed Wisdom Prompt (Copy to Claude Desktop)", 
+                        interactive=False,
+                        lines=5
+                    )
+                    wisdom_btn = gr.Button("Generate Wisdom Prompt", variant="primary")
+                    
+                with gr.Column():
+                    story_situation = gr.Textbox(
+                        label="Story Request",
+                        placeholder="mysterious, business, quarry..."
+                    )
+                    story_output = gr.Textbox(
+                        label="Creed Story Prompt (Copy to Claude Desktop)",
+                        interactive=False,
+                        lines=5
+                    )
+                    story_btn = gr.Button("Generate Story Prompt", variant="primary")
+            
+            with gr.Row():
+                with gr.Column():
+                    analysis_subject = gr.Textbox(
+                        label="Analysis Subject",
+                        placeholder="business strategy, market trends..."
+                    )
+                    analysis_output = gr.Textbox(
+                        label="Creed Analysis Prompt",
+                        interactive=False,
+                        lines=4
+                    )
+                    analysis_btn = gr.Button("Generate Analysis Prompt", variant="primary")
+                    
+                with gr.Column():
+                    conspiracy_topic = gr.Textbox(
+                        label="Conspiracy Topic", 
+                        placeholder="government, corporations, technology..."
+                    )
+                    conspiracy_output = gr.Textbox(
+                        label="Creed Conspiracy Prompt",
+                        interactive=False,
+                        lines=4
+                    )
+                    conspiracy_btn = gr.Button("Generate Conspiracy Prompt", variant="primary")
+                    
+            with gr.Row():
+                activate_output = gr.Textbox(
+                    label="Creed Mode Activation Prompt (Full Transformation)",
+                    interactive=False,
+                    lines=6
+                )
+                activate_btn = gr.Button("🎸 Generate FULL Creed Mode Prompt", variant="secondary", size="lg")
+            gr.HTML('<div class="tools-title">🛠️ MCP Tools</div>')
+            
+            with gr.Row():
+                with gr.Column():
+                    wisdom_topic = gr.Textbox(
+                        label="Wisdom Topic",
+                        placeholder="life, business, relationships..."
+                    )
+                    wisdom_output = gr.Textbox(
+                        label="Creed's Response", 
+                        interactive=False,
+                        lines=3
+                    )
+                    wisdom_btn = gr.Button("Ask Creed", variant="primary")
+                    
+                with gr.Column():
+                    story_situation = gr.Textbox(
+                        label="Story Request",
+                        placeholder="Tell me about..."
+                    )
+                    story_output = gr.Textbox(
+                        label="Creed's Story",
+                        interactive=False,
+                        lines=3
+                    )
+                    story_btn = gr.Button("Get Story", variant="primary")
+        
+        # Wire up the tools - now they're Claude transformation tools
+        wisdom_btn.click(
+            lambda topic: creed_ai.creed_wisdom_tool(topic or "life"),
+            inputs=[wisdom_topic],
+            outputs=[wisdom_output]
+        )
+        
+        story_btn.click(
+            lambda situation: creed_ai.creed_story_tool(situation or "mysterious"),
+            inputs=[story_situation],
+            outputs=[story_output]
+        )
+        
+        # Modern footer
+        gr.HTML("""
+        <div class="footer">
+            <strong>Creed Bratton AI</strong><br>
+            Model: phxdev/creed-qwen-0.5b-lora • Trained by Mark Scott<br>
+            <em>"Sometimes a guy's gotta ride the bull, am I right?"</em>
+        </div>
+        """)
+    
+    # Launch with modern styling and public sharing
+    print("🚀 Launching Real Creed AI with modern glassmorphism design...")
+    
+    demo.launch(
+        ssr_mode=False,
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=True,  # Create public link
+        show_error=True,
+        mcp_server=True
+    )
 
 if __name__ == "__main__":
-    demo.launch(mcp_server=True)
+    main()