app.py

import os
import gradio as gr
import json
import time
from datetime import datetime
from llama_cpp import Llama
from faster_whisper import WhisperModel
from huggingface_hub import hf_hub_download

# ===== CONFIG =====
MODELS_DIR = "/data/models"
MAX_TOKENS = 2048
CONTEXT_SIZE = 4096

# Model repos for auto-download (5 models to stay under 50GB)
MODEL_REPOS = {
    "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf": "bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF",
    "qwen2.5-coder-7b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-7B-Instruct-GGUF",
    "qwen2.5-coder-3b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-3B-Instruct-GGUF",
    "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF",
    "qwen2.5-coder-0.5b-instruct-q4_k_m.gguf": "Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF",
}

MODELS = {
    "🧠 DeepSeek V2 Lite (Best)": "DeepSeek-Coder-V2-Lite-Instruct-Q4_K_M.gguf",
    "⚖️ Qwen2.5 Coder 7B (Balanced)": "qwen2.5-coder-7b-instruct-q4_k_m.gguf",
    "🚀 Qwen2.5 Coder 3B (Fast)": "qwen2.5-coder-3b-instruct-q4_k_m.gguf",
    "💨 Qwen2.5 Coder 1.5B (Quick)": "qwen2.5-coder-1.5b-instruct-q4_k_m.gguf",
    "🔬 Qwen2.5 Coder 0.5B (Instant)": "qwen2.5-coder-0.5b-instruct-q4_k_m.gguf",
}

LANGUAGES = [
    "Python", "JavaScript", "TypeScript", "Go", "Rust", "Java", "C++", "C#", "C",
    "PHP", "Ruby", "Swift", "Kotlin", "Scala", "R", "Julia", "Perl", "HTML/CSS",
    "SQL", "Bash", "PowerShell", "Lua"
]

# ===== GLOBAL STATE =====
loaded_models = {}
current_model_name = None
whisper_model = None

# ===== INITIALIZATION =====
def init_directories():
    os.makedirs(MODELS_DIR, exist_ok=True)
    print(f"📁 Models directory: {MODELS_DIR}")

def get_cached_models():
    if not os.path.exists(MODELS_DIR):
        return []
    return [f for f in os.listdir(MODELS_DIR) if f.endswith('.gguf')]

def download_model(filename):
    """Download a model if not cached"""
    model_path = os.path.join(MODELS_DIR, filename)
    if os.path.exists(model_path):
        return model_path
    
    repo = MODEL_REPOS.get(filename)
    if not repo:
        print(f"❌ No repo found for {filename}")
        return None
    
    print(f"⬇️ Downloading {filename}...")
    try:
        path = hf_hub_download(
            repo_id=repo,
            filename=filename,
            local_dir=MODELS_DIR
        )
        print(f"✅ Downloaded {filename}")
        return path
    except Exception as e:
        print(f"❌ Download failed: {e}")
        return None

def load_model(model_name):
    global loaded_models, current_model_name
    
    if model_name == current_model_name and model_name in loaded_models:
        return loaded_models[model_name]
    
    # Unload previous model to save RAM
    if current_model_name and current_model_name != model_name:
        if current_model_name in loaded_models:
            del loaded_models[current_model_name]
            print(f"🗑️ Unloaded {current_model_name}")
    
    filename = MODELS.get(model_name)
    if not filename:
        return None
    
    # Download if needed
    model_path = download_model(filename)
    if not model_path or not os.path.exists(model_path):
        return None
    
    print(f"📥 Loading {model_name}...")
    try:
        llm = Llama(
            model_path=model_path,
            n_ctx=CONTEXT_SIZE,
            n_threads=4,
            n_batch=512,
            verbose=False
        )
        loaded_models[model_name] = llm
        current_model_name = model_name
        print(f"✅ {model_name} loaded!")
        return llm
    except Exception as e:
        print(f"❌ Failed to load: {e}")
        return None

def init_whisper():
    global whisper_model
    try:
        print("🎤 Loading Whisper...")
        whisper_model = WhisperModel("tiny", device="cpu", compute_type="int8")
        print("✅ Whisper ready!")
    except Exception as e:
        print(f"❌ Whisper failed: {e}")

# ===== LLM HELPERS =====
def generate_response(prompt, model_name, max_tokens=MAX_TOKENS, temperature=0.7):
    llm = load_model(model_name)
    if not llm:
        return "❌ Failed to load model. Please try again."
    
    try:
        response = llm(
            prompt,
            max_tokens=max_tokens,
            temperature=temperature,
            stop=["<|endoftext|>", "<|im_end|>", "</s>"],
            echo=False
        )
        return response["choices"][0]["text"].strip()
    except Exception as e:
        return f"❌ Error: {str(e)}"

def generate_stream(prompt, model_name, max_tokens=MAX_TOKENS, temperature=0.7):
    llm = load_model(model_name)
    if not llm:
        yield "❌ Failed to load model. Please try again."
        return
    
    try:
        stream = llm(
            prompt,
            max_tokens=max_tokens,
            temperature=temperature,
            stop=["<|endoftext|>", "<|im_end|>", "</s>"],
            stream=True,
            echo=False
        )
        for chunk in stream:
            if "choices" in chunk and chunk["choices"]:
                text = chunk["choices"][0].get("text", "")
                if text:
                    yield text
    except Exception as e:
        yield f"❌ Error: {str(e)}"

def format_prompt(system, user):
    return f"<|im_start|>system\n{system}<|im_end|>\n<|im_start|>user\n{user}<|im_end|>\n<|im_start|>assistant\n"

# ===== WHISPER =====
def transcribe_audio(audio_path):
    if not whisper_model or not audio_path:
        return ""
    try:
        segments, _ = whisper_model.transcribe(audio_path, beam_size=1)
        return " ".join([s.text for s in segments]).strip()
    except Exception as e:
        return f"❌ Transcription error: {str(e)}"

# ===== TOOL FUNCTIONS =====

# --- CORE TOOLS ---
def chat_stream(message, history, model, temperature, max_tokens):
    if not message.strip():
        yield history
        return
    
    history = history or []
    
    # Build conversation context from history
    system = "You are Axon, an expert AI coding assistant. Be helpful, concise, and provide working code examples when appropriate."
    conversation = ""
    for msg in history[-10:]:  # Last 10 messages
        if msg["role"] == "user":
            conversation += f"<|im_start|>user\n{msg['content']}<|im_end|>\n"
        elif msg["role"] == "assistant":
            conversation += f"<|im_start|>assistant\n{msg['content']}<|im_end|>\n"
    
    prompt = f"<|im_start|>system\n{system}<|im_end|>\n{conversation}<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
    
    # Add user message to history
    history = history + [{"role": "user", "content": message}]
    
    response = ""
    for chunk in generate_stream(prompt, model, max_tokens, temperature):
        response += chunk
        # Yield history with partial assistant response
        yield history + [{"role": "assistant", "content": response}]
    
    # Final yield with complete response
    yield history + [{"role": "assistant", "content": response}]

def generate_code(description, language, model, max_tokens):
    if not description.strip():
        return "Please provide a description."
    
    system = f"You are an expert {language} programmer. Generate clean, well-commented, production-ready code."
    prompt = format_prompt(system, f"Generate {language} code for: {description}")
    return generate_response(prompt, model, max_tokens)

def explain_code(code, model, detail_level, max_tokens):
    if not code.strip():
        return "Please provide code to explain."
    
    detail_map = {
        "Brief": "Give a brief 2-3 sentence explanation.",
        "Normal": "Explain the code clearly with key points.",
        "Detailed": "Provide a comprehensive explanation covering logic, patterns, and potential improvements."
    }
    
    system = "You are an expert code explainer. " + detail_map.get(detail_level, detail_map["Normal"])
    prompt = format_prompt(system, f"Explain this code:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def debug_code(code, error, model, max_tokens):
    if not code.strip():
        return "Please provide code to debug."
    
    system = "You are an expert debugger. Identify bugs, explain the issues, and provide corrected code."
    user = f"Debug this code:\n\n```\n{code}\n```"
    if error.strip():
        user += f"\n\nError message:\n{error}"
    
    prompt = format_prompt(system, user)
    return generate_response(prompt, model, max_tokens)

def review_code(code, model, max_tokens):
    if not code.strip():
        return "Please provide code to review."
    
    system = """You are a senior code reviewer. Review the code for:
1. Code quality and readability
2. Potential bugs or issues
3. Security vulnerabilities
4. Performance concerns
5. Best practices
Provide specific, actionable feedback."""
    
    prompt = format_prompt(system, f"Review this code:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

# --- ADVANCED TOOLS ---
def security_scan(code, model, max_tokens):
    if not code.strip():
        return "Please provide code to scan."
    
    system = """You are a security expert. Scan the code for vulnerabilities including:
- SQL injection
- XSS (Cross-site scripting)
- CSRF vulnerabilities
- Insecure data handling
- Authentication issues
- Input validation problems
- Secrets/credentials exposure
Rate severity (Critical/High/Medium/Low) and provide fixes."""
    
    prompt = format_prompt(system, f"Security scan this code:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def analyze_complexity(code, model, max_tokens):
    if not code.strip():
        return "Please provide code to analyze."
    
    system = """You are an algorithms expert. Analyze the code complexity:
1. Time complexity (Big O notation)
2. Space complexity (Big O notation)
3. Explain the analysis step by step
4. Suggest optimizations if possible"""
    
    prompt = format_prompt(system, f"Analyze complexity:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def convert_code(code, from_lang, to_lang, model, max_tokens):
    if not code.strip():
        return "Please provide code to convert."
    
    system = f"You are an expert polyglot programmer. Convert code from {from_lang} to {to_lang}. Preserve functionality and use idiomatic patterns for the target language."
    prompt = format_prompt(system, f"Convert this {from_lang} code to {to_lang}:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def generate_tests(code, language, framework, model, max_tokens):
    if not code.strip():
        return "Please provide code to test."
    
    system = f"You are a testing expert. Generate comprehensive {framework} unit tests for {language} code. Include edge cases, error cases, and happy path tests."
    prompt = format_prompt(system, f"Generate {framework} tests for:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def document_code(code, language, style, model, max_tokens):
    if not code.strip():
        return "Please provide code to document."
    
    style_map = {
        "Docstrings": "Add comprehensive docstrings to all functions/classes.",
        "Inline Comments": "Add helpful inline comments explaining the logic.",
        "Both": "Add both docstrings and inline comments."
    }
    
    system = f"You are a documentation expert. {style_map.get(style, style_map['Both'])} Use {language} documentation conventions."
    prompt = format_prompt(system, f"Document this code:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def optimize_code(code, language, focus, model, max_tokens):
    if not code.strip():
        return "Please provide code to optimize."
    
    system = f"You are a performance optimization expert. Optimize the {language} code focusing on {focus}. Show the optimized code and explain the improvements."
    prompt = format_prompt(system, f"Optimize this code for {focus}:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def code_diff(code1, code2, model, max_tokens):
    if not code1.strip() or not code2.strip():
        return "Please provide both code snippets to compare."
    
    system = "You are a code analysis expert. Compare the two code snippets and explain the differences, which is better, and why."
    prompt = format_prompt(system, f"Compare these code snippets:\n\n**Code 1:**\n```\n{code1}\n```\n\n**Code 2:**\n```\n{code2}\n```")
    return generate_response(prompt, model, max_tokens)

def to_pseudocode(code, output_type, model, max_tokens):
    if not code.strip():
        return "Please provide code to convert."
    
    if output_type == "Pseudocode":
        system = "Convert the code to clear, readable pseudocode that anyone can understand."
    else:
        system = "Convert the code to a text-based flowchart using ASCII art or a structured description."
    
    prompt = format_prompt(system, f"Convert to {output_type}:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def interview_challenge(topic, difficulty, language, model, max_tokens):
    system = f"You are a technical interviewer. Create a {difficulty} {language} coding challenge about {topic}. Include: problem statement, examples, constraints, hints, and a solution with explanation."
    prompt = format_prompt(system, f"Create a {difficulty} {topic} challenge in {language}")
    return generate_response(prompt, model, max_tokens)

# --- BUILDERS ---
def build_sql(description, sql_type, model, max_tokens):
    if not description.strip():
        return "Please describe what SQL you need."
    
    system = f"You are a SQL expert. Generate {sql_type} SQL queries. Include comments explaining the query."
    prompt = format_prompt(system, f"Generate SQL for: {description}")
    return generate_response(prompt, model, max_tokens)

def build_shell(description, shell_type, model, max_tokens):
    if not description.strip():
        return "Please describe what command you need."
    
    system = f"You are a {shell_type} expert. Generate safe, well-commented shell commands. Warn about any dangerous operations."
    prompt = format_prompt(system, f"Generate {shell_type} command for: {description}")
    return generate_response(prompt, model, max_tokens)

def build_cron(description, model, max_tokens):
    if not description.strip():
        return "Please describe the schedule you need."
    
    system = "You are a cron expert. Generate cron expressions with clear explanations. Include the cron format: minute hour day month weekday"
    prompt = format_prompt(system, f"Generate cron expression for: {description}")
    return generate_response(prompt, model, max_tokens)

def build_regex(description, model, max_tokens):
    if not description.strip():
        return "Please describe the pattern you need."
    
    system = "You are a regex expert. Generate regular expressions with explanations and test examples. Support multiple regex flavors if relevant."
    prompt = format_prompt(system, f"Generate regex for: {description}")
    return generate_response(prompt, model, max_tokens)

def build_api(description, framework, model, max_tokens):
    if not description.strip():
        return "Please describe the API endpoint you need."
    
    system = f"You are an API expert. Generate a complete {framework} REST API endpoint with error handling, validation, and documentation."
    prompt = format_prompt(system, f"Generate {framework} API endpoint for: {description}")
    return generate_response(prompt, model, max_tokens)

# --- DATA TOOLS ---
def generate_mock_data(schema, count, format_type, model, max_tokens):
    if not schema.strip():
        return "Please provide a schema or description."
    
    system = f"You are a data generation expert. Generate {count} realistic mock data records in {format_type} format based on the schema."
    prompt = format_prompt(system, f"Generate {count} mock records in {format_type}:\n\nSchema: {schema}")
    return generate_response(prompt, model, max_tokens)

def convert_data_format(data, from_format, to_format, model, max_tokens):
    if not data.strip():
        return "Please provide data to convert."
    
    system = f"Convert the data from {from_format} to {to_format}. Preserve all information and use proper formatting."
    prompt = format_prompt(system, f"Convert from {from_format} to {to_format}:\n\n```\n{data}\n```")
    return generate_response(prompt, model, max_tokens)

# --- NEW v26 TOOLS ---
def refactor_code(code, language, pattern, model, max_tokens):
    if not code.strip():
        return "Please provide code to refactor."
    
    system = f"You are a software architect. Refactor the {language} code using {pattern} patterns. Explain the improvements and show before/after."
    prompt = format_prompt(system, f"Refactor using {pattern}:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def generate_benchmark(code, language, model, max_tokens):
    if not code.strip():
        return "Please provide code to benchmark."
    
    system = f"You are a performance testing expert. Generate comprehensive {language} benchmark code to measure performance. Include setup, warmup, multiple iterations, and statistics."
    prompt = format_prompt(system, f"Generate benchmark for:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def analyze_dependencies(code, language, model, max_tokens):
    if not code.strip():
        return "Please provide code to analyze."
    
    system = f"""You are a dependency analysis expert. Analyze the {language} code for:
1. Direct imports/dependencies
2. Potential circular dependencies
3. Unused imports
4. Missing dependencies
5. Version compatibility concerns
6. Security advisories for known packages"""
    
    prompt = format_prompt(system, f"Analyze dependencies:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

def generate_changelog(diff_or_commits, version, model, max_tokens):
    if not diff_or_commits.strip():
        return "Please provide diff or commit messages."
    
    system = """You are a technical writer. Generate a professional changelog entry with:
- Version number and date
- Categories: Added, Changed, Fixed, Removed, Security
- Clear, user-friendly descriptions
- Breaking changes highlighted"""
    
    prompt = format_prompt(system, f"Generate changelog for version {version}:\n\n{diff_or_commits}")
    return generate_response(prompt, model, max_tokens)

def suggest_improvements(code, language, model, max_tokens):
    if not code.strip():
        return "Please provide code to analyze."
    
    system = f"""You are a senior {language} developer. Provide actionable improvement suggestions:
1. Code quality improvements
2. Performance optimizations
3. Better error handling
4. Modern language features to use
5. Design pattern recommendations
6. Testing suggestions
Rate each suggestion by impact (High/Medium/Low) and effort."""
    
    prompt = format_prompt(system, f"Suggest improvements:\n\n```\n{code}\n```")
    return generate_response(prompt, model, max_tokens)

# ===== UI =====
def create_ui():
    # Check cached models
    cached = get_cached_models()
    total_models = len(MODELS)
    
    with gr.Blocks(title="Axon v26") as demo:
        # Header
        gr.HTML("""
        <div class="main-header">
            <h1>⚡ Axon v26</h1>
            <p>The Ultimate Free AI Coding Assistant • 5 Models • 25 Tools • 100% Local</p>
        </div>
        """)
        
        # Status bar
        status_emoji = "🟢" if len(cached) >= total_models else "🟡"
        gr.HTML(f"""
        <div class="status-bar" style="background: linear-gradient(135deg, #667eea22, #764ba222);">
            {status_emoji} <strong>{len(cached)}/{total_models}</strong> models cached • 
            {"All models ready!" if len(cached) >= total_models else "Models download on first use (~1-9GB each)"}
        </div>
        """)
        
        # Global controls
        with gr.Row():
            model_dropdown = gr.Dropdown(
                choices=list(MODELS.keys()),
                value=list(MODELS.keys())[2],  # Default to 3B
                label="🤖 Model",
                scale=2
            )
            max_tokens = gr.Slider(256, 4096, value=2048, step=256, label="Max Tokens", scale=1)
        
        # Tabs
        with gr.Tabs():
            # === CHAT TAB ===
            with gr.Tab("💬 Chat"):
                gr.HTML("<p class='tool-description'>Have a conversation about code, get help, ask questions.</p>")
                chatbot = gr.Chatbot(height=400)
                with gr.Row():
                    msg = gr.Textbox(placeholder="Ask me anything about code...", scale=4, show_label=False)
                    send_btn = gr.Button("Send", variant="primary")
                with gr.Row():
                    temperature = gr.Slider(0, 1, value=0.7, step=0.1, label="Temperature")
                    audio_input = gr.Audio(type="filepath", label="🎤 Voice Input")
                    transcribe_btn = gr.Button("Transcribe")
                clear_btn = gr.Button("Clear Chat")
            
            # === GENERATE TAB ===
            with gr.Tab("⚡ Generate"):
                gr.HTML("<p class='tool-description'>Describe what you want, get working code.</p>")
                gen_prompt = gr.Textbox(lines=3, placeholder="Describe the code you want...", label="Description")
                gen_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                gen_btn = gr.Button("Generate", variant="primary")
                gen_output = gr.Code(label="Generated Code", language="python")
            
            # === EXPLAIN TAB ===
            with gr.Tab("🔍 Explain"):
                gr.HTML("<p class='tool-description'>Understand any code with detailed explanations.</p>")
                explain_input = gr.Code(lines=10, label="Code to Explain", language="python")
                explain_detail = gr.Radio(["Brief", "Normal", "Detailed"], value="Normal", label="Detail Level")
                explain_btn = gr.Button("Explain", variant="primary")
                explain_output = gr.Markdown(label="Explanation")
            
            # === DEBUG TAB ===
            with gr.Tab("🔧 Debug"):
                gr.HTML("<p class='tool-description'>Find and fix bugs with AI assistance.</p>")
                debug_input = gr.Code(lines=10, label="Buggy Code", language="python")
                debug_error = gr.Textbox(lines=3, placeholder="Paste error message (optional)", label="Error Message")
                debug_btn = gr.Button("Debug", variant="primary")
                debug_output = gr.Markdown(label="Debug Results")
            
            # === REVIEW TAB ===
            with gr.Tab("📋 Review"):
                gr.HTML("<p class='tool-description'>Get comprehensive code review feedback.</p>")
                review_input = gr.Code(lines=10, label="Code to Review", language="python")
                review_btn = gr.Button("Review", variant="primary")
                review_output = gr.Markdown(label="Review Results")
            
            # === SECURITY TAB ===
            with gr.Tab("🔐 Security"):
                gr.HTML("<p class='tool-description'>Scan code for security vulnerabilities.</p>")
                security_input = gr.Code(lines=10, label="Code to Scan", language="python")
                security_btn = gr.Button("Scan", variant="primary")
                security_output = gr.Markdown(label="Security Report")
            
            # === COMPLEXITY TAB ===
            with gr.Tab("📊 Complexity"):
                gr.HTML("<p class='tool-description'>Analyze time and space complexity (Big O).</p>")
                complexity_input = gr.Code(lines=10, label="Code to Analyze", language="python")
                complexity_btn = gr.Button("Analyze", variant="primary")
                complexity_output = gr.Markdown(label="Complexity Analysis")
            
            # === CONVERT TAB ===
            with gr.Tab("🔄 Convert"):
                gr.HTML("<p class='tool-description'>Translate code between programming languages.</p>")
                convert_input = gr.Code(lines=10, label="Code to Convert", language="python")
                with gr.Row():
                    convert_from = gr.Dropdown(choices=LANGUAGES, value="Python", label="From")
                    convert_to = gr.Dropdown(choices=LANGUAGES, value="JavaScript", label="To")
                convert_btn = gr.Button("Convert", variant="primary")
                convert_output = gr.Code(label="Converted Code", language="javascript")
            
            # === TEST TAB ===
            with gr.Tab("🧪 Test"):
                gr.HTML("<p class='tool-description'>Generate comprehensive unit tests.</p>")
                test_input = gr.Code(lines=10, label="Code to Test", language="python")
                with gr.Row():
                    test_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                    test_fw = gr.Dropdown(choices=["pytest", "unittest", "Jest", "Mocha", "JUnit", "RSpec"], value="pytest", label="Framework")
                test_btn = gr.Button("Generate Tests", variant="primary")
                test_output = gr.Code(label="Generated Tests", language="python")
            
            # === DOCUMENT TAB ===
            with gr.Tab("📝 Document"):
                gr.HTML("<p class='tool-description'>Add documentation to your code.</p>")
                doc_input = gr.Code(lines=10, label="Code to Document", language="python")
                with gr.Row():
                    doc_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                    doc_style = gr.Dropdown(choices=["Docstrings", "Inline Comments", "Both"], value="Both", label="Style")
                doc_btn = gr.Button("Document", variant="primary")
                doc_output = gr.Code(label="Documented Code", language="python")
            
            # === OPTIMIZE TAB ===
            with gr.Tab("🚀 Optimize"):
                gr.HTML("<p class='tool-description'>Improve code performance.</p>")
                opt_input = gr.Code(lines=10, label="Code to Optimize", language="python")
                with gr.Row():
                    opt_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                    opt_focus = gr.Dropdown(choices=["Speed", "Memory", "Readability", "All"], value="All", label="Focus")
                opt_btn = gr.Button("Optimize", variant="primary")
                opt_output = gr.Markdown(label="Optimized Code")
            
            # === DIFF TAB ===
            with gr.Tab("🔀 Diff"):
                gr.HTML("<p class='tool-description'>Compare two code snippets.</p>")
                with gr.Row():
                    diff_code1 = gr.Code(lines=8, label="Code 1", language="python")
                    diff_code2 = gr.Code(lines=8, label="Code 2", language="python")
                diff_btn = gr.Button("Compare", variant="primary")
                diff_output = gr.Markdown(label="Comparison")
            
            # === PSEUDOCODE TAB ===
            with gr.Tab("📐 Pseudo"):
                gr.HTML("<p class='tool-description'>Convert code to pseudocode or flowcharts.</p>")
                pseudo_input = gr.Code(lines=10, label="Code", language="python")
                pseudo_type = gr.Radio(["Pseudocode", "Flowchart"], value="Pseudocode", label="Output Type")
                pseudo_btn = gr.Button("Convert", variant="primary")
                pseudo_output = gr.Markdown(label="Output")
            
            # === INTERVIEW TAB ===
            with gr.Tab("🎓 Interview"):
                gr.HTML("<p class='tool-description'>Generate coding interview challenges.</p>")
                with gr.Row():
                    interview_topic = gr.Textbox(placeholder="e.g., binary trees, dynamic programming", label="Topic")
                    interview_diff = gr.Dropdown(choices=["Easy", "Medium", "Hard"], value="Medium", label="Difficulty")
                    interview_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                interview_btn = gr.Button("Generate Challenge", variant="primary")
                interview_output = gr.Markdown(label="Challenge")
            
            # === SQL TAB ===
            with gr.Tab("🗄️ SQL"):
                gr.HTML("<p class='tool-description'>Generate SQL queries from natural language.</p>")
                sql_desc = gr.Textbox(lines=3, placeholder="Describe your query...", label="Description")
                sql_type = gr.Dropdown(choices=["MySQL", "PostgreSQL", "SQLite", "SQL Server", "Oracle"], value="PostgreSQL", label="Database")
                sql_btn = gr.Button("Generate SQL", variant="primary")
                sql_output = gr.Code(label="Generated SQL", language="sql")
            
            # === SHELL TAB ===
            with gr.Tab("🐚 Shell"):
                gr.HTML("<p class='tool-description'>Generate shell commands from descriptions.</p>")
                shell_desc = gr.Textbox(lines=3, placeholder="Describe what you want to do...", label="Description")
                shell_type = gr.Dropdown(choices=["Bash", "PowerShell", "Zsh", "Fish"], value="Bash", label="Shell")
                shell_btn = gr.Button("Generate", variant="primary")
                shell_output = gr.Code(label="Generated Command", language="shell")
            
            # === CRON TAB ===
            with gr.Tab("⏰ Cron"):
                gr.HTML("<p class='tool-description'>Generate cron schedule expressions.</p>")
                cron_desc = gr.Textbox(lines=2, placeholder="e.g., Every Monday at 9am", label="Schedule Description")
                cron_btn = gr.Button("Generate", variant="primary")
                cron_output = gr.Markdown(label="Cron Expression")
            
            # === REGEX TAB ===
            with gr.Tab("🎯 Regex"):
                gr.HTML("<p class='tool-description'>Generate regular expressions with explanations.</p>")
                regex_desc = gr.Textbox(lines=2, placeholder="e.g., Match email addresses", label="Pattern Description")
                regex_btn = gr.Button("Generate", variant="primary")
                regex_output = gr.Markdown(label="Regex Pattern")
            
            # === API TAB ===
            with gr.Tab("🔗 API"):
                gr.HTML("<p class='tool-description'>Generate REST API endpoint boilerplate.</p>")
                api_desc = gr.Textbox(lines=3, placeholder="Describe your API endpoint...", label="Description")
                api_fw = gr.Dropdown(choices=["FastAPI", "Flask", "Express", "Django", "Spring Boot", "Go Gin"], value="FastAPI", label="Framework")
                api_btn = gr.Button("Generate", variant="primary")
                api_output = gr.Code(label="Generated API", language="python")
            
            # === MOCK DATA TAB ===
            with gr.Tab("📦 Mock"):
                gr.HTML("<p class='tool-description'>Generate realistic test data.</p>")
                mock_schema = gr.Textbox(lines=3, placeholder="e.g., users with name, email, age", label="Schema Description")
                with gr.Row():
                    mock_count = gr.Number(value=5, label="Count", precision=0)
                    mock_format = gr.Dropdown(choices=["JSON", "CSV", "SQL INSERT", "YAML"], value="JSON", label="Format")
                mock_btn = gr.Button("Generate", variant="primary")
                mock_output = gr.Code(label="Generated Data", language="json")
            
            # === FORMAT TAB ===
            with gr.Tab("🔄 Format"):
                gr.HTML("<p class='tool-description'>Convert between data formats.</p>")
                format_input = gr.Code(lines=10, label="Input Data", language="json")
                with gr.Row():
                    format_from = gr.Dropdown(choices=["JSON", "YAML", "XML", "CSV", "TOML"], value="JSON", label="From")
                    format_to = gr.Dropdown(choices=["JSON", "YAML", "XML", "CSV", "TOML"], value="YAML", label="To")
                format_btn = gr.Button("Convert", variant="primary")
                format_output = gr.Code(label="Converted Data", language="python")
            
            # === REFACTOR TAB (NEW) ===
            with gr.Tab("🎨 Refactor"):
                gr.HTML("<p class='tool-description'>Restructure code using design patterns.</p>")
                refactor_input = gr.Code(lines=10, label="Code to Refactor", language="python")
                with gr.Row():
                    refactor_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                    refactor_pattern = gr.Dropdown(choices=["Clean Code", "SOLID", "DRY", "Factory", "Singleton", "Observer", "Strategy"], value="Clean Code", label="Pattern")
                refactor_btn = gr.Button("Refactor", variant="primary")
                refactor_output = gr.Markdown(label="Refactored Code")
            
            # === BENCHMARK TAB (NEW) ===
            with gr.Tab("📊 Bench"):
                gr.HTML("<p class='tool-description'>Generate performance benchmark code.</p>")
                bench_input = gr.Code(lines=10, label="Code to Benchmark", language="python")
                bench_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                bench_btn = gr.Button("Generate Benchmark", variant="primary")
                bench_output = gr.Code(label="Benchmark Code", language="python")
            
            # === DEPENDENCIES TAB (NEW) ===
            with gr.Tab("🔗 Deps"):
                gr.HTML("<p class='tool-description'>Analyze imports and dependencies.</p>")
                deps_input = gr.Code(lines=10, label="Code to Analyze", language="python")
                deps_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                deps_btn = gr.Button("Analyze", variant="primary")
                deps_output = gr.Markdown(label="Dependency Analysis")
            
            # === CHANGELOG TAB (NEW) ===
            with gr.Tab("📋 Change"):
                gr.HTML("<p class='tool-description'>Generate changelogs from diffs or commits.</p>")
                changelog_input = gr.Textbox(lines=10, placeholder="Paste git diff, commit messages, or describe changes...", label="Changes")
                changelog_version = gr.Textbox(value="1.0.0", label="Version")
                changelog_btn = gr.Button("Generate Changelog", variant="primary")
                changelog_output = gr.Markdown(label="Changelog")
            
            # === SUGGEST TAB (NEW) ===
            with gr.Tab("💡 Suggest"):
                gr.HTML("<p class='tool-description'>Get AI-powered improvement suggestions.</p>")
                suggest_input = gr.Code(lines=10, label="Code to Analyze", language="python")
                suggest_lang = gr.Dropdown(choices=LANGUAGES, value="Python", label="Language")
                suggest_btn = gr.Button("Get Suggestions", variant="primary")
                suggest_output = gr.Markdown(label="Suggestions")
        
        # Footer
        gr.HTML("""
        <div style="text-align: center; margin-top: 20px; padding: 15px; background: linear-gradient(135deg, #667eea11, #764ba211); border-radius: 10px;">
            <p><strong>⚡ Axon v26</strong> • Built with ❤️ by <a href="https://huggingface.co/AIencoder" target="_blank">AIencoder</a></p>
            <p style="font-size: 0.85em; color: #666;">
                <a href="https://huggingface.co/datasets/AIencoder/llama-cpp-wheels" target="_blank">🛞 Pre-built Wheels</a> • 
                100% Local • No API Keys • MIT License
            </p>
        </div>
        """)
        
        # === EVENT HANDLERS ===
        
        # Chat
        def respond(message, history, model, temp, tokens):
            history = history or []
            for updated_history in chat_stream(message, history, model, temp, tokens):
                yield updated_history, ""
        
        msg.submit(respond, [msg, chatbot, model_dropdown, temperature, max_tokens], [chatbot, msg])
        send_btn.click(respond, [msg, chatbot, model_dropdown, temperature, max_tokens], [chatbot, msg])
        clear_btn.click(lambda: ([], ""), None, [chatbot, msg])
        transcribe_btn.click(transcribe_audio, audio_input, msg)
        
        # Core tools
        gen_btn.click(generate_code, [gen_prompt, gen_lang, model_dropdown, max_tokens], gen_output)
        explain_btn.click(explain_code, [explain_input, model_dropdown, explain_detail, max_tokens], explain_output)
        debug_btn.click(debug_code, [debug_input, debug_error, model_dropdown, max_tokens], debug_output)
        review_btn.click(review_code, [review_input, model_dropdown, max_tokens], review_output)
        
        # Advanced tools
        security_btn.click(security_scan, [security_input, model_dropdown, max_tokens], security_output)
        complexity_btn.click(analyze_complexity, [complexity_input, model_dropdown, max_tokens], complexity_output)
        convert_btn.click(convert_code, [convert_input, convert_from, convert_to, model_dropdown, max_tokens], convert_output)
        test_btn.click(generate_tests, [test_input, test_lang, test_fw, model_dropdown, max_tokens], test_output)
        doc_btn.click(document_code, [doc_input, doc_lang, doc_style, model_dropdown, max_tokens], doc_output)
        opt_btn.click(optimize_code, [opt_input, opt_lang, opt_focus, model_dropdown, max_tokens], opt_output)
        diff_btn.click(code_diff, [diff_code1, diff_code2, model_dropdown, max_tokens], diff_output)
        pseudo_btn.click(to_pseudocode, [pseudo_input, pseudo_type, model_dropdown, max_tokens], pseudo_output)
        interview_btn.click(interview_challenge, [interview_topic, interview_diff, interview_lang, model_dropdown, max_tokens], interview_output)
        
        # Builders
        sql_btn.click(build_sql, [sql_desc, sql_type, model_dropdown, max_tokens], sql_output)
        shell_btn.click(build_shell, [shell_desc, shell_type, model_dropdown, max_tokens], shell_output)
        cron_btn.click(build_cron, [cron_desc, model_dropdown, max_tokens], cron_output)
        regex_btn.click(build_regex, [regex_desc, model_dropdown, max_tokens], regex_output)
        api_btn.click(build_api, [api_desc, api_fw, model_dropdown, max_tokens], api_output)
        
        # Data tools
        mock_btn.click(generate_mock_data, [mock_schema, mock_count, mock_format, model_dropdown, max_tokens], mock_output)
        format_btn.click(convert_data_format, [format_input, format_from, format_to, model_dropdown, max_tokens], format_output)
        
        # NEW v26 tools
        refactor_btn.click(refactor_code, [refactor_input, refactor_lang, refactor_pattern, model_dropdown, max_tokens], refactor_output)
        bench_btn.click(generate_benchmark, [bench_input, bench_lang, model_dropdown, max_tokens], bench_output)
        deps_btn.click(analyze_dependencies, [deps_input, deps_lang, model_dropdown, max_tokens], deps_output)
        changelog_btn.click(generate_changelog, [changelog_input, changelog_version, model_dropdown, max_tokens], changelog_output)
        suggest_btn.click(suggest_improvements, [suggest_input, suggest_lang, model_dropdown, max_tokens], suggest_output)
    
    return demo

# ===== MAIN =====
if __name__ == "__main__":
    print("⚡ Axon v26 - The FINAL Version")
    print("=" * 40)
    
    init_directories()
    init_whisper()
    
    cached = get_cached_models()
    print(f"📦 Cached models: {len(cached)}/{len(MODELS)}")
    for m in cached:
        print(f"   ✅ {m}")
    
    print("\n🚀 Starting Axon...")
    demo = create_ui()
    demo.launch(server_name="0.0.0.0", server_port=7860, css="""
    .main-header { text-align: center; margin-bottom: 20px; }
    .tool-description { color: #666; font-size: 0.9em; margin-bottom: 10px; }
    .status-bar { padding: 10px; border-radius: 8px; margin-bottom: 15px; }
    """)