import gradio as gr
import anthropic
import openai
from groq import Groq
import google.generativeai as genai
import requests
import time
import os
import pandas as pd
from datetime import datetime
import httpx

# Create httpx client without proxies for HuggingFace Spaces
http_client_no_proxy = httpx.Client(proxies={})

# Model configurations
MODEL_CONFIGS = {
    # === TIER 1: PREMIUM (Highest Quality) ===
    "Claude Sonnet 4.5 💎": {
        "provider": "anthropic",
        "model": "claude-sonnet-4-20250514",
        "api_key_env": "ANTHROPIC_API_KEY",
        "cost_per_1M_tokens": 3.00,
        "context_window": 32000,
        "tier": "premium",
        "description": "Best for complex architecture"
    },
    "GPT-4o 💎": {
        "provider": "openai",
        "model": "gpt-4o-2024-11-20",
        "api_key_env": "OPENAI_API_KEY",
        "cost_per_1M_tokens": 2.50,
        "context_window": 128000,
        "tier": "premium",
        "description": "Best for general purpose"
    },


    # === TIER 2: FREE GITHUB MODELS ===
    "Mistral Medium 3 (GitHub) 🆓": {
        "provider": "github",
        "model": "Mistral-Medium-3",
        "api_key_env": "GITHUB_TOKEN",
        "cost_per_1M_tokens": 0.00,
        "context_window": 131072,
        "tier": "free-github",
        "description": "Advanced reasoning + vision via GitHub Models (FREE)"
    },
    "GPT-4o-mini (GitHub) 🆓": {
        "provider": "github",
        "model": "gpt-4o-mini",
        "api_key_env": "GITHUB_TOKEN",
        "cost_per_1M_tokens": 0.00,
        "context_window": 128000,
        "tier": "free-github",
        "description": "Fast GPT-4o-mini via GitHub Models (FREE)"
    },

    # === TIER 3: FREE GROQ MODELS ===
    "Llama 3.3 70B (Groq) 🚀": {
        "provider": "groq",
        "model": "llama-3.3-70b-versatile",
        "api_key_env": "GROQ_API_KEY",
        "cost_per_1M_tokens": 0.00,
        "context_window": 131072,
        "tier": "free-groq",
        "description": "Latest Llama model via Groq (Ultra-fast)"
    },
    "Llama 3.1 8B (Groq) 🚀": {
        "provider": "groq",
        "model": "llama-3.1-8b-instant",
        "api_key_env": "GROQ_API_KEY",
        "cost_per_1M_tokens": 0.00,
        "context_window": 128000,
        "tier": "free-groq",
        "description": "Fast & efficient via Groq (FREE)"
    },
    "Gemma 2 9B (Groq) 🚀": {
        "provider": "groq",
        "model": "gemma2-9b-it",
        "api_key_env": "GROQ_API_KEY",
        "cost_per_1M_tokens": 0.00,
        "context_window": 8192,
        "tier": "free-groq",
        "description": "Efficient code generation via Groq"
    },

    # === TIER 4: FREE GOOGLE MODELS ===
    "Gemini 2.0 Flash 🔥": {
        "provider": "google",
        "model": "gemini-2.0-flash-exp",
        "api_key_env": "GOOGLE_API_KEY",
        "cost_per_1M_tokens": 0.00,
        "context_window": 1000000,
        "tier": "free-google",
        "description": "Experimental - Ultra-fast generation (1M context)"
    },


    # === TIER 5: FREE HUGGINGFACE MODELS ===
    "Qwen2.5-Coder-32B 🤗": {
        "provider": "huggingface",
        "model": "Qwen/Qwen2.5-Coder-32B-Instruct",
        "api_key_env": "HF_TOKEN",
        "cost_per_1M_tokens": 0.00,
        "context_window": 32768,
        "tier": "free-hf",
        "description": "32B code specialist via HF Inference API (FREE)"
    },
    "Phi-4-mini 🤗": {
        "provider": "huggingface",
        "model": "microsoft/Phi-4-mini-instruct",
        "api_key_env": "HF_TOKEN",
        "cost_per_1M_tokens": 0.00,
        "context_window": 16384,
        "tier": "free-hf",
        "description": "Microsoft's efficient code model via HF Inference API"
    },
    "TinyLlama 1.1B 🤗": {
        "provider": "huggingface",
        "model": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
        "api_key_env": "HF_TOKEN",
        "cost_per_1M_tokens": 0.00,
        "context_window": 2048,
        "tier": "free-hf",
        "description": "Ultra-fast 1.1B model for simple tasks (FREE)"
    }
}

SYSTEM_PROMPT = """You are Ectus-R, an expert autonomous software engineer powered by AION-R.
Your task is to generate production-ready code based on user requirements.

REQUIREMENTS:
1. Read the user's instructions carefully and decide language, framework, and architecture accordingly
2. Write clean, idiomatic code following best practices
3. Include comprehensive error handling
4. Add inline comments explaining complex logic
5. Generate unit tests when appropriate
6. Create deployment configuration (Dockerfile) when needed
7. Use modern language features and libraries

OUTPUT FORMAT:
1. Main source code with complete implementation
2. Unit tests (if requested or beneficial)
3. Dockerfile (if deployment mentioned)
4. Brief README with usage instructions

Context window: 32,000 tokens output (demo limit) - you can generate comprehensive solutions.

Be complete and thorough. Focus on quality and production-readiness."""

def generate_code_with_model(prompt: str, model_name: str, temperature: float = 0.7):
    """Generate code using specified model"""
    config = MODEL_CONFIGS[model_name]

    # Check if API key is available
    api_key = os.getenv(config["api_key_env"])
    if not api_key:
        return {
            "code": f"❌ API Key Missing\n\nPlease configure {config['api_key_env']} in Space settings to use {model_name}.\n\nGo to: https://huggingface.co/spaces/Yatro/Ectus-R_Code_Generation-Demo/settings\nAdd secret: {config['api_key_env']}\n\nFor FREE models (Qwen, DeepSeek, CodeLlama, WizardCoder, StarCoder2), only HF_TOKEN is needed.",
            "elapsed_time": 0,
            "loc": 0,
            "input_tokens": 0,
            "output_tokens": 0,
            "cost": 0,
            "tokens_per_sec": 0
        }

    start_time = time.time()

    try:
        if config["provider"] == "anthropic":
            client = anthropic.Anthropic(api_key=os.getenv(config["api_key_env"]))
            response = client.messages.create(
                model=config["model"],
                max_tokens=32000,  # Limited for demo stability
                temperature=temperature,
                system=SYSTEM_PROMPT,
                messages=[{"role": "user", "content": prompt}]
            )
            generated_code = response.content[0].text
            input_tokens = response.usage.input_tokens
            output_tokens = response.usage.output_tokens

        elif config["provider"] == "openai":
            client = openai.OpenAI(
                api_key=os.getenv(config["api_key_env"]),
                http_client=http_client_no_proxy
            )
            response = client.chat.completions.create(
                model=config["model"],
                messages=[
                    {"role": "system", "content": SYSTEM_PROMPT},
                    {"role": "user", "content": prompt}
                ],
                temperature=temperature,
                max_tokens=16000  # GPT-4o limit
            )
            generated_code = response.choices[0].message.content
            input_tokens = response.usage.prompt_tokens
            output_tokens = response.usage.completion_tokens

        elif config["provider"] == "groq":
            client = Groq(api_key=os.getenv(config["api_key_env"]))
            response = client.chat.completions.create(
                model=config["model"],
                messages=[
                    {"role": "system", "content": SYSTEM_PROMPT},
                    {"role": "user", "content": prompt}
                ],
                temperature=temperature,
                max_tokens=min(8192, config.get("context_window", 8192))  # Use model-specific limit (Gemma2=8192, Llama=32K)
            )
            generated_code = response.choices[0].message.content
            input_tokens = response.usage.prompt_tokens
            output_tokens = response.usage.completion_tokens

        elif config["provider"] == "google":
            # Use Google Generative AI API
            genai.configure(api_key=os.getenv(config["api_key_env"]))
            model = genai.GenerativeModel(config["model"])
            response = model.generate_content(
                f"{SYSTEM_PROMPT}\n\nUser request: {prompt}",
                generation_config={"temperature": temperature, "max_output_tokens": 32000}  # Gemini 2.0 Flash: 1M context, using 32K for demo
            )
            generated_code = response.text
            input_tokens = response.usage_metadata.prompt_token_count
            output_tokens = response.usage_metadata.candidates_token_count

        elif config["provider"] == "github":
            # GitHub Models API (OpenAI-compatible)
            client = openai.OpenAI(
                base_url="https://models.inference.ai.azure.com",
                api_key=os.getenv(config["api_key_env"]),
                http_client=http_client_no_proxy
            )
            response = client.chat.completions.create(
                model=config["model"],
                messages=[
                    {"role": "system", "content": SYSTEM_PROMPT},
                    {"role": "user", "content": prompt}
                ],
                temperature=temperature,
                max_tokens=4096
            )
            generated_code = response.choices[0].message.content
            input_tokens = response.usage.prompt_tokens
            output_tokens = response.usage.completion_tokens

        elif config["provider"] == "huggingface":
            api_url = f"https://api-inference.huggingface.co/models/{config['model']}"
            headers = {"Authorization": f"Bearer {os.getenv(config['api_key_env'])}"}
            payload = {
                "inputs": f"{SYSTEM_PROMPT}\n\nUser: {prompt}\n\nAssistant:",
                "parameters": {
                    "temperature": temperature,
                    "max_new_tokens": 4096,
                    "return_full_text": False
                }
            }
            response = requests.post(api_url, headers=headers, json=payload, timeout=60)

            if response.status_code == 200:
                result = response.json()
                if isinstance(result, list) and len(result) > 0:
                    generated_code = result[0].get("generated_text", "Error: No text generated")
                else:
                    generated_code = str(result)
                # HF doesn't always return token counts
                input_tokens = len(prompt.split()) * 1.3  # estimate
                output_tokens = len(generated_code.split()) * 1.3
            else:
                generated_code = f"Error: HF API returned {response.status_code}\n{response.text}"
                input_tokens = 0
                output_tokens = 0

        else:
            generated_code = f"Error: Unknown provider {config['provider']}"
            input_tokens = 0
            output_tokens = 0

    except Exception as e:
        generated_code = f"Error generating code: {str(e)}"
        input_tokens = 0
        output_tokens = 0

    elapsed_time = time.time() - start_time

    # Calculate metrics
    loc = len(generated_code.split('\n'))
    cost = (input_tokens + output_tokens) / 1_000_000 * config["cost_per_1M_tokens"]
    tokens_per_sec = output_tokens / elapsed_time if elapsed_time > 0 else 0

    return {
        "code": generated_code,
        "elapsed_time": elapsed_time,
        "loc": loc,
        "input_tokens": int(input_tokens),
        "output_tokens": int(output_tokens),
        "cost": cost,
        "tokens_per_sec": tokens_per_sec
    }

def single_model_generation(prompt: str, model: str):
    """Generate code with selected model - pure prompt evaluation"""

    if not prompt.strip():
        return "Please enter a project description."

    # Use prompt directly - let AI decide everything from instructions
    result = generate_code_with_model(prompt, model, temperature=0.7)

    output = f"""# Generated Code: {model}

**Generation Time:** {result['elapsed_time']:.2f}s
**Lines of Code:** {result['loc']}
**Tokens:** {result['input_tokens']} in → {result['output_tokens']} out
**Speed:** {result['tokens_per_sec']:.0f} tokens/sec
**Cost:** ${result['cost']:.4f}

---

{result['code']}
    """

    return output

def multi_model_comparison(prompt: str):
    """Compare all models on same prompt - pure prompt evaluation"""

    if not prompt.strip():
        return pd.DataFrame(), "Please enter a project description."

    # Use prompt directly - let AI decide everything from instructions
    results = []

    for model_name in MODEL_CONFIGS.keys():
        result = generate_code_with_model(prompt, model_name, temperature=0.7)

        results.append({
            "Model": model_name,
            "Time (s)": f"{result['elapsed_time']:.2f}",
            "LOC": result['loc'],
            "Tokens/s": f"{result['tokens_per_sec']:.0f}",
            "Cost ($)": f"{result['cost']:.4f}",
            "Quality": "✅" if result['loc'] > 50 else "⚠️"
        })

    df = pd.DataFrame(results)

    # Find best performers
    df_numeric = df.copy()
    df_numeric['Time (s)'] = df_numeric['Time (s)'].astype(float)
    df_numeric['Tokens/s'] = df_numeric['Tokens/s'].astype(float)

    fastest = df_numeric.loc[df_numeric['Time (s)'].idxmin(), 'Model']
    highest_speed = df_numeric.loc[df_numeric['Tokens/s'].idxmax(), 'Model']
    most_code = df_numeric.loc[df_numeric['LOC'].idxmax(), 'Model']

    summary = f"""## Performance Summary

🏆 **Fastest Generation:** {fastest}
⚡ **Highest Throughput:** {highest_speed}
📝 **Most Code Generated:** {most_code}

**Ectus-R Score:** 173.0/255 (Super-Autónomo)
**QA Success Rate:** 95.6%
**Speed vs Manual:** 50-400x faster
    """

    return df, summary

# Gradio Interface
with gr.Blocks(
    title="Ectus-R Code Generation Demo",
    theme=gr.themes.Soft(primary_hue="purple")
) as demo:

    gr.Markdown("""
    # Ectus-R - Autonomous Software Engineering Platform

    **AGI-AEF Score:** 173.0/255 (Super-Autónomo - Top 5% globally)
    **Powered by AION-R** | **Multi-LLM Orchestration** | **95.6% QA Success Rate**
    """)

    with gr.Tab("🚀 Single Model Generation"):
        gr.Markdown("""
        Generate production-ready code with your choice of AI model.

        **Pure prompt evaluation:** Describe your requirements in detail. The AI will decide language, framework, and architecture based on your instructions.

        **Context Window:** 32,000 tokens output
        """)

        with gr.Row():
            with gr.Column(scale=1):
                prompt_input = gr.Textbox(
                    label="Project Description",
                    placeholder="Example: Create a REST API in Rust using Axum for a blog with users and posts. Include JWT authentication, PostgreSQL database, unit tests, and Docker deployment with multi-stage build.",
                    lines=10,
                    value="Create a minimal REST API for a TODO list with create, read, update, delete operations. Use best practices and include tests."
                )

                model_select = gr.Dropdown(
                    choices=list(MODEL_CONFIGS.keys()),
                    value="Claude Sonnet 4.5 💎",
                    label="AI Model",
                    info="Select the model to generate code"
                )

                generate_btn = gr.Button("Generate Code", variant="primary", size="lg")

            with gr.Column(scale=2):
                output_single = gr.Markdown(
                    value="Generated code will appear here...",
                    line_breaks=True
                )

        generate_btn.click(
            single_model_generation,
            inputs=[prompt_input, model_select],
            outputs=output_single
        )

        gr.Examples(
            examples=[
                ["Create a REST API in Rust using Axum for a blog with users and posts. Include JWT authentication, PostgreSQL database, unit tests, and Docker deployment.", "Claude Sonnet 4.5 💎"],
                ["Build a CLI tool in Python for file encryption using AES-256 with Click framework. Include progress bars and error handling.", "GPT-4o 💎"],
                ["Implement a rate limiter middleware in TypeScript for Express web APIs. Support Redis backend and configurable limits per endpoint.", "Llama 3.3 70B (Groq) 🚀"],
            ],
            inputs=[prompt_input, model_select]
        )

    with gr.Tab("⚡ Multi-Model Comparison"):
        gr.Markdown("""
        Compare all 6 AI models side-by-side on the same task.

        **Pure prompt evaluation:** Each model reads the same instructions and decides implementation details independently.

        **Context Window:** 32,000 tokens output per model
        """)

        with gr.Row():
            with gr.Column(scale=1):
                prompt_compare = gr.Textbox(
                    label="Project Description (tested on ALL models)",
                    placeholder="Example: Create a REST API in Python using FastAPI for a TODO list with create, read, update, delete operations. Include SQLAlchemy models, Pydantic schemas, and basic tests.",
                    lines=8,
                    value="Create a minimal REST API for a TODO list with create, read, update, delete operations. Use best practices and include tests."
                )

                compare_btn = gr.Button("Compare All Models", variant="primary", size="lg")

            with gr.Column(scale=2):
                comparison_table = gr.Dataframe(
                    headers=["Model", "Time (s)", "LOC", "Tokens/s", "Cost ($)", "Quality"],
                    label="Real-time Performance Metrics"
                )
                winner_msg = gr.Markdown()

        compare_btn.click(
            multi_model_comparison,
            inputs=[prompt_compare],
            outputs=[comparison_table, winner_msg]
        )

    with gr.Tab("📊 Benchmarks & Performance"):
        gr.Markdown("""
        ## Real-World Performance Metrics

        ### Ectus-R vs Manual Development

        | Task Type | Ectus-R Time | Manual Time | Speedup | Cost Savings |
        |-----------|-------------|-------------|---------|--------------|
        | Simple REST API | 11.3 seconds | 2-4 hours | **640x faster** | 99.93% |
        | Microservices App | 4 hours | 6 weeks | **240x faster** | 99.88% |
        | Full Stack App | 2 days | 3 months | **45x faster** | 99.74% |

        ### Quality Metrics

        - **QA Success Rate:** 95.6% (tests pass on first generation)
        - **Code Quality:** Industry-standard (linting, formatting, best practices)
        - **Error Rate:** <0.1% (production-ready code)

        ### Multi-LLM Performance Comparison (10 Models)

        | Model | Speed (tok/s) | HumanEval | Quality | Cost | Use Case |
        |-------|---------------|-----------|---------|------|----------|
        | **🏆 Qwen2.5-Coder-32B** | 45 | **92.7%** | 9.5/10 | **FREE** | SOTA code generation |
        | DeepSeek-Coder-V2 | 40 | 90.2% | 9.3/10 | **FREE** | Code optimization |
        | Claude Sonnet 4.5 💎 | 50 | ~85% | 9.7/10 | $3/1M | Complex architecture |
        | GPT-4o 💎 | 65 | 85.4% | 9.5/10 | $2.50/1M | General purpose |
        | CodeLlama-70B | 50 | 67.8% | 7.5/10 | **FREE** | Python/Rust reliable |
        | WizardCoder-Python | 45 | 73.2% | 8.0/10 | **FREE** | Python specialist |
        | StarCoder2-15B | 100 | 72.3% | 7.8/10 | **FREE** | Fast generation |
        | Llama 3.1 70B | 120 | ~65% | 8.8/10 | **FREE** | Fast prototyping |
        | Gemini 2.0 Flash | 150 | ~80% | 9.0/10 | **FREE** | Real-time + 1M context |

        **Key Insight:** Free models (Qwen, DeepSeek) outperform paid GPT-4 on code benchmarks!

        ### Cost Analysis

        **Traditional Development:**
        - Developer salary: $100,000/year = $48/hour
        - Simple API (4 hours): $192

        **Ectus-R:**
        - Claude Sonnet generation: $0.12 (11.3s @ $3/1M tokens)
        - **Savings:** $191.88 (99.93%)

        ---

        ## AGI-AEF Autonomy Assessment

        **Overall Score:** 173.0/255 (67.8% - Super-Autónomo)

        | Dimension | Score | Category |
        |-----------|-------|----------|
        | Adaptability Cognitiva | 20.1/27 | Muy Bueno |
        | Razonamiento Simbólico | 19.8/25 | Muy Bueno |
        | Autonomía Operacional | 22.4/28 | Excelente |
        | Creatividad Generativa | 21.3/24 | Excelente |
        | **Velocidad Procesamiento** | **23.7/27** | **Sobresaliente** |

        **Ranking:** Top 5% globally among code generation systems
        """)

    with gr.Tab("ℹ️ About Ectus-R"):
        gr.Markdown("""
        ## Ectus-R: Autonomous Software Engineering Platform

        Ectus-R is an enterprise-grade platform for transforming business requirements into production-ready code
        through autonomous development processes.

        ### Core Capabilities

        ✅ **10 AI Models** - 2 premium + 5 FREE code specialists + 3 FREE general
        ✅ **FREE SOTA Models** - Qwen2.5-Coder (92.7% HumanEval) beats GPT-4 (85.4%)
        ✅ **Autonomous QA Cycle** - 95.6% success rate (industry-leading)
        ✅ **Full-Stack Generation** - Frontend, backend, databases, infrastructure
        ✅ **DevOps Automation** - Docker, Kubernetes, CI/CD pipelines
        ✅ **50-400x Faster** - Compared to manual development

        ### Technology Stack

        - **Core Engine:** Rust (89%), Python (7%), TypeScript (4%)
        - **Lines of Code:** 142,366 LOC
        - **Powered by:** AION-R AI infrastructure platform
        - **Security:** OWASP Top 10 compliant

        ### Commercial Tiers

        | Tier | Revenue Range | Price | Features |
        |------|--------------|-------|----------|
        | **Startup** | < $1M ARR | **FREE** (MIT) | Unlimited developers, basic support |
        | **Growth** | $1-10M ARR | **$499/month** | Priority support, SLA 99.5% |
        | **Enterprise** | $10M+ ARR | **$2,499/month** | Dedicated support, SLA 99.9%, custom |

        ### Links

        - 🌐 **Website:** [Coming soon]
        - 💻 **GitHub:** [github.com/Yatrogenesis/Ectus-R](https://github.com/Yatrogenesis/Ectus-R)
        - 📚 **Documentation:** [Ectus-R Docs](https://github.com/Yatrogenesis/Ectus-R/blob/main/README.md)
        - 📄 **License:** [MIT / Commercial](https://github.com/Yatrogenesis/Ectus-R/blob/main/LICENSE-COMMERCIAL.md)
        - 📊 **Benchmarks:** [BENCHMARKS.md](https://github.com/Yatrogenesis/Ectus-R/blob/main/BENCHMARKS.md)

        ### Support

        - 💬 **Community:** [GitHub Discussions](https://github.com/Yatrogenesis/Ectus-R/discussions)
        - 🐛 **Issues:** [GitHub Issues](https://github.com/Yatrogenesis/Ectus-R/issues)
        - 📧 **Enterprise:** enterprise@yatrogenesis.com

        ---

        **Built with Rust** • **Powered by AION-R** • **Enterprise-Ready**

        *Ectus-R: The future of autonomous software engineering*
        """)

# Launch with optimized settings
demo.launch(
    server_name="0.0.0.0",
    server_port=7860,
    show_error=True,
    show_api=False
)