app.py

"""
MOUSE Workflow - Visual Workflow Builder with UI Execution
@Powered by VIDraft
✓ Visual workflow designer with drag-and-drop
✓ Import/Export JSON with copy-paste support
✓ Auto-generate UI from workflow for end-user execution
"""

import os, json, typing, tempfile, traceback
import gradio as gr
from gradio_workflowbuilder import WorkflowBuilder

# Optional imports for LLM APIs
try:
    from openai import OpenAI
    OPENAI_AVAILABLE = True
except ImportError:
    OPENAI_AVAILABLE = False
    print("OpenAI library not available. Install with: pip install openai")

# Anthropic 관련 코드 주석 처리
# try:
#     import anthropic
#     ANTHROPIC_AVAILABLE = True
# except ImportError:
#     ANTHROPIC_AVAILABLE = False
#     print("Anthropic library not available. Install with: pip install anthropic")
ANTHROPIC_AVAILABLE = False

try:
    import requests
    REQUESTS_AVAILABLE = True
except ImportError:
    REQUESTS_AVAILABLE = False
    print("Requests library not available. Install with: pip install requests")

try:
    from huggingface_hub import HfApi, create_repo, upload_file
    HF_HUB_AVAILABLE = True
except ImportError:
    HF_HUB_AVAILABLE = False
    print("Huggingface Hub not available. Install with: pip install huggingface-hub")


# ───────────────────────────────────────────────────────────
from gradio_workflowbuilder import WorkflowBuilder as _WB

if not getattr(_WB, "_patched_for_custom_palette", False):
    # 1) __init__ 확장 ─ custom_palette 보관
    _orig_init = _WB.__init__
    def _patched_init(self, *args, custom_palette=None, **kwargs):
        self._custom_palette = custom_palette or []
        _orig_init(self, *args, **kwargs)
    _WB.__init__ = _patched_init

    # 2) config 생성 시 팔레트 합치기
    cfg_method = "_get_config" if hasattr(_WB, "_get_config") else "get_config"
    _orig_get_cfg = getattr(_WB, cfg_method)

    def _patched_get_cfg(self, *args, **kwargs):
        cfg = _orig_get_cfg(self, *args, **kwargs)
        if getattr(self, "_custom_palette", None):
            cfg["palette"] = cfg.get("palette", []) + self._custom_palette
        return cfg

    setattr(_WB, cfg_method, _patched_get_cfg)
    _WB._patched_for_custom_palette = True
# ───────────────────────────────────────────────────────────


import json, pathlib

palette_path = pathlib.Path(__file__).parent / "best_ai_palette.json"
with open(palette_path, "r", encoding="utf-8") as f:
    best_ai_palette = json.load(f)

# -------------------------------------------------------------------
# 🛠️  헬퍼 함수들
# -------------------------------------------------------------------
def export_pretty(data: typing.Dict[str, typing.Any]) -> str:
    return json.dumps(data, indent=2, ensure_ascii=False) if data else "No workflow to export"

def export_file(data: typing.Dict[str, typing.Any]) -> typing.Optional[str]:
    """워크플로우를 JSON 파일로 내보내기"""
    if not data:
        return None
    
    try:
        # 임시 파일 생성
        fd, path = tempfile.mkstemp(suffix=".json", prefix="workflow_", text=True)
        with os.fdopen(fd, "w", encoding="utf-8") as f:
            json.dump(data, f, ensure_ascii=False, indent=2)
        return path
    except Exception as e:
        print(f"Error exporting file: {e}")
        return None

def load_json_from_text_or_file(json_text: str, file_obj) -> typing.Tuple[typing.Dict[str, typing.Any], str]:
    """텍스트 또는 파일에서 JSON 로드"""
    # 파일이 있으면 파일 우선
    if file_obj is not None:
        try:
            with open(file_obj.name, "r", encoding="utf-8") as f:
                json_text = f.read()
        except Exception as e:
            return None, f"❌ Error reading file: {str(e)}"
    
    # JSON 텍스트가 없거나 비어있으면
    if not json_text or json_text.strip() == "":
        return None, "No JSON data provided"
    
    try:
        # JSON 파싱
        data = json.loads(json_text.strip())
        
        # 데이터 검증
        if not isinstance(data, dict):
            return None, "Invalid format: not a dictionary"
        
        # 필수 필드 확인
        if 'nodes' not in data:
            data['nodes'] = []
        if 'edges' not in data:
            data['edges'] = []
            
        nodes_count = len(data.get('nodes', []))
        edges_count = len(data.get('edges', []))
        
        return data, f"✅ Loaded: {nodes_count} nodes, {edges_count} edges"
        
    except json.JSONDecodeError as e:
        return None, f"❌ JSON parsing error: {str(e)}"
    except Exception as e:
        return None, f"❌ Error: {str(e)}"

def create_sample_workflow(example_type="basic"):
    """샘플 워크플로우 생성"""
    
    if example_type == "basic":
        # 기본 예제: 간단한 Q&A - VIDraft 사용
        return {
            "nodes": [

                {   # 🆕 캔버스에 기본으로 배치될 Best-AI 노드
                    "id": "best_ai_default",
                    "type": "llmNode",
                    "position": {"x": 80, "y": 40},
                    "data": {
                        "label": "AI Processing",
                        "template": {
                            "provider":     {"value": "VIDraft"},
                            "model":        {"value": "Gemma-3-r1984-27B"},
                            "temperature":  {"value": 0.7},
                            "system_prompt":{"value": "You are a helpful assistant."}
                        }
                    }
                },
    

                
                {
                    "id": "input_1",
                    "type": "ChatInput",
                    "position": {"x": 100, "y": 200},
                    "data": {
                        "label": "User Question",
                        "template": {
                            "input_value": {"value": "What is the capital of Korea?"}
                        }
                    }
                },
                {
                    "id": "llm_1",
                    "type": "llmNode",
                    "position": {"x": 400, "y": 200},
                    "data": {
                        "label": "AI Processing",
                        "template": {
                            "provider": {"value": "VIDraft"},  # 기본값을 VIDraft로 변경
                            "model": {"value": "Gemma-3-r1984-27B"},
                            "temperature": {"value": 0.7},
                            "system_prompt": {"value": "You are a helpful assistant."}
                        }
                    }
                },
                {
                    "id": "output_1",
                    "type": "ChatOutput",
                    "position": {"x": 700, "y": 200},
                    "data": {"label": "Answer"}
                }
            ],
            "edges": [
                {"id": "e1", "source": "input_1", "target": "llm_1"},
                {"id": "e2", "source": "llm_1", "target": "output_1"}
            ]
        }
    
    elif example_type == "vidraft":
        # VIDraft 예제
        return {
            "nodes": [
                {
                    "id": "input_1",
                    "type": "ChatInput",
                    "position": {"x": 100, "y": 200},
                    "data": {
                        "label": "User Input",
                        "template": {
                            "input_value": {"value": "AI와 머신러닝의 차이점을 설명해주세요."}
                        }
                    }
                },
                {
                    "id": "llm_1",
                    "type": "llmNode",
                    "position": {"x": 400, "y": 200},
                    "data": {
                        "label": "VIDraft AI (Gemma)",
                        "template": {
                            "provider": {"value": "VIDraft"},
                            "model": {"value": "Gemma-3-r1984-27B"},
                            "temperature": {"value": 0.8},
                            "system_prompt": {"value": "당신은 전문적이고 친절한 AI 교육자입니다. 복잡한 개념을 쉽게 설명해주세요."}
                        }
                    }
                },
                {
                    "id": "output_1",
                    "type": "ChatOutput",
                    "position": {"x": 700, "y": 200},
                    "data": {"label": "AI Explanation"}
                }
            ],
            "edges": [
                {"id": "e1", "source": "input_1", "target": "llm_1"},
                {"id": "e2", "source": "llm_1", "target": "output_1"}
            ]
        }
    
    elif example_type == "multi_input":
        # 다중 입력 예제
        return {
            "nodes": [
                {
                    "id": "name_input",
                    "type": "textInput",
                    "position": {"x": 100, "y": 100},
                    "data": {
                        "label": "Your Name",
                        "template": {
                            "input_value": {"value": "John"}
                        }
                    }
                },
                {
                    "id": "topic_input",
                    "type": "textInput",
                    "position": {"x": 100, "y": 250},
                    "data": {
                        "label": "Topic",
                        "template": {
                            "input_value": {"value": "Python programming"}
                        }
                    }
                },
                {
                    "id": "level_input",
                    "type": "textInput",
                    "position": {"x": 100, "y": 400},
                    "data": {
                        "label": "Skill Level",
                        "template": {
                            "input_value": {"value": "beginner"}
                        }
                    }
                },
                {
                    "id": "combiner",
                    "type": "textNode",
                    "position": {"x": 350, "y": 250},
                    "data": {
                        "label": "Combine Inputs",
                        "template": {
                            "text": {"value": "Create a personalized learning plan"}
                        }
                    }
                },
                {
                    "id": "llm_1",
                    "type": "llmNode",
                    "position": {"x": 600, "y": 250},
                    "data": {
                        "label": "Generate Learning Plan",
                        "template": {
                            "provider": {"value": "VIDraft"},  # 기본값을 VIDraft로 변경
                            "model": {"value": "Gemma-3-r1984-27B"},
                            "temperature": {"value": 0.7},
                            "system_prompt": {"value": "You are an expert educational consultant. Create personalized learning plans based on the user's name, topic of interest, and skill level."}
                        }
                    }
                },
                {
                    "id": "output_1",
                    "type": "ChatOutput",
                    "position": {"x": 900, "y": 250},
                    "data": {"label": "Your Learning Plan"}
                }
            ],
            "edges": [
                {"id": "e1", "source": "name_input", "target": "combiner"},
                {"id": "e2", "source": "topic_input", "target": "combiner"},
                {"id": "e3", "source": "level_input", "target": "combiner"},
                {"id": "e4", "source": "combiner", "target": "llm_1"},
                {"id": "e5", "source": "llm_1", "target": "output_1"}
            ]
        }
    
    elif example_type == "chain":
        # 체인 처리 예제
        return {
            "nodes": [
                {
                    "id": "input_1",
                    "type": "ChatInput",
                    "position": {"x": 50, "y": 200},
                    "data": {
                        "label": "Original Text",
                        "template": {
                            "input_value": {"value": "The quick brown fox jumps over the lazy dog."}
                        }
                    }
                },
                {
                    "id": "translator",
                    "type": "llmNode",
                    "position": {"x": 300, "y": 200},
                    "data": {
                        "label": "Translate to Korean",
                        "template": {
                            "provider": {"value": "VIDraft"},
                            "model": {"value": "Gemma-3-r1984-27B"},
                            "temperature": {"value": 0.3},
                            "system_prompt": {"value": "You are a professional translator. Translate the given English text to Korean accurately."}
                        }
                    }
                },
                {
                    "id": "analyzer",
                    "type": "llmNode",
                    "position": {"x": 600, "y": 200},
                    "data": {
                        "label": "Analyze Translation",
                        "template": {
                            "provider": {"value": "OpenAI"},
                            "model": {"value": "gpt-4.1-mini"},
                            "temperature": {"value": 0.5},
                            "system_prompt": {"value": "You are a linguistic expert. Analyze the Korean translation and explain its nuances and cultural context."}
                        }
                    }
                },
                {
                    "id": "output_translation",
                    "type": "ChatOutput",
                    "position": {"x": 450, "y": 350},
                    "data": {"label": "Korean Translation"}
                },
                {
                    "id": "output_analysis",
                    "type": "ChatOutput",
                    "position": {"x": 900, "y": 200},
                    "data": {"label": "Translation Analysis"}
                }
            ],
            "edges": [
                {"id": "e1", "source": "input_1", "target": "translator"},
                {"id": "e2", "source": "translator", "target": "analyzer"},
                {"id": "e3", "source": "translator", "target": "output_translation"},
                {"id": "e4", "source": "analyzer", "target": "output_analysis"}
            ]
        }
    
    # 기본값은 basic
    return create_sample_workflow("basic")

# 배포를 위한 독립 앱 생성 함수
def generate_standalone_app(workflow_data: dict, app_name: str, app_description: str) -> str:
    """워크플로우를 독립적인 Gradio 앱으로 변환"""
    
    # JSON 데이터를 문자열로 변환
    workflow_json = json.dumps(workflow_data, indent=2)
    
    app_code = f'''"""
{app_name}
{app_description}
Generated by MOUSE Workflow
"""

import os
import json
import gradio as gr
import requests

# Workflow configuration
WORKFLOW_DATA = {workflow_json}

def execute_workflow(*input_values):
    """Execute the workflow with given inputs"""
    
    # API keys from environment
    vidraft_token = os.getenv("FRIENDLI_TOKEN")
    openai_key = os.getenv("OPENAI_API_KEY")
    
    nodes = WORKFLOW_DATA.get("nodes", [])
    edges = WORKFLOW_DATA.get("edges", [])
    
    results = {{}}
    
    # Get input nodes
    input_nodes = [n for n in nodes if n.get("type") in ["ChatInput", "textInput", "Input", "numberInput"]]
    
    # Map inputs to node IDs
    for i, node in enumerate(input_nodes):
        if i < len(input_values):
            results[node["id"]] = input_values[i]
    
    # Process nodes
    for node in nodes:
        node_id = node.get("id")
        node_type = node.get("type", "")
        node_data = node.get("data", {{}})
        template = node_data.get("template", {{}})
        
        if node_type == "textNode":
            # Combine connected inputs
            base_text = template.get("text", {{}}).get("value", "")
            connected_inputs = []
            
            for edge in edges:
                if edge.get("target") == node_id:
                    source_id = edge.get("source")
                    if source_id in results:
                        connected_inputs.append(f"{{source_id}}: {{results[source_id]}}")
            
            if connected_inputs:
                results[node_id] = f"{{base_text}}\\n\\nInputs:\\n" + "\\n".join(connected_inputs)
            else:
                results[node_id] = base_text
                
        elif node_type in ["llmNode", "OpenAIModel", "ChatModel"]:
            # Get provider and model - VIDraft as default
            provider = template.get("provider", {{}}).get("value", "VIDraft")
            if provider not in ["VIDraft", "OpenAI"]:
                provider = "VIDraft"  # Default to VIDraft
            temperature = template.get("temperature", {{}}).get("value", 0.7)
            system_prompt = template.get("system_prompt", {{}}).get("value", "")
            
            # Get input text
            input_text = ""
            for edge in edges:
                if edge.get("target") == node_id:
                    source_id = edge.get("source")
                    if source_id in results:
                        input_text = results[source_id]
                        break
            
            # Call API
            if provider == "OpenAI" and openai_key:
                try:
                    from openai import OpenAI
                    client = OpenAI(api_key=openai_key)
                    
                    messages = []
                    if system_prompt:
                        messages.append({{"role": "system", "content": system_prompt}})
                    messages.append({{"role": "user", "content": input_text}})
                    
                    response = client.chat.completions.create(
                        model="gpt-4.1-mini",
                        messages=messages,
                        temperature=temperature,
                        max_tokens=1000
                    )
                    
                    results[node_id] = response.choices[0].message.content
                except Exception as e:
                    results[node_id] = f"[OpenAI Error: {{str(e)}}]"
                    
            elif provider == "VIDraft" and vidraft_token:
                try:
                    headers = {{
                        "Authorization": f"Bearer {{vidraft_token}}",
                        "Content-Type": "application/json"
                    }}
                    
                    messages = []
                    if system_prompt:
                        messages.append({{"role": "system", "content": system_prompt}})
                    messages.append({{"role": "user", "content": input_text}})
                    
                    payload = {{
                        "model": "dep89a2fld32mcm",
                        "messages": messages,
                        "max_tokens": 16384,
                        "temperature": temperature,
                        "top_p": 0.8,
                        "stream": False
                    }}
                    
                    response = requests.post(
                        "https://api.friendli.ai/dedicated/v1/chat/completions",
                        headers=headers,
                        json=payload,
                        timeout=30
                    )
                    
                    if response.status_code == 200:
                        results[node_id] = response.json()["choices"][0]["message"]["content"]
                    else:
                        results[node_id] = f"[VIDraft Error: {{response.status_code}}]"
                except Exception as e:
                    results[node_id] = f"[VIDraft Error: {{str(e)}}]"
            else:
                # Show which API key is missing
                if provider == "OpenAI":
                    results[node_id] = "[OpenAI API key not found. Please set OPENAI_API_KEY in Space secrets]"
                elif provider == "VIDraft":
                    results[node_id] = "[VIDraft API key not found. Please set FRIENDLI_TOKEN in Space secrets]"
                else:
                    results[node_id] = f"[No API key found for {{provider}}. Using simulated response: {{input_text[:50]}}...]"
                
        elif node_type in ["ChatOutput", "textOutput", "Output"]:
            # Get connected result
            for edge in edges:
                if edge.get("target") == node_id:
                    source_id = edge.get("source")
                    if source_id in results:
                        results[node_id] = results[source_id]
                        break
    
    # Return outputs
    output_nodes = [n for n in nodes if n.get("type") in ["ChatOutput", "textOutput", "Output"]]
    return [results.get(n["id"], "") for n in output_nodes]

# Build UI
with gr.Blocks(title="{app_name}", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# {app_name}")
    gr.Markdown("{app_description}")
    
    # API Status Check
    vidraft_token = os.getenv("FRIENDLI_TOKEN")
    openai_key = os.getenv("OPENAI_API_KEY")
    
    with gr.Accordion("🔑 API Status", open=False):
        if vidraft_token:
            gr.Markdown("✅ **VIDraft API**: Connected (Gemma-3-r1984-27B)")
        else:
            gr.Markdown("❌ **VIDraft API**: Not configured")
            
        if openai_key:
            gr.Markdown("✅ **OpenAI API**: Connected (gpt-4.1-mini)")
        else:
            gr.Markdown("⚠️ **OpenAI API**: Not configured (optional)")
        
        if not vidraft_token:
            gr.Markdown("""
            **⚠️ Important**: Please add FRIENDLI_TOKEN to Space secrets for the app to work properly.
            
            Go to: Space settings → Repository secrets → Add secret
            """)
        elif not openai_key:
            gr.Markdown("""
            **💡 Tip**: The app will work with VIDraft alone. Add OPENAI_API_KEY if you need OpenAI features.
            """)
        else:
            gr.Markdown("**✨ All APIs configured! Your app is fully functional.**")
    
    # Extract nodes
    nodes = WORKFLOW_DATA.get("nodes", [])
    input_nodes = [n for n in nodes if n.get("type") in ["ChatInput", "textInput", "Input", "numberInput"]]
    output_nodes = [n for n in nodes if n.get("type") in ["ChatOutput", "textOutput", "Output"]]
    
    # Create inputs
    inputs = []
    if input_nodes:
        gr.Markdown("### 📥 Inputs")
        for node in input_nodes:
            label = node.get("data", {{}}).get("label", node.get("id"))
            template = node.get("data", {{}}).get("template", {{}})
            default_value = template.get("input_value", {{}}).get("value", "")
            
            if node.get("type") == "numberInput":
                inp = gr.Number(label=label, value=float(default_value) if default_value else 0)
            else:
                inp = gr.Textbox(label=label, value=default_value, lines=2)
            inputs.append(inp)
    
    # Execute button
    btn = gr.Button("🚀 Execute Workflow", variant="primary")
    
    # Create outputs
    outputs = []
    if output_nodes:
        gr.Markdown("### 📤 Outputs")
        for node in output_nodes:
            label = node.get("data", {{}}).get("label", node.get("id"))
            out = gr.Textbox(label=label, interactive=False, lines=3)
            outputs.append(out)
    
    # Connect
    btn.click(fn=execute_workflow, inputs=inputs, outputs=outputs)
    
    gr.Markdown("---")
    gr.Markdown("*Powered by MOUSE Workflow*")

if __name__ == "__main__":
    demo.launch()
'''
    
    return app_code

def generate_requirements_txt() -> str:
    """Generate requirements.txt for the standalone app"""
    return """gradio==5.34.2
openai
requests
"""

def deploy_to_huggingface(workflow_data: dict, app_name: str, app_description: str, 
                         hf_token: str, space_name: str, is_private: bool = False,
                         api_keys: dict = None) -> dict:
    """Deploy workflow to Hugging Face Space with API keys"""
    
    if not HF_HUB_AVAILABLE:
        return {"success": False, "error": "huggingface-hub library not installed"}
    
    if api_keys is None:
        api_keys = {}
    
    try:
        # Initialize HF API
        api = HfApi(token=hf_token)
        
        # Create repository
        repo_id = api.create_repo(
            repo_id=space_name,
            repo_type="space",
            space_sdk="gradio",
            private=is_private,
            exist_ok=True
        )
        
        # Detect which providers are used in the workflow
        providers_used = set()
        nodes = workflow_data.get("nodes", [])
        for node in nodes:
            if node.get("type") in ["llmNode", "OpenAIModel", "ChatModel"]:
                template = node.get("data", {}).get("template", {})
                provider = template.get("provider", {}).get("value", "")
                if provider:
                    providers_used.add(provider)
        
        # Generate files
        app_code = generate_standalone_app(workflow_data, app_name, app_description)
        requirements = generate_requirements_txt()
        
        # README with API setup instructions
        api_status = []
        if "FRIENDLI_TOKEN" in api_keys and api_keys["FRIENDLI_TOKEN"]:
            api_status.append("- **FRIENDLI_TOKEN**: ✅ Will be configured automatically")
        else:
            api_status.append("- **FRIENDLI_TOKEN**: ⚠️ Not provided (VIDraft won't work)")
            
        if "OPENAI_API_KEY" in api_keys and api_keys["OPENAI_API_KEY"]:
            api_status.append("- **OPENAI_API_KEY**: ✅ Will be configured automatically")
        elif "OpenAI" in providers_used:
            api_status.append("- **OPENAI_API_KEY**: ❌ Required but not provided")
        
        readme = f"""---
title: {app_name}
emoji: 🐭
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.34.2
app_file: app.py
pinned: false
---

# {app_name}

{app_description}

## 🔑 API Configuration Status

{chr(10).join(api_status)}

## 📝 Providers Used in This Workflow

{', '.join(providers_used) if providers_used else 'No LLM providers detected'}

## 🚀 Default Configuration

This app is configured to use **VIDraft (Gemma-3-r1984-27B)** as the default LLM provider for optimal performance.

---
Generated by MOUSE Workflow
"""
        
        # Upload files
        api.upload_file(
            path_or_fileobj=app_code.encode(),
            path_in_repo="app.py",
            repo_id=repo_id.repo_id,
            repo_type="space"
        )
        
        api.upload_file(
            path_or_fileobj=requirements.encode(),
            path_in_repo="requirements.txt",
            repo_id=repo_id.repo_id,
            repo_type="space"
        )
        
        api.upload_file(
            path_or_fileobj=readme.encode(),
            path_in_repo="README.md",
            repo_id=repo_id.repo_id,
            repo_type="space"
        )
        
        # Add all provided API keys as secrets
        added_secrets = []
        failed_secrets = []
        
        for key_name, key_value in api_keys.items():
            if key_value:  # Only add non-empty keys
                try:
                    api.add_space_secret(
                        repo_id=repo_id.repo_id,
                        key=key_name,
                        value=key_value
                    )
                    added_secrets.append(key_name)
                except Exception as e:
                    failed_secrets.append(f"{key_name}: {str(e)}")
                    print(f"Warning: Could not add {key_name} secret: {e}")
        
        space_url = f"https://huggingface.co/spaces/{repo_id.repo_id}"
        
        return {
            "success": True,
            "space_url": space_url,
            "message": f"Successfully deployed to {space_url}",
            "added_secrets": added_secrets,
            "failed_secrets": failed_secrets,
            "providers_used": list(providers_used)
        }
        
    except Exception as e:
        return {
            "success": False,
            "error": str(e)
        }

# UI 실행을 위한 실제 워크플로우 실행 함수
def execute_workflow_simple(workflow_data: dict, input_values: dict) -> dict:
    """워크플로우 실제 실행"""
    import traceback
    
    # API 키 확인
    vidraft_token = os.getenv("FRIENDLI_TOKEN")  # VIDraft/Friendli token
    openai_key = os.getenv("OPENAI_API_KEY")
    # anthropic_key = os.getenv("ANTHROPIC_API_KEY")  # 주석 처리
    
    # OpenAI 라이브러리 확인
    try:
        from openai import OpenAI
        openai_available = True
    except ImportError:
        openai_available = False
        print("OpenAI library not available")
    
    # Anthropic 라이브러리 확인 - 주석 처리
    # try:
    #     import anthropic
    #     anthropic_available = True
    # except ImportError:
    #     anthropic_available = False
    #     print("Anthropic library not available")
    anthropic_available = False
    
    results = {}
    nodes = workflow_data.get("nodes", [])
    edges = workflow_data.get("edges", [])
    
    # 노드를 순서대로 처리
    for node in nodes:
        node_id = node.get("id")
        node_type = node.get("type", "")
        node_data = node.get("data", {})
        
        try:
            if node_type in ["ChatInput", "textInput", "Input"]:
                # UI에서 제공된 입력값 사용
                if node_id in input_values:
                    results[node_id] = input_values[node_id]
                else:
                    # 기본값 사용
                    template = node_data.get("template", {})
                    default_value = template.get("input_value", {}).get("value", "")
                    results[node_id] = default_value
            
            elif node_type == "textNode":
                # 텍스트 노드는 연결된 모든 입력을 결합
                template = node_data.get("template", {})
                base_text = template.get("text", {}).get("value", "")
                
                # 연결된 입력들 수집
                connected_inputs = []
                for edge in edges:
                    if edge.get("target") == node_id:
                        source_id = edge.get("source")
                        if source_id in results:
                            connected_inputs.append(f"{source_id}: {results[source_id]}")
                
                # 결합된 텍스트 생성
                if connected_inputs:
                    combined_text = f"{base_text}\n\nInputs:\n" + "\n".join(connected_inputs)
                    results[node_id] = combined_text
                else:
                    results[node_id] = base_text
            
            elif node_type in ["llmNode", "OpenAIModel", "ChatModel"]:
                # LLM 노드 처리
                template = node_data.get("template", {})
                
                # 프로바이더 정보 추출 - VIDraft 또는 OpenAI만 허용
                provider_info = template.get("provider", {})
                provider = provider_info.get("value", "VIDraft") if isinstance(provider_info, dict) else "VIDraft"  # 기본값 VIDraft
                
                # provider가 VIDraft 또는 OpenAI가 아닌 경우 VIDraft로 기본 설정
                if provider not in ["VIDraft", "OpenAI"]:
                    provider = "VIDraft"
                
                # 모델 정보 추출
                if provider == "OpenAI":
                    # OpenAI는 gpt-4.1-mini로 고정
                    model = "gpt-4.1-mini"
                elif provider == "VIDraft":
                    # VIDraft는 Gemma-3-r1984-27B로 고정
                    model = "Gemma-3-r1984-27B"
                else:
                    model = "Gemma-3-r1984-27B"  # 기본값 VIDraft 모델
                
                # 온도 정보 추출
                temp_info = template.get("temperature", {})
                temperature = temp_info.get("value", 0.7) if isinstance(temp_info, dict) else 0.7
                
                # 시스템 프롬프트 추출
                prompt_info = template.get("system_prompt", {})
                system_prompt = prompt_info.get("value", "") if isinstance(prompt_info, dict) else ""
                
                # 입력 텍스트 찾기
                input_text = ""
                for edge in edges:
                    if edge.get("target") == node_id:
                        source_id = edge.get("source")
                        if source_id in results:
                            input_text = results[source_id]
                            break
                
                # 실제 API 호출
                if provider == "OpenAI" and openai_key and openai_available:
                    try:
                        client = OpenAI(api_key=openai_key)
                        
                        messages = []
                        if system_prompt:
                            messages.append({"role": "system", "content": system_prompt})
                        messages.append({"role": "user", "content": input_text})
                        
                        response = client.chat.completions.create(
                            model="gpt-4.1-mini",  # 고정된 모델명
                            messages=messages,
                            temperature=temperature,
                            max_tokens=1000
                        )
                        
                        results[node_id] = response.choices[0].message.content
                        
                    except Exception as e:
                        results[node_id] = f"[OpenAI Error: {str(e)}]"
                
                # Anthropic 관련 코드 주석 처리
                # elif provider == "Anthropic" and anthropic_key and anthropic_available:
                #     try:
                #         client = anthropic.Anthropic(api_key=anthropic_key)
                #         
                #         message = client.messages.create(
                #             model="claude-3-haiku-20240307",
                #             max_tokens=1000,
                #             temperature=temperature,
                #             system=system_prompt if system_prompt else None,
                #             messages=[{"role": "user", "content": input_text}]
                #         )
                #         
                #         results[node_id] = message.content[0].text
                #         
                #     except Exception as e:
                #         results[node_id] = f"[Anthropic Error: {str(e)}]"
                
                elif provider == "VIDraft" and vidraft_token:
                    try:
                        import requests
                        
                        headers = {
                            "Authorization": f"Bearer {vidraft_token}",
                            "Content-Type": "application/json"
                        }
                        
                        # 메시지 구성
                        messages = []
                        if system_prompt:
                            messages.append({"role": "system", "content": system_prompt})
                        messages.append({"role": "user", "content": input_text})
                        
                        payload = {
                            "model": "dep89a2fld32mcm",  # VIDraft 모델 ID
                            "messages": messages,
                            "max_tokens": 16384,
                            "temperature": temperature,
                            "top_p": 0.8,
                            "stream": False  # 동기 실행을 위해 False로 설정
                        }
                        
                        # VIDraft API endpoint
                        response = requests.post(
                            "https://api.friendli.ai/dedicated/v1/chat/completions",
                            headers=headers,
                            json=payload,
                            timeout=30
                        )
                        
                        if response.status_code == 200:
                            response_json = response.json()
                            results[node_id] = response_json["choices"][0]["message"]["content"]
                        else:
                            results[node_id] = f"[VIDraft API Error: {response.status_code} - {response.text}]"
                            
                    except Exception as e:
                        results[node_id] = f"[VIDraft Error: {str(e)}]"
                
                else:
                    # API 키가 없는 경우 시뮬레이션
                    results[node_id] = f"[Simulated {provider} Response to: {input_text[:50]}...]"
            
            elif node_type in ["ChatOutput", "textOutput", "Output"]:
                # 출력 노드는 연결된 노드의 결과를 가져옴
                for edge in edges:
                    if edge.get("target") == node_id:
                        source_id = edge.get("source")
                        if source_id in results:
                            results[node_id] = results[source_id]
                            break
                        
        except Exception as e:
            results[node_id] = f"[Node Error: {str(e)}]"
            print(f"Error processing node {node_id}: {traceback.format_exc()}")
    
    return results

# -------------------------------------------------------------------
# 🎨 CSS
# -------------------------------------------------------------------
CSS = """
.main-container{max-width:1600px;margin:0 auto;}
.workflow-section{margin-bottom:2rem;min-height:500px;}
.button-row{display:flex;gap:1rem;justify-content:center;margin:1rem 0;}
.status-box{
    padding:10px;border-radius:5px;margin-top:10px;
    background:#f0f9ff;border:1px solid #3b82f6;color:#1e40af;
}
.component-description{
  padding:24px;background:linear-gradient(135deg,#f8fafc 0%,#e2e8f0 100%);
  border-left:4px solid #3b82f6;border-radius:12px;
  box-shadow:0 2px 8px rgba(0,0,0,.05);margin:16px 0;
}
.workflow-container{position:relative;}
.ui-execution-section{
    background:linear-gradient(135deg,#f0fdf4 0%,#dcfce7 100%);
    padding:24px;border-radius:12px;margin:24px 0;
    border:1px solid #86efac;
}
.powered-by{
    text-align:center;color:#64748b;font-size:14px;
    margin-top:8px;font-style:italic;
}
.sample-buttons{
    display:grid;grid-template-columns:1fr 1fr;gap:0.5rem;
    margin-top:0.5rem;
}
.deploy-section{
    background:linear-gradient(135deg,#fef3c7 0%,#fde68a 100%);
    padding:24px;border-radius:12px;margin:24px 0;
    border:1px solid #fbbf24;
}
.save-indicator{
    text-align:right;
    font-size:14px;
    color:#16a34a;
    padding:8px 16px;
    background:#f0fdf4;
    border-radius:20px;
    display:inline-block;
    margin-left:auto;
}
.workflow-info{
    font-size:14px;
    color:#475569;
    background:#f8fafc;
    padding:8px 16px;
    border-radius:8px;
    display:inline-block;
    margin-bottom:16px;
}
"""

# -------------------------------------------------------------------
# 🖥️  Gradio 앱
# -------------------------------------------------------------------
with gr.Blocks(title="🐭 MOUSE Workflow", theme=gr.themes.Soft(), css=CSS) as demo:
    
    with gr.Column(elem_classes=["main-container"]):
        gr.Markdown("# 🐭 MOUSE Workflow")
        gr.Markdown("**Visual Workflow Builder with Interactive UI Execution**")
        gr.HTML('<p class="powered-by">@Powered by VIDraft & Huggingface gradio</p>')
        
        html_content = """<div class="component-description">
            <p style="font-size:16px;margin:0;">Build sophisticated workflows visually • Import/Export JSON • Generate interactive UI for end-users • Default LLM: VIDraft (Gemma-3-r1984-27B)</p>
            <p style="font-size:14px;margin-top:8px;color:#64748b;">💡 Tip: Your workflow is automatically saved as you make changes. The JSON preview updates in real-time!</p>
        </div>"""
        gr.HTML(html_content)
        
        # API Status Display
        with gr.Accordion("🔌 API Status", open=False):
            gr.Markdown(f"""
            **Available APIs:**
            - FRIENDLI_TOKEN (VIDraft): {'✅ Connected' if os.getenv("FRIENDLI_TOKEN") else '❌ Not found'}
            - OPENAI_API_KEY: {'✅ Connected' if os.getenv("OPENAI_API_KEY") else '❌ Not found'}
            
            **Libraries:**
            - OpenAI: {'✅ Installed' if OPENAI_AVAILABLE else '❌ Not installed'}
            - Requests: {'✅ Installed' if REQUESTS_AVAILABLE else '❌ Not installed'}
            - Hugging Face Hub: {'✅ Installed' if HF_HUB_AVAILABLE else '❌ Not installed (needed for deployment)'}
            
            **Available Models:**
            - OpenAI: gpt-4.1-mini (fixed)
            - VIDraft: Gemma-3-r1984-27B (model ID: dep89a2fld32mcm)
            
            **Sample Workflows:**
            - Basic Q&A: Simple question-answer flow (VIDraft)
            - VIDraft: Korean language example with Gemma model
            - Multi-Input: Combine multiple inputs for personalized output (VIDraft)
            - Chain: Sequential processing with multiple outputs (VIDraft + OpenAI)
            
            **Note**: All examples prioritize VIDraft for optimal performance. Friendli API token will be automatically configured during deployment.
            """)
        
        # State for storing workflow data
        loaded_data = gr.State(None)
        trigger_update = gr.State(False)
        save_status = gr.State("Ready")




        
# ─── Dynamic Workflow Container ─
        # ─── Dynamic Workflow Container ───
        with gr.Column(elem_classes=["workflow-container"]):

            # Auto-save status indicator
            with gr.Row():
                gr.Markdown("### 🎨 Visual Workflow Designer")
                save_indicator = gr.Markdown("💾 Auto-save: Ready",
                                             elem_classes=["save-indicator"])

            @gr.render(inputs=[loaded_data, trigger_update])
            def render_workflow(data, trigger):
                """
                동적으로 WorkflowBuilder 렌더링 + 10초 자동 저장
                """
                workflow_value = data if data else {"nodes": [], "edges": []}

                # WorkflowBuilder (🔌 custom_palette 제거)
                wb = WorkflowBuilder(
                    label="",
                    info="Drag nodes → Connect edges → Edit properties → Auto-save!",
                    value=workflow_value,
                    elem_id="main_workflow",
                    custom_palette=best_ai_palette   # ← 새로 추가

                )

                # ---------- 저장 로직 ----------
                def periodic_save(workflow_data):
                    import json, time, copy
                    if isinstance(workflow_data, str):
                        try:
                            workflow_data = json.loads(workflow_data)
                        except json.JSONDecodeError:
                            workflow_data = {"nodes": [], "edges": []}
                    ts = time.strftime("%H:%M:%S")
                    return copy.deepcopy(workflow_data), \
                           f"💾 Auto-save: Saved ✓ ({ts})"

                # 1) 즉각 저장
                wb.change(
                    fn=periodic_save,
                    inputs=[wb],
                    outputs=[loaded_data, save_indicator]
                )

                # 2) 5초 주기 백업
                auto_timer = gr.Timer(10)
                auto_timer.tick(
                    fn=periodic_save,
                    inputs=[wb],
                    outputs=[loaded_data, save_indicator]
                )
                # -------------------------------

                return wb








        
        # ─── Import Section ───
        with gr.Accordion("📥 Import Workflow", open=True):
            gr.Markdown("*Load an existing workflow from JSON or start with a sample template*")
            with gr.Row():
                with gr.Column(scale=2):
                    import_json_text = gr.Code(
                        language="json",
                        label="Paste JSON here",
                        lines=8,
                        value='{\n  "nodes": [],\n  "edges": []\n}'
                    )
                with gr.Column(scale=1):
                    file_upload = gr.File(
                        label="Or upload JSON file",
                        file_types=[".json"],
                        type="filepath"
                    )
                    btn_load = gr.Button("📥 Load Workflow", variant="primary", size="lg")
                    
                    # Sample buttons
                    gr.Markdown("**Sample Workflows:**")
                    with gr.Row():
                        btn_sample_basic = gr.Button("🎯 Basic Q&A", variant="secondary", scale=1)
                        btn_sample_vidraft = gr.Button("🤖 VIDraft", variant="secondary", scale=1)
                    with gr.Row():
                        btn_sample_multi = gr.Button("📝 Multi-Input", variant="secondary", scale=1)
                        btn_sample_chain = gr.Button("🔗 Chain", variant="secondary", scale=1)
                    
                    # Status
                    status_text = gr.Textbox(
                        label="Status", 
                        value="Ready", 
                        elem_classes=["status-box"],
                        interactive=False
                    )
        
        # ─── Export Section ───
        gr.Markdown("## 💾 Export / Live Preview")
        gr.Markdown("*Your workflow is automatically saved. The JSON below shows your current workflow in real-time.*")
        
        # Workflow info display
        workflow_info = gr.Markdown("📊 Empty workflow", elem_classes=["workflow-info"])
        
        with gr.Row():
            with gr.Column(scale=3):
                export_preview = gr.Code(
                    language="json", 
                    label="Current Workflow JSON (Live Preview)", 
                    lines=8,
                    interactive=False
                )
                gr.Markdown("*💡 This JSON updates automatically as you modify the workflow above*")
            with gr.Column(scale=1):
                btn_preview = gr.Button("🔄 Force Refresh", size="lg", variant="secondary")
                btn_download = gr.DownloadButton(
                    "💾 Download JSON", 
                    size="lg",
                    variant="primary",
                    visible=True
                )
        
        # ─── Deploy Section ───
        with gr.Accordion("🚀 Deploy to Hugging Face Space", open=False, elem_classes=["deploy-section"]):
            gr.Markdown("""
            Deploy your **current workflow** as an independent Hugging Face Space app. 
            The workflow shown in the JSON preview above will be deployed exactly as is.
            """)
            
            gr.Markdown("*⚠️ Make sure to save/finalize your workflow design before deploying!*")
            
            with gr.Row():
                with gr.Column(scale=2):
                    deploy_name = gr.Textbox(
                        label="App Name",
                        placeholder="My Awesome Workflow App",
                        value="My Workflow App"
                    )
                    deploy_description = gr.Textbox(
                        label="App Description",
                        placeholder="Describe what your workflow does...",
                        lines=3,
                        value="A workflow application created with MOUSE Workflow builder."
                    )
                    deploy_space_name = gr.Textbox(
                        label="Space Name (your-username/space-name)",
                        placeholder="username/my-workflow-app",
                        info="This will be the URL of your Space"
                    )
                    
                with gr.Column(scale=1):
                    deploy_token = gr.Textbox(
                        label="Hugging Face Token",
                        type="password",
                        placeholder="hf_...",
                        info="Get your token from huggingface.co/settings/tokens"
                    )
                    
                    # API Keys 설정 섹션
                    gr.Markdown("### 🔑 API Keys Configuration")
                    
                    # FRIENDLI_TOKEN 설정
                    friendli_token_input = gr.Textbox(
                        label="FRIENDLI_TOKEN (VIDraft/Gemma)",
                        type="password",
                        placeholder="flp_...",
                        value=os.getenv("FRIENDLI_TOKEN", ""),
                        info="Required for VIDraft. Will be added as secret."
                    )
                    
                    # OpenAI API Key 설정
                    openai_token_input = gr.Textbox(
                        label="OPENAI_API_KEY (Optional)",
                        type="password",
                        placeholder="sk-...",
                        value=os.getenv("OPENAI_API_KEY", ""),
                        info="Optional. Leave empty if not using OpenAI."
                    )
                    
                    deploy_private = gr.Checkbox(
                        label="Make Space Private",
                        value=False
                    )
                    
                    btn_deploy = gr.Button("🚀 Deploy to HF Space", variant="primary", size="lg")
                    
                    # Deploy status
                    deploy_status = gr.Markdown("")
            
            # Preview generated code
            with gr.Accordion("📄 Preview Generated Code", open=False):
                generated_code_preview = gr.Code(
                    language="python",
                    label="app.py (This will be deployed)",
                    lines=20
                )
        
        # ─── UI Execution Section ───
        with gr.Column(elem_classes=["ui-execution-section"]):
            gr.Markdown("## 🚀 UI Execution")
            gr.Markdown("Test your workflow instantly! Click below to generate and run the UI from your current workflow design.")
            
            btn_execute_ui = gr.Button("▶️ Generate & Run UI from Current Workflow", variant="primary", size="lg")
            
            # UI execution state
            ui_workflow_data = gr.State(None)
            
            # Dynamic UI container
            @gr.render(inputs=[ui_workflow_data])
            def render_execution_ui(workflow_data):
                if not workflow_data or not workflow_data.get("nodes"):
                    gr.Markdown("*Load a workflow first, then click 'Generate & Run UI'*")
                    return
                
                gr.Markdown("### 📋 Generated UI")
                
                # Extract input and output nodes
                input_nodes = []
                output_nodes = []
                
                for node in workflow_data.get("nodes", []):
                    node_type = node.get("type", "")
                    if node_type in ["ChatInput", "textInput", "Input", "numberInput"]:
                        input_nodes.append(node)
                    elif node_type in ["ChatOutput", "textOutput", "Output"]:
                        output_nodes.append(node)
                    elif node_type == "textNode":
                        # textNode는 중간 처리 노드로, UI에는 표시하지 않음
                        pass
                
                # Create input components
                input_components = {}
                
                if input_nodes:
                    gr.Markdown("#### 📥 Inputs")
                    for node in input_nodes:
                        node_id = node.get("id")
                        label = node.get("data", {}).get("label", node_id)
                        node_type = node.get("type")
                        
                        # Get default value
                        template = node.get("data", {}).get("template", {})
                        default_value = template.get("input_value", {}).get("value", "")
                        
                        if node_type == "numberInput":
                            input_components[node_id] = gr.Number(
                                label=label,
                                value=float(default_value) if default_value else 0
                            )
                        else:
                            input_components[node_id] = gr.Textbox(
                                label=label,
                                value=default_value,
                                lines=2,
                                placeholder="Enter your input..."
                            )
                
                # Execute button
                execute_btn = gr.Button("🎯 Execute", variant="primary")
                
                # Create output components
                output_components = {}
                
                if output_nodes:
                    gr.Markdown("#### 📤 Outputs")
                    for node in output_nodes:
                        node_id = node.get("id")
                        label = node.get("data", {}).get("label", node_id)
                        
                        output_components[node_id] = gr.Textbox(
                            label=label,
                            interactive=False,
                            lines=3
                        )
                
                # Execution log
                gr.Markdown("#### 📊 Execution Log")
                log_output = gr.Textbox(
                    label="Log",
                    interactive=False,
                    lines=5
                )
                
                # Define execution handler
                def execute_ui_workflow(*input_values):
                    # Create input dictionary
                    inputs_dict = {}
                    input_keys = list(input_components.keys())
                    for i, key in enumerate(input_keys):
                        if i < len(input_values):
                            inputs_dict[key] = input_values[i]
                    
                    # Check API status
                    log = "=== Workflow Execution Started ===\n"
                    log += f"Inputs provided: {len(inputs_dict)}\n"
                    
                    # API 상태 확인
                    vidraft_token = os.getenv("FRIENDLI_TOKEN")
                    openai_key = os.getenv("OPENAI_API_KEY")
                    
                    log += "\nAPI Status:\n"
                    log += f"- FRIENDLI_TOKEN (VIDraft): {'✅ Found' if vidraft_token else '❌ Not found'}\n"
                    log += f"- OPENAI_API_KEY: {'✅ Found' if openai_key else '❌ Not found'}\n"
                    
                    if not vidraft_token and not openai_key:
                        log += "\n⚠️ No API keys found. Results will be simulated.\n"
                        log += "To get real AI responses, set API keys in environment variables.\n"
                        log += "Minimum requirement: FRIENDLI_TOKEN for VIDraft\n"
                    elif vidraft_token and not openai_key:
                        log += "\n✅ VIDraft API connected - Basic functionality available\n"
                        log += "💡 Add OPENAI_API_KEY for full functionality\n"
                    
                    log += "\n--- Processing Nodes ---\n"
                    
                    try:
                        results = execute_workflow_simple(workflow_data, inputs_dict)
                        
                        # Prepare outputs
                        output_values = []
                        for node_id in output_components.keys():
                            value = results.get(node_id, "No output")
                            output_values.append(value)
                            
                            # Log 길이 제한
                            display_value = value[:100] + "..." if len(str(value)) > 100 else value
                            log += f"\nOutput [{node_id}]: {display_value}\n"
                        
                        log += "\n=== Execution Completed Successfully! ===\n"
                        output_values.append(log)
                        
                        return output_values
                        
                    except Exception as e:
                        error_msg = f"❌ Error: {str(e)}"
                        log += f"\n{error_msg}\n"
                        log += "=== Execution Failed ===\n"
                        return [error_msg] * len(output_components) + [log]
                
                # Connect execution
                all_inputs = list(input_components.values())
                all_outputs = list(output_components.values()) + [log_output]
                
                execute_btn.click(
                    fn=execute_ui_workflow,
                    inputs=all_inputs,
                    outputs=all_outputs
                )
        
        # ─── Event Handlers ───
        
        # Load workflow (from text or file)
        def load_workflow(json_text, file_obj):
            data, status = load_json_from_text_or_file(json_text, file_obj)
            if data:
                # 로드 성공시 자동으로 미리보기 업데이트
                return data, status, json_text if not file_obj else export_pretty(data), "💾 Auto-save: Loaded ✓"
            else:
                return None, status, gr.update(), gr.update()
        
        btn_load.click(
            fn=load_workflow,
            inputs=[import_json_text, file_upload],
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        # Auto-load when file is uploaded
        file_upload.change(
            fn=load_workflow,
            inputs=[import_json_text, file_upload],
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        # Load samples
        btn_sample_basic.click(
            fn=lambda: (create_sample_workflow("basic"), "✅ Basic Q&A sample loaded", export_pretty(create_sample_workflow("basic")), "💾 Auto-save: Sample loaded ✓"),
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        btn_sample_vidraft.click(
            fn=lambda: (create_sample_workflow("vidraft"), "✅ VIDraft sample loaded", export_pretty(create_sample_workflow("vidraft")), "💾 Auto-save: Sample loaded ✓"),
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        btn_sample_multi.click(
            fn=lambda: (create_sample_workflow("multi_input"), "✅ Multi-input sample loaded", export_pretty(create_sample_workflow("multi_input")), "💾 Auto-save: Sample loaded ✓"),
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        btn_sample_chain.click(
            fn=lambda: (create_sample_workflow("chain"), "✅ Chain processing sample loaded", export_pretty(create_sample_workflow("chain")), "💾 Auto-save: Sample loaded ✓"),
            outputs=[loaded_data, status_text, import_json_text, save_indicator]
        ).then(
            fn=lambda current_trigger: not current_trigger,
            inputs=trigger_update,
            outputs=trigger_update
        )
        
        # Preview current workflow - 강제 새로고침
        def force_refresh_preview(current_data):
            """현재 워크플로우 데이터를 강제로 새로고침"""
            if current_data:
                node_count = len(current_data.get("nodes", []))
                edge_count = len(current_data.get("edges", []))
                info = f"📊 Workflow contains {node_count} nodes and {edge_count} edges"
                return export_pretty(current_data), "💾 Auto-save: Refreshed ✓", info
            return "No workflow data available", "💾 Auto-save: No data", "📊 Empty workflow"
        
        btn_preview.click(
            fn=force_refresh_preview,
            inputs=loaded_data,
            outputs=[export_preview, save_indicator, workflow_info]
        )
        
        # Download workflow는 이미 loaded_data.change에서 처리됨
        
        # Auto-update export preview when workflow changes
        def update_preview_and_download(data):
            """워크플로우 변경시 미리보기와 다운로드 업데이트"""
            if data:
                preview = export_pretty(data)
                download_file = export_file(data)
                node_count = len(data.get("nodes", []))
                edge_count = len(data.get("edges", []))
                status = f"📊 Workflow contains {node_count} nodes and {edge_count} edges"
                return preview, download_file, status
            return "No workflow data", None, "📊 Empty workflow"
        
        loaded_data.change(
            fn=update_preview_and_download,
            inputs=loaded_data,
            outputs=[export_preview, btn_download, workflow_info]
        )
        
        # Generate UI execution - 현재 워크플로우 사용
        def prepare_ui_execution(current_data):
            """현재 워크플로우를 UI 실행용으로 준비"""
            if not current_data or not current_data.get("nodes"):
                gr.Warning("Please create a workflow first!")
                return None
            return current_data
        
        btn_execute_ui.click(
            fn=prepare_ui_execution,
            inputs=loaded_data,
            outputs=ui_workflow_data
        )
        
        # ─── Deploy Event Handlers ───
        
        # Preview generated code
        def preview_generated_code(workflow_data, app_name, app_description):
            if not workflow_data:
                return "# No workflow loaded\n# Create or load a workflow first"
            
            if not workflow_data.get("nodes"):
                return "# Empty workflow\n# Add some nodes to see the generated code"
            
            try:
                code = generate_standalone_app(workflow_data, app_name, app_description)
                return code
            except Exception as e:
                return f"# Error generating code\n# {str(e)}"
        
        # Update preview when inputs change
        deploy_name.change(
            fn=preview_generated_code,
            inputs=[loaded_data, deploy_name, deploy_description],
            outputs=generated_code_preview
        )
        
        deploy_description.change(
            fn=preview_generated_code,
            inputs=[loaded_data, deploy_name, deploy_description],
            outputs=generated_code_preview
        )
        
        # Update preview when workflow changes too
        loaded_data.change(
            fn=preview_generated_code,
            inputs=[loaded_data, deploy_name, deploy_description],
            outputs=generated_code_preview
        )
        
        # Deploy handler
        def handle_deploy(workflow_data, app_name, app_description, hf_token, space_name, 
                         friendli_token, openai_token, is_private):
            if not workflow_data:
                return "❌ No workflow loaded. Please create or load a workflow first."
            
            if not workflow_data.get("nodes"):
                return "❌ Empty workflow. Please add some nodes to your workflow."
            
            if not hf_token:
                return "❌ Hugging Face token is required. Get yours at huggingface.co/settings/tokens"
            
            if not space_name:
                return "❌ Space name is required. Format: username/space-name"
            
            # Validate space name format
            if "/" not in space_name:
                return "❌ Invalid space name format. Use: username/space-name"
            
            # Check if huggingface-hub is available
            if not HF_HUB_AVAILABLE:
                return "❌ huggingface-hub library not installed. Install with: pip install huggingface-hub"
            
            # Show deploying status
            yield "🔄 Deploying to Hugging Face Space..."
            
            # Prepare API keys
            api_keys = {}
            
            # Always include FRIENDLI_TOKEN (even if empty)
            if not friendli_token:
                friendli_token = os.getenv("FRIENDLI_TOKEN", "")
            if friendli_token:
                api_keys["FRIENDLI_TOKEN"] = friendli_token
            
            # Include OpenAI key if provided
            if not openai_token:
                openai_token = os.getenv("OPENAI_API_KEY", "")
            if openai_token:
                api_keys["OPENAI_API_KEY"] = openai_token
            
            # Deploy
            result = deploy_to_huggingface(
                workflow_data=workflow_data,
                app_name=app_name,
                app_description=app_description,
                hf_token=hf_token,
                space_name=space_name,
                is_private=is_private,
                api_keys=api_keys
            )
            
            if result["success"]:
                # Build secrets status message
                secrets_msg = "\n\n**🔑 API Keys Status:**"
                
                if result.get("added_secrets"):
                    for secret in result["added_secrets"]:
                        secrets_msg += f"\n- {secret}: ✅ Successfully added"
                
                if result.get("failed_secrets"):
                    for failure in result["failed_secrets"]:
                        secrets_msg += f"\n- {failure}: ❌ Failed to add"
                
                # Check for missing required keys
                providers = result.get("providers_used", [])
                if "VIDraft" in providers and "FRIENDLI_TOKEN" not in result.get("added_secrets", []):
                    secrets_msg += "\n- FRIENDLI_TOKEN: ⚠️ Required for VIDraft but not provided"
                if "OpenAI" in providers and "OPENAI_API_KEY" not in result.get("added_secrets", []):
                    secrets_msg += "\n- OPENAI_API_KEY: ⚠️ Required for OpenAI but not provided"
                
                yield f"""✅ **Deployment Successful!**
                
🎉 Your workflow has been deployed to:
[{result['space_url']}]({result['space_url']})

⏱️ The Space will be ready in a few minutes. Building usually takes 2-5 minutes.

{secrets_msg}

📝 **Providers Detected in Workflow:**
{', '.join(result.get('providers_used', [])) if result.get('providers_used') else 'No LLM providers detected'}

🚀 **Default Configuration:**
The app is configured to prioritize VIDraft (Gemma-3-r1984-27B) for optimal performance.

📚 **Space Management:**
- To update secrets: Go to Space settings → Repository secrets
- To restart Space: Go to Space settings → Factory reboot
- To make changes: Edit files directly in the Space repository
"""
            else:
                yield f"❌ **Deployment Failed**\n\nError: {result['error']}"
        
        btn_deploy.click(
            fn=handle_deploy,
            inputs=[loaded_data, deploy_name, deploy_description, deploy_token, deploy_space_name, 
                   friendli_token_input, openai_token_input, deploy_private],
            outputs=deploy_status
        )
        

# -------------------------------------------------------------------
# 🚀 실행
# -------------------------------------------------------------------
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", show_error=True)