app.py

import json
import os
import time
import uuid
import hashlib
import base64
from datetime import datetime
from concurrent.futures import ThreadPoolExecutor

import requests
from flask import Flask, request, jsonify, Response, stream_with_context
from flask_cors import CORS
from dotenv import load_dotenv

# 加载环境变量
load_dotenv()

# ==================== 配置管理类 ====================
class Config:
    """全局配置管理"""
    
    # 服务器配置
    PORT = int(os.getenv('PORT', 5200))
    MAX_WORKERS = int(os.getenv('MAX_WORKERS', 20))
    
    # 认证配置
    API_KEY = os.getenv('API_KEY', 'sk-123456')
    RAYCAST_TOKEN = os.getenv('RAYCAST_TOKEN', 'rca_9455afe4694f6d63194263810fa7e93a659e21b0eb2e384e18b26dec53c0ee21')
    
    # Raycast API 配置
    RAYCAST_BASE_URL = 'https://backend.raycast.com/api/v1'
    RAYCAST_CHAT_URL = f'{RAYCAST_BASE_URL}/ai/chat_completions'
    RAYCAST_FILES_URL = f'{RAYCAST_BASE_URL}/ai/files/'
    
    # Raycast 请求头配置
    RAYCAST_HEADERS = {
        'Content-Type': 'application/json',
        'accept-language': 'zh-CN,zh-Hans;q=0.9',
        'x-raycast-deviceid': 'c86ec3d4b2c9a66de6d1a19fc5bada76fc15af8f117dc1b69ba993391f0ad531',
        'accept-encoding': 'gzip, deflate, br',
        'user-agent': 'Raycast/1.0.4/747 (iOS Version 18.2.1 (Build 22C161))',
        'Cookie': '__raycast_session=4eb4e28abc9196e140b1980c79b75cdc'
    }
    
    # 系统偏好设置
    DEFAULT_SYSTEM_INSTRUCTIONS = f"""<user-preferences>

  The user has the following system preferences:

  - Locale: en-CN

  - Timezone: Asia/Shanghai

  - Current Date: {datetime.now().strftime('%Y-%m-%d')}

  - Unit Currency: ¥

  - Unit Temperature: °C

  - Unit Length: m

  - Unit Mass: kg

  - Decimal Separator: .

  - Grouping Separator: ,

  Use the system preferences to format your answers accordingly.

</user-preferences>"""
    
    @classmethod
    def get_raycast_headers(cls, include_auth=True):
        """获取Raycast请求头"""
        headers = cls.RAYCAST_HEADERS.copy()
        if include_auth:
            headers['authorization'] = f'Bearer {cls.RAYCAST_TOKEN}'
        return headers

# 配置Flask应用
app = Flask(__name__)
CORS(app)

# 创建线程池
executor = ThreadPoolExecutor(max_workers=Config.MAX_WORKERS)

# ==================== 认证装饰器 ====================
def require_auth(f):
    """认证装饰器"""
    def decorated_function(*args, **kwargs):
        auth_header = request.headers.get('Authorization')
        if not auth_header:
            return jsonify({
                'error': {
                    'message': '缺少认证头',
                    'type': 'authentication_error',
                    'code': 'missing_authorization'
                }
            }), 401
        
        # 检查Bearer token格式
        if not auth_header.startswith('Bearer '):
            return jsonify({
                'error': {
                    'message': '认证格式错误',
                    'type': 'authentication_error',
                    'code': 'invalid_authorization_format'
                }
            }), 401
        
        token = auth_header[7:]
        if token != Config.API_KEY:
            return jsonify({
                'error': {
                    'message': '认证失败',
                    'type': 'authentication_error',
                    'code': 'invalid_api_key'
                }
            }), 401
        
        return f(*args, **kwargs)
    decorated_function.__name__ = f.__name__
    return decorated_function

# ==================== 工具类 ====================
class UtilsHelper:
    @staticmethod
    def generate_uuid():
        return str(uuid.uuid4())

    @staticmethod
    def get_current_timestamp():
        return int(time.time())

    @staticmethod
    def generate_md5(data):
        if isinstance(data, str):
            data = data.encode('utf-8')
        return base64.b64encode(hashlib.md5(data).digest()).decode('utf-8')

    @staticmethod
    def is_search_model(model):
        return model.endswith('-search')

    @staticmethod
    def get_base_model(model):
        return model[:-7] if model.endswith('-search') else model

# ==================== 模型映射类 ====================
class ModelMapper:
    BASE_MODELS = {
        "ray1": "raycast",
        "ray1-mini": "raycast",
        "gpt-4.1": "openai",
        "gpt-4.1-mini": "openai",
        "gpt-4.1-nano": "openai",
        "gpt-4": "openai",
        "gpt-4-turbo": "openai",
        "gpt-4o": "openai",
        "gpt-4o-mini": "openai",
        "o3": "openai_o1",
        "o4-mini": "openai_o1",
        "o1-mini": "openai_o1",
        "o1-2024-12-17": "openai_o1",
        "o3-mini": "openai_o1",
        "claude-3-5-haiku-latest": "anthropic",
        "claude-3-5-sonnet-latest": "anthropic",
        "claude-3-7-sonnet-latest": "anthropic",
        "claude-3-7-sonnet-latest-reasoning": "anthropic",
        "claude-3-opus-20240229": "anthropic",
        "claude-sonnet-4-20250514": "anthropic",
        "claude-opus-4-20250514": "anthropic",
        "claude-sonnet-4-20250514-reasoning": "anthropic",
        "claude-opus-4-20250514-reasoning": "anthropic",
        "sonar": "perplexity",
        "sonar-pro": "perplexity",
        "sonar-reasoning": "perplexity",
        "sonar-reasoning-pro": "perplexity",
        "meta-llama/llama-4-scout-17b-16e-instruct": "groq",
        "llama-3.3-70b-versatile": "groq",
        "llama-3.1-8b-instant": "groq",
        "llama3-70b-8192": "groq",
        "meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo": "together",
        "open-mistral-nemo": "mistral",
        "mistral-large-latest": "mistral",
        "mistral-medium-latest": "mistral",
        "mistral-small-latest": "mistral",
        "codestral-latest": "mistral",
        "deepseek-r1-distill-llama-70b": "groq",
        "gemini-2.5-pro-preview-06-05": "google",
        "gemini-1.5-flash": "google",
        "gemini-2.5-flash-preview-04-17": "google",
        "gemini-2.0-flash": "google",
        "gemini-2.0-flash-thinking-exp-01-21": "google",
        "deepseek-ai/DeepSeek-R1": "together",
        "deepseek-ai/DeepSeek-V3": "together",
        "grok-3-fast-beta": "xai",
        "grok-3-mini-fast-beta": "xai",
        "grok-2-latest": "xai"
    }

    # 生成完整的模型映射表（包含搜索版本）
    @classmethod
    def get_model_map(cls):
        model_map = cls.BASE_MODELS.copy()
        # 为每个基础模型添加带搜索功能的版本
        for model in cls.BASE_MODELS.keys():
            model_map[f"{model}-search"] = cls.BASE_MODELS[model]
        return model_map

    @classmethod
    def get_provider(cls, model):
        base_model = UtilsHelper.get_base_model(model)
        return cls.get_model_map().get(base_model, 'google')

    @classmethod
    def get_actual_model(cls, model):
        base_model = UtilsHelper.get_base_model(model)
        provider = cls.get_provider(model)
        
        if provider == 'raycast':
            return 'gemini-2.5-flash-preview-04-17'
        else:
            return base_model

    @classmethod
    def get_all_models(cls):
        return list(cls.get_model_map().keys())

# ==================== 工具功能类 ====================
class ToolsManager:
    @staticmethod
    def get_tools(use_search=False):
        if not use_search:
            return []
        
        return [
            {
                "name": "search_images",
                "type": "remote_tool"
            },
            {
                "name": "web_search",
                "type": "remote_tool"
            }
        ]

# ==================== 文件上传类 ====================
class FileUploader:
    @classmethod
    def upload_file(cls, file_data):
        try:
            filename = file_data['filename']
            content = file_data['content']
            content_type = file_data['contentType']
            
            buffer = base64.b64decode(content)
            byte_size = len(buffer)
            checksum = UtilsHelper.generate_md5(buffer)

            # 创建文件元数据
            create_file_payload = {
                'blob': {
                    'byte_size': byte_size,
                    'checksum': checksum,
                    'content_type': content_type,
                    'filename': filename
                }
            }

            headers = Config.get_raycast_headers()
            headers['x-raycast-timestamp'] = str(UtilsHelper.get_current_timestamp())
            headers['x-request-id'] = UtilsHelper.generate_uuid().upper()

            create_response = requests.post(
                Config.RAYCAST_FILES_URL,
                headers=headers,
                json=create_file_payload,
                timeout=30
            )

            if not create_response.ok:
                raise Exception(f'文件元数据创建失败: {create_response.status_code}')

            create_result = create_response.json()
            upload_url = create_result['direct_upload']['url']
            file_id = create_result['id']

            # 上传文件
            upload_headers = {
                'Content-Type': content_type,
                'Content-MD5': checksum,
                'Content-Length': str(byte_size),
                'Content-Disposition': f'inline; filename="{filename}"; filename*=UTF-8\'\'{filename}',
                'Upload-Complete': '?1'
            }

            upload_response = requests.put(
                upload_url,
                headers=upload_headers,
                data=buffer,
                timeout=60
            )

            if not upload_response.ok:
                raise Exception(f'文件上传失败: {upload_response.status_code}')

            return {
                'id': file_id,
                'type': 'file'
            }

        except Exception as error:
            print(f'文件上传错误: {error}')
            raise error

    @classmethod
    def extract_files_from_openai(cls, messages):
        files = []
        
        for message in messages:
            if message.get('role') == 'user' and isinstance(message.get('content'), list):
                for content in message['content']:
                    if content.get('type') == 'image_url' and content.get('image_url'):
                        url = content['image_url']['url']
                        if url.startswith('data:'):
                            # 处理base64图片
                            header, data = url.split(',', 1)
                            mime_match = header.split(':')[1].split(';')[0] if ':' in header else 'image/jpeg'
                            content_type = mime_match
                            
                            files.append({
                                'filename': f'image_{UtilsHelper.generate_uuid()}.{content_type.split("/")[1]}',
                                'content': data,
                                'contentType': content_type
                            })
        
        return files

# ==================== 转换类 ====================
class MessageConverter:
    @classmethod
    def merge_consecutive_messages(cls, messages):
        """合并连续相同角色的消息"""
        if not messages:
            return messages
        
        merged_messages = []
        current_message = None
        
        for message in messages:
            role = message.get('role')
            content = message.get('content', '')
            
            # 处理content为list的情况
            if isinstance(content, list):
                content = ''.join([
                    c.get('text', '') for c in content 
                    if c.get('type') == 'text'
                ])
            
            if current_message is None:
                # 第一条消息
                current_message = {
                    'role': role,
                    'content': content
                }
            elif current_message['role'] == role:
                # 相同角色，合并内容
                current_message['content'] += '\n' + content
            else:
                # 不同角色，保存当前消息并开始新消息
                merged_messages.append(current_message)
                current_message = {
                    'role': role,
                    'content': content
                }
        
        # 添加最后一条消息
        if current_message:
            merged_messages.append(current_message)
        
        return merged_messages

    @classmethod
    def process_system_messages(cls, messages):
        # 先合并连续相同角色的消息
        merged_messages = cls.merge_consecutive_messages(messages)
        
        processed_messages = []
        additional_system_instructions = ''
        system_collection_stopped = False
        
        for message in merged_messages:
            if message.get('role') == 'system':
                if not system_collection_stopped:
                    # 连续的 system 消息收集到 additional_system_instructions
                    if additional_system_instructions:
                        additional_system_instructions += '\n' + message['content']
                    else:
                        additional_system_instructions = message['content']
                else:
                    # 后续的 system 消息转换为 user 消息
                    processed_messages.append({
                        'author': 'user',
                        'content': {
                            'references': [],
                            'text': message['content']
                        }
                    })
            else:
                # 遇到非 system 消息，停止收集 system 消息
                system_collection_stopped = True
                
                processed_message = {
                    'author': 'user' if message.get('role') == 'user' else 'assistant',
                    'content': {
                        'references': [],
                        'text': message['content']
                    }
                }
                
                processed_messages.append(processed_message)
        
        return processed_messages, additional_system_instructions

    @classmethod
    def convert_to_raycast_format(cls, openai_request):
        processed_messages, additional_system_instructions = cls.process_system_messages(
            openai_request['messages']
        )
        
        # 处理文件上传
        files = FileUploader.extract_files_from_openai(openai_request['messages'])
        attachments = []
        
        for file in files:
            try:
                uploaded_file = FileUploader.upload_file(file)
                attachments.append(uploaded_file)
            except Exception as error:
                print(f'文件上传失败: {error}')

        # 如果有附件，添加到最后一个用户消息中
        if attachments and processed_messages:
            last_message = processed_messages[-1]
            if last_message['author'] == 'user':
                last_message['content']['attachments'] = attachments
        
        actual_model = ModelMapper.get_actual_model(openai_request['model'])
        provider = ModelMapper.get_provider(openai_request['model'])
        use_search = UtilsHelper.is_search_model(openai_request['model'])
        
        raycast_request = {
            'additional_system_instructions': additional_system_instructions or Config.DEFAULT_SYSTEM_INSTRUCTIONS,
            'debug': False,
            'locale': 'en_CN',
            'message_id': UtilsHelper.generate_uuid(),
            'messages': processed_messages,
            'model': actual_model,
            'provider': 'google' if provider == 'raycast' else provider,
            'source': 'ai_chat',
            'tools': ToolsManager.get_tools(use_search)
        }
        
        return raycast_request

# ==================== 响应处理类 ====================
class ResponseProcessor:
    def __init__(self):
        self.is_thinking = False
        self.thinking_content = ''

    def process_raycast_chunk(self, chunk):
        content = ''
        
        # 处理思考内容
        if chunk.get('reasoning'):
            if not self.is_thinking:
                # 开始思考
                self.is_thinking = True
                content += '<think>'
            content += chunk['reasoning']
            self.thinking_content += chunk['reasoning']
        
        # 处理普通文本内容
        if chunk.get('text'):
            if self.is_thinking:
                # 结束思考
                content += '</think>'
                self.is_thinking = False
                self.thinking_content = ''
            content += chunk['text']
        
        return content
    
    def convert_to_openai_format(self, raycast_chunk, model, is_stream=False):
        content = self.process_raycast_chunk(raycast_chunk)
        
        if is_stream:
            return {
                'id': 'chatcmpl-' + UtilsHelper.generate_uuid(),
                'object': 'chat.completion.chunk',
                'created': UtilsHelper.get_current_timestamp(),
                'model': model,
                'choices': [{
                    'index': 0,
                    'delta': {
                        'content': content
                    },
                    'finish_reason': None
                }]
            }
        else:
            return {
                'id': 'chatcmpl-' + UtilsHelper.generate_uuid(),
                'object': 'chat.completion',
                'created': UtilsHelper.get_current_timestamp(),
                'model': model,
                'choices': [{
                    'index': 0,
                    'message': {
                        'role': 'assistant',
                        'content': content
                    },
                    'finish_reason': 'stop'
                }],
                'usage': {
                    'prompt_tokens': 0,
                    'completion_tokens': 0,
                    'total_tokens': 0
                }
            }
    
    def finish_thinking(self):
        if self.is_thinking:
            self.is_thinking = False
            return '</think>'
        return ''

# ==================== API服务类 ====================
class RaycastAPIService:
    @classmethod
    def send_request(cls, raycast_request):
        headers = Config.get_raycast_headers()
        headers['x-raycast-timestamp'] = str(UtilsHelper.get_current_timestamp())
        
        print(f'发送到 Raycast: {json.dumps(raycast_request, indent=2, ensure_ascii=False)}')
        
        response = requests.post(
            Config.RAYCAST_CHAT_URL,
            headers=headers,
            json=raycast_request,
            stream=True,
            timeout=120
        )
        
        if not response.ok:
            error_text = response.text
            print(f'Raycast API 错误响应: {error_text}')
            raise Exception(f'Raycast API 响应错误: {response.status_code} {response.reason}')
        
        return response

# ==================== 处理函数 ====================
def handle_chat_completion(request_data):
    try:
        print(f'收到请求: {json.dumps(request_data, indent=2, ensure_ascii=False)}')
        
        # 转换请求格式
        raycast_request = MessageConverter.convert_to_raycast_format(request_data)
        
        # 发送请求到 Raycast
        response = RaycastAPIService.send_request(raycast_request)
        
        return response, request_data
        
    except Exception as error:
        print(f'代理错误: {error}')
        raise error

def process_stream_response(response, request_data):
    processor = ResponseProcessor()
    
    def generate():
        try:
            buffer = ''
            for chunk in response.iter_lines():
                chunk = chunk.decode("utf-8").strip()
                if chunk:
                    buffer += chunk + '\n'
                    lines = buffer.split('\n')
                    buffer = lines.pop() if lines else ''  
                    
                    for line in lines:
                        if line.strip():
                            try:
                                if line.startswith('data: '):
                                    data = line[6:]  
                                    if data == '[DONE]':
                                        # 检查是否需要关闭thinking标签
                                        finish_content = processor.finish_thinking()
                                        if finish_content:
                                            finish_response = processor.convert_to_openai_format(
                                                {'text': finish_content}, request_data['model'], True
                                            )
                                            yield f"data: {json.dumps(finish_response)}\n\n"
                                        yield 'data: [DONE]\n\n'
                                        return
                                    
                                    parsed = json.loads(data)
                                    openai_response = processor.convert_to_openai_format(
                                        parsed, request_data['model'], True
                                    )
                                    yield f"data: {json.dumps(openai_response)}\n\n"
                            except Exception as err:
                                print(f'解析流式响应错误: {err}, 原始行: {line}')
            
            yield 'data: [DONE]\n\n'
                    
        except Exception as err:
            print(f'流式响应错误: {err}')
            yield f'data: {json.dumps({"error": "流式响应处理错误"})}\n\n'
        finally:
            response.close()
    
    return generate()

def process_non_stream_response(response, request_data):
    processor = ResponseProcessor()
    full_content = ''
    
    try:
        buffer = ''
        for chunk in response.iter_lines():
            chunk = chunk.decode("utf-8").strip()
            if chunk:
                buffer += chunk + '\n'
                lines = buffer.split('\n')
                buffer = lines.pop() if lines else '' 
                
                for line in lines:
                    if line.strip():
                        try:
                            if line.startswith('data: '):
                                data = line[6:] 
                                if data == '[DONE]':
                                    break  # 结束处理
                                
                                parsed = json.loads(data)
                                content = processor.process_raycast_chunk(parsed)
                                full_content += content
                        except Exception as err:
                            print(f'解析非流式响应错误: {err}, 原始行: {line}')
        
        # 确保thinking标签正确关闭
        finish_content = processor.finish_thinking()
        full_content += finish_content
        
        return {
            'id': 'chatcmpl-' + UtilsHelper.generate_uuid(),
            'object': 'chat.completion',
            'created': UtilsHelper.get_current_timestamp(),
            'model': request_data['model'],
            'choices': [{
                'index': 0,
                'message': {
                    'role': 'assistant',
                    'content': full_content
                },
                'finish_reason': 'stop'
            }],
            'usage': {
                'prompt_tokens': 0,
                'completion_tokens': 0,
                'total_tokens': 0
            }
        }
        
    except Exception as err:
        print(f'非流式响应错误: {err}')
        raise err
    finally:
        response.close()

# ==================== 路由处理 ====================

@app.route('/v1/chat/completions', methods=['POST'])
@require_auth
def chat_completions():
    try:
        request_data = request.get_json()
        if not request_data:
            return jsonify({
                'error': {
                    'message': '请求数据为空',
                    'type': 'invalid_request',
                    'code': 'invalid_request'
                }
            }), 400
        
        is_stream = request_data.get('stream', False)
        
        # 在线程池中处理请求
        future = executor.submit(handle_chat_completion, request_data)
        response, req_data = future.result()
        
        if is_stream:
            return Response(
                stream_with_context(process_stream_response(response, req_data)),
                content_type='text/event-stream',
                headers={
                    'Cache-Control': 'no-cache',
                    'Connection': 'keep-alive',
                    'Access-Control-Allow-Origin': '*'
                }
            )
        else:
            future = executor.submit(process_non_stream_response, response, req_data)
            result = future.result()
            return jsonify(result)
            
    except Exception as error:
        return jsonify({
            'error': {
                'message': str(error) or '内部服务器错误',
                'type': 'internal_error',
                'code': 'internal_error'
            }
        }), 500

@app.route('/v1/models', methods=['GET'])
def list_models():
    models = [
        {
            'id': model,
            'object': 'model',
            'created': UtilsHelper.get_current_timestamp(),
            'owned_by': 'raycast-proxy'
        }
        for model in ModelMapper.get_all_models()
    ]
    
    return jsonify({
        'object': 'list',
        'data': models
    })

@app.route('/health', methods=['GET'])
def health_check():
    return jsonify({
        'status': 'ok',
        'timestamp': datetime.now().isoformat(),
        'models_count': len(ModelMapper.get_all_models()),
        'config': {
            'port': Config.PORT,
            'max_workers': Config.MAX_WORKERS,
            'auth_required': bool(Config.API_KEY)
        }
    })

@app.route('/', methods=['OPTIONS'])
@app.route('/v1/chat/completions', methods=['OPTIONS'])
@app.route('/v1/models', methods=['OPTIONS'])
def handle_options():
    return '', 200

if __name__ == '__main__':
    print(f'🚀 Raycast 代理服务器运行在端口 {Config.PORT}')
    print(f'🔗 OpenAI 兼容端点: http://localhost:{Config.PORT}/v1/chat/completions')
    print(f'📜 模型列表: http://localhost:{Config.PORT}/v1/models')
    print(f'⚡ 最大工作线程数: {Config.MAX_WORKERS}')
    
    # 使用支持多线程的WSGI服务器
    app.run(
        host='0.0.0.0',
        port=Config.PORT,
        debug=False,
        threaded=True,
        processes=1
    )