# main.py import os import re import random import string import uuid import json import logging import asyncio import time from collections import defaultdict from typing import List, Dict, Any, Optional, AsyncGenerator, Union, Tuple from datetime import datetime from aiohttp import ClientSession, ClientTimeout, ClientError from fastapi import FastAPI, HTTPException, Request, Depends, Header from fastapi.responses import StreamingResponse, JSONResponse, RedirectResponse from pydantic import BaseModel, validator from io import BytesIO import base64 from dotenv import load_dotenv # Load environment variables load_dotenv() # Configure logging logging.basicConfig( level=logging.INFO, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s", handlers=[logging.StreamHandler()] ) logger = logging.getLogger(__name__) # Load environment variables API_KEYS = os.getenv('API_KEYS', '').split(',') # Comma-separated API keys RATE_LIMIT = int(os.getenv('RATE_LIMIT', '60')) # Requests per minute AVAILABLE_MODELS = os.getenv('AVAILABLE_MODELS', '') # Comma-separated available models if not API_KEYS or API_KEYS == ['']: logger.error("No API keys found. Please set the API_KEYS environment variable.") raise Exception("API_KEYS environment variable not set.") # Process available models if AVAILABLE_MODELS: AVAILABLE_MODELS = [model.strip() for model in AVAILABLE_MODELS.split(',') if model.strip()] else: AVAILABLE_MODELS = [] # If empty, all models are available # Simple in-memory rate limiter based solely on IP addresses rate_limit_store = defaultdict(lambda: {"count": 0, "timestamp": time.time()}) # Define cleanup interval and window CLEANUP_INTERVAL = 60 # seconds RATE_LIMIT_WINDOW = 60 # seconds async def cleanup_rate_limit_stores(): """ Periodically cleans up stale entries in the rate_limit_store to prevent memory bloat. """ while True: current_time = time.time() ips_to_delete = [ip for ip, value in rate_limit_store.items() if current_time - value["timestamp"] > RATE_LIMIT_WINDOW * 2] for ip in ips_to_delete: del rate_limit_store[ip] logger.debug(f"Cleaned up rate_limit_store for IP: {ip}") await asyncio.sleep(CLEANUP_INTERVAL) async def rate_limiter_per_ip(request: Request): """ Rate limiter that enforces a limit based on the client's IP address. """ client_ip = request.client.host current_time = time.time() # Initialize or update the count and timestamp if current_time - rate_limit_store[client_ip]["timestamp"] > RATE_LIMIT_WINDOW: rate_limit_store[client_ip] = {"count": 1, "timestamp": current_time} else: if rate_limit_store[client_ip]["count"] >= RATE_LIMIT: logger.warning(f"Rate limit exceeded for IP address: {client_ip}") raise HTTPException(status_code=429, detail='Rate limit exceeded for IP address | NiansuhAI') rate_limit_store[client_ip]["count"] += 1 async def get_api_key(request: Request, authorization: str = Header(None)) -> str: """ Dependency to extract and validate the API key from the Authorization header. """ client_ip = request.client.host if authorization is None or not authorization.startswith('Bearer '): logger.warning(f"Invalid or missing authorization header from IP: {client_ip}") raise HTTPException(status_code=401, detail='Invalid authorization header format') api_key = authorization[7:] if api_key not in API_KEYS: logger.warning(f"Invalid API key attempted: {api_key} from IP: {client_ip}") raise HTTPException(status_code=401, detail='Invalid API key') return api_key # Custom exception for model not working class ModelNotWorkingException(Exception): def __init__(self, model: str): self.model = model self.message = f"The model '{model}' is currently not working. Please try another model or wait for it to be fixed." super().__init__(self.message) # Image Handling Functions ALLOWED_EXTENSIONS = {'png', 'jpg', 'jpeg', 'gif', 'webp', 'svg'} def is_allowed_extension(filename: str) -> bool: """ Checks if the given filename has an allowed extension. """ return '.' in filename and \ filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS def is_data_uri_an_image(data_uri: str) -> bool: """ Checks if the given data URI represents an image. """ match = re.match(r'data:image/(\w+);base64,', data_uri) if not match: raise ValueError("Invalid data URI image.") image_format = match.group(1).lower() if image_format not in ALLOWED_EXTENSIONS and image_format != "svg+xml": raise ValueError("Invalid image format (from MIME type).") return True def extract_data_uri(data_uri: str) -> bytes: """ Extracts the binary data from the given data URI. """ return base64.b64decode(data_uri.split(",")[1]) def to_data_uri(image: str) -> str: """ Validates and returns the data URI for an image. """ is_data_uri_an_image(image) return image class ImageResponseCustom: def __init__(self, url: str, alt: str): self.url = url self.alt = alt # Blackbox AI Integration (Placeholder for actual implementation) class Blackbox: url = "https://www.blackbox.ai" api_endpoint = "https://www.blackbox.ai/api/chat" # Placeholder endpoint working = True supports_stream = True supports_system_message = True supports_message_history = True default_model = 'blackboxai' image_models = ['ImageGeneration'] models = [ default_model, 'blackboxai-pro', *image_models, "llama-3.1-8b", 'llama-3.1-70b', 'llama-3.1-405b', 'gpt-4o', 'gemini-pro', 'gemini-1.5-flash', 'claude-sonnet-3.5', 'PythonAgent', 'JavaAgent', 'JavaScriptAgent', 'HTMLAgent', 'GoogleCloudAgent', 'AndroidDeveloper', 'SwiftDeveloper', 'Next.jsAgent', 'MongoDBAgent', 'PyTorchAgent', 'ReactAgent', 'XcodeAgent', 'AngularJSAgent', ] agentMode = { 'ImageGeneration': {'mode': True, 'id': "ImageGenerationLV45LJp", 'name': "Image Generation"}, 'Niansuh': {'mode': True, 'id': "NiansuhAIk1HgESy", 'name': "Niansuh"}, } trendingAgentMode = { "blackboxai": {}, "gemini-1.5-flash": {'mode': True, 'id': 'Gemini'}, "llama-3.1-8b": {'mode': True, 'id': "llama-3.1-8b"}, 'llama-3.1-70b': {'mode': True, 'id': "llama-3.1-70b"}, 'llama-3.1-405b': {'mode': True, 'id': "llama-3.1-405b"}, 'blackboxai-pro': {'mode': True, 'id': "BLACKBOXAI-PRO"}, 'PythonAgent': {'mode': True, 'id': "Python Agent"}, 'JavaAgent': {'mode': True, 'id': "Java Agent"}, 'JavaScriptAgent': {'mode': True, 'id': "JavaScript Agent"}, 'HTMLAgent': {'mode': True, 'id': "HTML Agent"}, 'GoogleCloudAgent': {'mode': True, 'id': "Google Cloud Agent"}, 'AndroidDeveloper': {'mode': True, 'id': "Android Developer"}, 'SwiftDeveloper': {'mode': True, 'id': "Swift Developer"}, 'Next.jsAgent': {'mode': True, 'id': "Next.js Agent"}, 'MongoDBAgent': {'mode': True, 'id': "MongoDB Agent"}, 'PyTorchAgent': {'mode': True, 'id': "PyTorch Agent"}, 'ReactAgent': {'mode': True, 'id': "React Agent"}, 'XcodeAgent': {'mode': True, 'id': "Xcode Agent"}, 'AngularJSAgent': {'mode': True, 'id': "AngularJS Agent"}, } userSelectedModel = { "gpt-4o": "gpt-4o", "gemini-pro": "gemini-pro", 'claude-sonnet-3.5': "claude-sonnet-3.5", } model_prefixes = { 'gpt-4o': '@GPT-4o', 'gemini-pro': '@Gemini-PRO', 'claude-sonnet-3.5': '@Claude-Sonnet-3.5', 'PythonAgent': '@Python Agent', 'JavaAgent': '@Java Agent', 'JavaScriptAgent': '@JavaScript Agent', 'HTMLAgent': '@HTML Agent', 'GoogleCloudAgent': '@Google Cloud Agent', 'AndroidDeveloper': '@Android Developer', 'SwiftDeveloper': '@Swift Developer', 'Next.jsAgent': '@Next.js Agent', 'MongoDBAgent': '@MongoDB Agent', 'PyTorchAgent': '@PyTorch Agent', 'ReactAgent': '@React Agent', 'XcodeAgent': '@Xcode Agent', 'AngularJSAgent': '@AngularJS Agent', 'blackboxai-pro': '@BLACKBOXAI-PRO', 'ImageGeneration': '@Image Generation', 'Niansuh': '@Niansuh', } model_referers = { "blackboxai": f"{url}/?model=blackboxai", "gpt-4o": f"{url}/?model=gpt-4o", "gemini-pro": f"{url}/?model=gemini-pro", "claude-sonnet-3.5": f"{url}/?model=claude-sonnet-3.5" } model_aliases = { "gemini-flash": "gemini-1.5-flash", "claude-3.5-sonnet": "claude-sonnet-3.5", "flux": "ImageGeneration", "niansuh": "Niansuh", } @classmethod def get_model(cls, model: str) -> Optional[str]: if model in cls.models: return model elif model in cls.userSelectedModel and cls.userSelectedModel[model] in cls.models: return cls.userSelectedModel[model] elif model in cls.model_aliases and cls.model_aliases[model] in cls.models: return cls.model_aliases[model] else: return cls.default_model if cls.default_model in cls.models else None @classmethod async def create_async_generator( cls, model: str, messages: List[Dict[str, Any]], proxy: Optional[str] = None, image: Optional[str] = None, image_name: Optional[str] = None, webSearchMode: bool = False, **kwargs ) -> AsyncGenerator[Union[str, ImageResponseCustom], None]: model = cls.get_model(model) if model is None: logger.error(f"Model {model} is not available.") raise ModelNotWorkingException(model) logger.info(f"Selected model: {model}") if not cls.working or model not in cls.models: logger.error(f"Model {model} is not working or not supported.") raise ModelNotWorkingException(model) headers = { "accept": "*/*", "accept-language": "en-US,en;q=0.9", "cache-control": "no-cache", "content-type": "application/json", "origin": cls.url, "pragma": "no-cache", "priority": "u=1, i", "referer": cls.model_referers.get(model, cls.url), "sec-ch-ua": '"Chromium";v="129", "Not=A?Brand";v="8"', "sec-ch-ua-mobile": "?0", "sec-ch-ua-platform": '"Linux"', "sec-fetch-dest": "empty", "sec-fetch-mode": "cors", "sec-fetch-site": "same-origin", "user-agent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/129.0.0.0 Safari/537.36", } if model in cls.model_prefixes: prefix = cls.model_prefixes[model] if not messages[0]['content'].startswith(prefix): logger.debug(f"Adding prefix '{prefix}' to the first message.") messages[0]['content'] = f"{prefix} {messages[0]['content']}" random_id = ''.join(random.choices(string.ascii_letters + string.digits, k=7)) messages[-1]['id'] = random_id messages[-1]['role'] = 'user' # Don't log the full message content for privacy logger.debug(f"Generated message ID: {random_id} for model: {model}") if image is not None: messages[-1]['data'] = { 'fileText': '', 'imageBase64': image, 'title': image_name } messages[-1]['content'] = 'FILE:BB\n$#$\n\n$#$\n' + messages[-1]['content'] logger.debug("Image data added to the message.") data = { "messages": messages, "id": random_id, "previewToken": None, "userId": None, "codeModelMode": True, "agentMode": {}, "trendingAgentMode": {}, "isMicMode": False, "userSystemPrompt": None, "maxTokens": 99999999, "playgroundTopP": 0.9, "playgroundTemperature": 0.5, "isChromeExt": False, "githubToken": None, "clickedAnswer2": False, "clickedAnswer3": False, "clickedForceWebSearch": False, "visitFromDelta": False, "mobileClient": False, "userSelectedModel": None, "webSearchMode": webSearchMode, } if model in cls.agentMode: data["agentMode"] = cls.agentMode[model] elif model in cls.trendingAgentMode: data["trendingAgentMode"] = cls.trendingAgentMode[model] elif model in cls.userSelectedModel: data["userSelectedModel"] = cls.userSelectedModel[model] logger.info(f"Sending request to {cls.api_endpoint} with data (excluding messages).") timeout = ClientTimeout(total=60) # Set an appropriate timeout retry_attempts = 10 # Set the number of retry attempts for attempt in range(retry_attempts): try: async with ClientSession(headers=headers, timeout=timeout) as session: async with session.post(cls.api_endpoint, json=data, proxy=proxy) as response: response.raise_for_status() logger.info(f"Received response with status {response.status}") if model in cls.image_models: response_text = await response.text() # Extract image URL from the response url_match = re.search(r'https://storage\.googleapis\.com/[^\s\)]+', response_text) if url_match: image_url = url_match.group(0) logger.info(f"Image URL found: {image_url}") yield ImageResponseCustom(url=image_url, alt=messages[-1]['content']) else: logger.error("Image URL not found in the response.") raise Exception("Image URL not found in the response") else: full_response = "" search_results_json = "" try: async for chunk, _ in response.content.iter_chunks(): if chunk: decoded_chunk = chunk.decode(errors='ignore') decoded_chunk = re.sub(r'\$@\$v=[^$]+\$@\$', '', decoded_chunk) if decoded_chunk.strip(): if '$~~~$' in decoded_chunk: search_results_json += decoded_chunk else: full_response += decoded_chunk yield decoded_chunk logger.info("Finished streaming response chunks.") except Exception as e: logger.exception("Error while iterating over response chunks.") raise e if data["webSearchMode"] and search_results_json: match = re.search(r'\$~~~\$(.*?)\$~~~\$', search_results_json, re.DOTALL) if match: try: search_results = json.loads(match.group(1)) formatted_results = "\n\n**Sources:**\n" for i, result in enumerate(search_results[:5], 1): formatted_results += f"{i}. [{result['title']}]({result['link']})\n" logger.info("Formatted search results.") yield formatted_results except json.JSONDecodeError as je: logger.error("Failed to parse search results JSON.") raise je break # Exit the retry loop if successful except ClientError as ce: logger.error(f"Client error occurred: {ce}. Retrying attempt {attempt + 1}/{retry_attempts}") if attempt == retry_attempts - 1: raise HTTPException(status_code=502, detail="Error communicating with the external API.") except asyncio.TimeoutError: logger.error(f"Request timed out. Retrying attempt {attempt + 1}/{retry_attempts}") if attempt == retry_attempts - 1: raise HTTPException(status_code=504, detail="External API request timed out.") except Exception as e: logger.error(f"Unexpected error: {e}. Retrying attempt {attempt + 1}/{retry_attempts}") if attempt == retry_attempts - 1: raise HTTPException(status_code=500, detail=str(e)) # FastAPI app setup app = FastAPI() # Add the cleanup task when the app starts @app.on_event("startup") async def startup_event(): asyncio.create_task(cleanup_rate_limit_stores()) logger.info("Started rate limit store cleanup task.") # Middleware to enhance security and enforce Content-Type for specific endpoints @app.middleware("http") async def security_middleware(request: Request, call_next): client_ip = request.client.host # Enforce that POST requests to /v1/chat/completions must have Content-Type: application/json if request.method == "POST" and request.url.path == "/v1/chat/completions": content_type = request.headers.get("Content-Type") if content_type != "application/json": logger.warning(f"Invalid Content-Type from IP: {client_ip} for path: {request.url.path}") return JSONResponse( status_code=400, content={ "error": { "message": "Content-Type must be application/json", "type": "invalid_request_error", "param": None, "code": None } }, ) response = await call_next(request) return response # Request Models class TextContent(BaseModel): type: str = "text" text: str @validator('type') def type_must_be_text(cls, v): if v != "text": raise ValueError("Type must be 'text'") return v class ImageContent(BaseModel): type: str = "image_url" image_url: Dict[str, str] @validator('type') def type_must_be_image_url(cls, v): if v != "image_url": raise ValueError("Type must be 'image_url'") return v ContentItem = Union[TextContent, ImageContent] class Message(BaseModel): role: str content: List[ContentItem] @validator('role') def role_must_be_valid(cls, v): if v not in {"system", "user", "assistant"}: raise ValueError("Role must be 'system', 'user', or 'assistant'") return v class ChatRequest(BaseModel): model: str messages: List[Message] temperature: Optional[float] = 1.0 top_p: Optional[float] = 1.0 n: Optional[int] = 1 stream: Optional[bool] = False stop: Optional[Union[str, List[str]]] = None max_tokens: Optional[int] = None presence_penalty: Optional[float] = 0.0 frequency_penalty: Optional[float] = 0.0 logit_bias: Optional[Dict[str, float]] = None user: Optional[str] = None webSearchMode: Optional[bool] = False # Custom parameter class TokenizerRequest(BaseModel): text: str def calculate_estimated_cost(prompt_tokens: int, completion_tokens: int) -> float: """ Calculate the estimated cost based on the number of tokens. Replace the pricing below with your actual pricing model. """ # Example pricing: $0.00000268 per token cost_per_token = 0.00000268 return round((prompt_tokens + completion_tokens) * cost_per_token, 8) def create_response(content: str, model: str, finish_reason: Optional[str] = None) -> Dict[str, Any]: return { "id": f"chatcmpl-{uuid.uuid4()}", "object": "chat.completion", "created": int(datetime.now().timestamp()), "model": model, "choices": [ { "index": 0, "message": { "role": "assistant", "content": content }, "finish_reason": finish_reason } ], "usage": None, # To be filled in non-streaming responses } def extract_all_images_from_content(content_items: List[ContentItem]) -> List[Tuple[str, str]]: """ Extracts all images from the content list. Returns a list of tuples containing (alt_text, image_data_uri). """ images = [] for item in content_items: if isinstance(item, ImageContent): alt_text = item.image_url.get('alt', '') # Optional alt text image_data_uri = item.image_url.get('url', '') if image_data_uri: images.append((alt_text, image_data_uri)) return images # Endpoint: POST /v1/chat/completions @app.post("/v1/chat/completions", dependencies=[Depends(rate_limiter_per_ip)]) async def chat_completions(request: ChatRequest, req: Request, api_key: str = Depends(get_api_key)): client_ip = req.client.host # Redact user messages only for logging purposes redacted_messages = [{"role": msg.role, "content": "[redacted]"} for msg in request.messages] logger.info(f"Received chat completions request from API key: {api_key} | IP: {client_ip} | Model: {request.model} | Messages: {redacted_messages}") try: # Validate that the requested model is available if request.model not in Blackbox.models and request.model not in Blackbox.model_aliases: logger.warning(f"Attempt to use unavailable model: {request.model} from IP: {client_ip}") raise HTTPException(status_code=400, detail="Requested model is not available.") # Initialize response content assistant_content = "" # Iterate through messages to find and process images for msg in request.messages: if msg.role == "user": # Extract all images from the message content images = extract_all_images_from_content(msg.content) for alt_text, image_data_uri in images: # Analyze the image analysis_result = await analyze_image(image_data_uri) assistant_content += analysis_result + "\n" # Example response content assistant_content += "Based on the image you provided, here are the insights..." # Calculate token usage (simple approximation) prompt_tokens = sum(len(" ".join([item.text if isinstance(item, TextContent) else item.image_url['url'] for item in msg.content]).split()) for msg in request.messages) completion_tokens = len(assistant_content.split()) total_tokens = prompt_tokens + completion_tokens estimated_cost = calculate_estimated_cost(prompt_tokens, completion_tokens) logger.info(f"Completed response generation for API key: {api_key} | IP: {client_ip}") if request.stream: async def generate(): try: for msg in request.messages: if msg.role == "user": images = extract_all_images_from_content(msg.content) for alt_text, image_data_uri in images: analysis_result = await analyze_image(image_data_uri) response_chunk = { "id": f"chatcmpl-{uuid.uuid4()}", "object": "chat.completion.chunk", "created": int(datetime.now().timestamp()), "model": request.model, "choices": [ { "index": 0, "delta": {"content": analysis_result + "\n", "role": "assistant"}, "finish_reason": None, } ], "usage": None, } yield f"data: {json.dumps(response_chunk)}\n\n" # Final message final_response = { "id": f"chatcmpl-{uuid.uuid4()}", "object": "chat.completion", "created": int(datetime.now().timestamp()), "model": request.model, "choices": [ { "message": { "role": "assistant", "content": assistant_content.strip() }, "finish_reason": "stop", "index": 0 } ], "usage": { "prompt_tokens": prompt_tokens, "completion_tokens": completion_tokens, "total_tokens": total_tokens, "estimated_cost": estimated_cost }, } yield f"data: {json.dumps(final_response)}\n\n" yield "data: [DONE]\n\n" except HTTPException as he: error_response = {"error": he.detail} yield f"data: {json.dumps(error_response)}\n\n" except Exception as e: logger.exception(f"Error during streaming response generation from IP: {client_ip}.") error_response = {"error": str(e)} yield f"data: {json.dumps(error_response)}\n\n" return StreamingResponse(generate(), media_type="text/event-stream") else: return { "id": f"chatcmpl-{uuid.uuid4()}", "object": "chat.completion", "created": int(datetime.now().timestamp()), "model": request.model, "choices": [ { "message": { "role": "assistant", "content": assistant_content.strip() }, "finish_reason": "stop", "index": 0 } ], "usage": { "prompt_tokens": prompt_tokens, "completion_tokens": completion_tokens, "total_tokens": total_tokens, "estimated_cost": estimated_cost }, } except ModelNotWorkingException as e: logger.warning(f"Model not working: {e} | IP: {client_ip}") raise HTTPException(status_code=503, detail=str(e)) except HTTPException as he: logger.warning(f"HTTPException: {he.detail} | IP: {client_ip}") raise he except Exception as e: logger.exception(f"An unexpected error occurred while processing the chat completions request from IP: {client_ip}.") raise HTTPException(status_code=500, detail=str(e)) # Endpoint: POST /v1/tokenizer @app.post("/v1/tokenizer", dependencies=[Depends(rate_limiter_per_ip)]) async def tokenizer(request: TokenizerRequest, req: Request): client_ip = req.client.host text = request.text token_count = len(text.split()) logger.info(f"Tokenizer requested from IP: {client_ip} | Text length: {len(text)}") return {"text": text, "tokens": token_count} # Endpoint: GET /v1/models @app.get("/v1/models", dependencies=[Depends(rate_limiter_per_ip)]) async def get_models(req: Request): client_ip = req.client.host logger.info(f"Fetching available models from IP: {client_ip}") return {"data": [{"id": model, "object": "model"} for model in Blackbox.models]} # Endpoint: GET /v1/models/{model}/status @app.get("/v1/models/{model}/status", dependencies=[Depends(rate_limiter_per_ip)]) async def model_status(model: str, req: Request): client_ip = req.client.host logger.info(f"Model status requested for '{model}' from IP: {client_ip}") if model in Blackbox.models: return {"model": model, "status": "available"} elif model in Blackbox.model_aliases and Blackbox.model_aliases[model] in Blackbox.models: actual_model = Blackbox.model_aliases[model] return {"model": actual_model, "status": "available via alias"} else: logger.warning(f"Model not found: {model} from IP: {client_ip}") raise HTTPException(status_code=404, detail="Model not found") # Endpoint: GET /v1/health @app.get("/v1/health", dependencies=[Depends(rate_limiter_per_ip)]) async def health_check(req: Request): client_ip = req.client.host logger.info(f"Health check requested from IP: {client_ip}") return {"status": "ok"} # Endpoint: GET /v1/chat/completions (GET method) @app.get("/v1/chat/completions") async def chat_completions_get(req: Request): client_ip = req.client.host logger.info(f"GET request made to /v1/chat/completions from IP: {client_ip}, redirecting to 'about:blank'") return RedirectResponse(url='about:blank') # Custom exception handler to match OpenAI's error format @app.exception_handler(HTTPException) async def http_exception_handler(request: Request, exc: HTTPException): client_ip = request.client.host logger.error(f"HTTPException: {exc.detail} | Path: {request.url.path} | IP: {client_ip}") return JSONResponse( status_code=exc.status_code, content={ "error": { "message": exc.detail, "type": "invalid_request_error", "param": None, "code": None } }, ) # Run the application if __name__ == "__main__": import uvicorn uvicorn.run("main:app", host="0.0.0.0", port=8000, reload=True)