main.py

import base64
import json
import os
import secrets
import string
import time
import tempfile
from typing import List, Optional, Union, Any

import httpx
from dotenv import load_dotenv
from fastapi import FastAPI, HTTPException
from fastapi.responses import JSONResponse, StreamingResponse
from pydantic import BaseModel, Field, model_validator

# New import for OCR
from gradio_client import Client, handle_file

# --- Configuration ---
load_dotenv()

# Env variables for external services
IMAGE_API_URL = os.environ.get("IMAGE_API_URL", "https://image.api.example.com")
SNAPZION_UPLOAD_URL = "https://upload.snapzion.com/api/public-upload"
SNAPZION_API_KEY = os.environ.get("SNAP", "")
CHAT_API_URL = "https://www.chatwithmono.xyz/api/chat"
IMAGE_GEN_API_URL = "https://www.chatwithmono.xyz/api/image"
MODERATION_API_URL = "https://www.chatwithmono.xyz/api/moderation"

# --- Model Definitions ---
# Added florence-2-ocr for the new endpoint
AVAILABLE_MODELS = [
    {"id": "gpt-4-turbo", "object": "model", "created": int(time.time()), "owned_by": "system"},
    {"id": "gpt-4o", "object": "model", "created": int(time.time()), "owned_by": "system"},
    {"id": "gpt-3.5-turbo", "object": "model", "created": int(time.time()), "owned_by": "system"},
    {"id": "dall-e-3", "object": "model", "created": int(time.time()), "owned_by": "system"},
    {"id": "text-moderation-stable", "object": "model", "created": int(time.time()), "owned_by": "system"},
    {"id": "florence-2-ocr", "object": "model", "created": int(time.time()), "owned_by": "system"},
]
MODEL_ALIASES = {}

# --- FastAPI Application & Global Clients ---
app = FastAPI(
    title="OpenAI Compatible API",
    description="An adapter for various services to be compatible with the OpenAI API specification.",
    version="1.1.0"
)

# Initialize Gradio client for OCR globally to avoid re-initialization on each request
try:
    ocr_client = Client("multimodalart/Florence-2-l4")
except Exception as e:
    print(f"Warning: Could not initialize Gradio client for OCR: {e}")
    ocr_client = None

# --- Pydantic Models ---

# /v1/chat/completions
class Message(BaseModel):
    role: str
    content: str

class ChatRequest(BaseModel):
    messages: List[Message]
    model: str
    stream: Optional[bool] = False
    tools: Optional[Any] = None

# /v1/images/generations
class ImageGenerationRequest(BaseModel):
    prompt: str
    aspect_ratio: Optional[str] = "1:1"
    n: Optional[int] = 1
    user: Optional[str] = None
    model: Optional[str] = "default"

# /v1/moderations
class ModerationRequest(BaseModel):
    input: Union[str, List[str]]
    model: Optional[str] = "text-moderation-stable"

# /v1/ocr
class OcrRequest(BaseModel):
    image_url: Optional[str] = Field(None, description="URL of the image to process.")
    image_b64: Optional[str] = Field(None, description="Base64 encoded string of the image to process.")

    @model_validator(mode='before')
    @classmethod
    def check_sources(cls, data: Any) -> Any:
        if isinstance(data, dict):
            url = data.get('image_url')
            b64 = data.get('image_b64')
            if not (url or b64):
                raise ValueError('Either image_url or image_b64 must be provided.')
            if url and b64:
                raise ValueError('Provide either image_url or image_b64, not both.')
        return data

class OcrResponse(BaseModel):
    ocr_text: str
    raw_response: dict


# --- Helper Function for Random ID Generation ---
def generate_random_id(prefix: str, length: int = 29) -> str:
    """Generates a cryptographically secure, random alphanumeric ID."""
    population = string.ascii_letters + string.digits
    random_part = "".join(secrets.choice(population) for _ in range(length))
    return f"{prefix}{random_part}"

# === API Endpoints ===

@app.get("/v1/models", tags=["Models"])
async def list_models():
    """Lists the available models."""
    return {"object": "list", "data": AVAILABLE_MODELS}

@app.post("/v1/chat/completions", tags=["Chat"])
async def chat_completion(request: ChatRequest):
    """Handles chat completion requests, supporting streaming and non-streaming."""
    model_id = MODEL_ALIASES.get(request.model, request.model)
    chat_id = generate_random_id("chatcmpl-")
    headers = {
        'accept': 'text/event-stream',
        'content-type': 'application/json',
        'origin': 'https://www.chatwithmono.xyz',
        'referer': 'https://www.chatwithmono.xyz/',
        'user-agent': 'Mozilla/5.0',
    }
    
    # Handle tool prompting
    if request.tools:
        tool_prompt = f"""You have access to the following tools. To call a tool, please respond with JSON for a tool call within <tool_call></tool_call> XML tags. Respond in the format {{"name": tool name, "parameters": dictionary of argument name and its value}}. Do not use variables.
Tools: {";".join(f"<tool>{tool}</tool>" for tool in request.tools)}
Response Format for tool call:
<tool_call>
{{"name": <function-name>, "arguments": <args-json-object>}}
</tool_call>"""
        if request.messages[0].role == "system":
            request.messages[0].content += "\n\n" + tool_prompt
        else:
            request.messages.insert(0, Message(role="system", content=tool_prompt))
    
    payload = {"messages": [msg.model_dump() for msg in request.messages], "model": model_id}

    if request.stream:
        async def event_stream():
            created = int(time.time())
            usage_info = None
            is_first_chunk = True
            tool_call_buffer = ""
            in_tool_call = False

            try:
                async with httpx.AsyncClient(timeout=120) as client:
                    async with client.stream("POST", CHAT_API_URL, headers=headers, json=payload) as response:
                        response.raise_for_status()
                        async for line in response.aiter_lines():
                            if not line: continue
                            if line.startswith("0:"):
                                try:
                                    content_piece = json.loads(line[2:])
                                except json.JSONDecodeError:
                                    continue
                                
                                current_buffer = content_piece
                                if in_tool_call:
                                    current_buffer = tool_call_buffer + content_piece

                                if "</tool_call>" in current_buffer:
                                    tool_str = current_buffer.split("<tool_call>")[1].split("</tool_call>")[0]
                                    tool_json = json.loads(tool_str.strip())
                                    delta = {
                                        "content": None,
                                        "tool_calls": [{"index": 0, "id": generate_random_id("call_"), "type": "function",
                                                        "function": {"name": tool_json["name"], "arguments": json.dumps(tool_json["parameters"])}}]
                                    }
                                    chunk = {"id": chat_id, "object": "chat.completion.chunk", "created": created, "model": model_id,
                                             "choices": [{"index": 0, "delta": delta, "finish_reason": None}], "usage": None}
                                    yield f"data: {json.dumps(chunk)}\n\n"
                                    
                                    in_tool_call = False
                                    tool_call_buffer = ""
                                    # Process text that might come after the tool call in the same chunk
                                    remaining_text = current_buffer.split("</tool_call>", 1)[1]
                                    if remaining_text:
                                        content_piece = remaining_text
                                    else:
                                        continue

                                if "<tool_call>" in content_piece:
                                    in_tool_call = True
                                    tool_call_buffer += content_piece.split("<tool_call>", 1)[1]
                                    # Process text that came before the tool call
                                    text_before = content_piece.split("<tool_call>", 1)[0]
                                    if text_before:
                                        # Send the text before the tool call starts
                                        delta = {"content": text_before, "tool_calls": None}
                                        chunk = {"id": chat_id, "object": "chat.completion.chunk", "created": created, "model": model_id,
                                                 "choices": [{"index": 0, "delta": delta, "finish_reason": None}], "usage": None}
                                        yield f"data: {json.dumps(chunk)}\n\n"
                                    if "</tool_call>" not in tool_call_buffer:
                                        continue # Wait for the closing tag
                                
                                if not in_tool_call:
                                    delta = {"content": content_piece}
                                    if is_first_chunk:
                                        delta["role"] = "assistant"
                                        is_first_chunk = False
                                    chunk = {"id": chat_id, "object": "chat.completion.chunk", "created": created, "model": model_id,
                                             "choices": [{"index": 0, "delta": delta, "finish_reason": None}], "usage": None}
                                    yield f"data: {json.dumps(chunk)}\n\n"

                            elif line.startswith(("e:", "d:")):
                                try:
                                    usage_info = json.loads(line[2:]).get("usage")
                                except (json.JSONDecodeError, AttributeError): pass
                                break
                        
                        # Finalize
                        final_usage = None
                        if usage_info:
                            final_usage = {"prompt_tokens": usage_info.get("promptTokens", 0), "completion_tokens": usage_info.get("completionTokens", 0), "total_tokens": usage_info.get("promptTokens", 0) + usage_info.get("completionTokens", 0)}
                        done_chunk = {"id": chat_id, "object": "chat.completion.chunk", "created": created, "model": model_id,
                                      "choices": [{"index": 0, "delta": {}, "finish_reason": "stop" if not in_tool_call else "tool_calls"}], "usage": final_usage}
                        yield f"data: {json.dumps(done_chunk)}\n\n"

            except httpx.HTTPStatusError as e:
                error_content = {"error": {"message": f"Upstream API error: {e.response.status_code}. Details: {e.response.text}", "type": "upstream_error", "code": str(e.response.status_code)}}
                yield f"data: {json.dumps(error_content)}\n\n"
            finally:
                yield "data: [DONE]\n\n"

        return StreamingResponse(event_stream(), media_type="text/event-stream")
    else: # Non-streaming
        full_response, usage_info = "", {}
        try:
            async with httpx.AsyncClient(timeout=120) as client:
                async with client.stream("POST", CHAT_API_URL, headers=headers, json=payload) as response:
                    response.raise_for_status()
                    async for chunk in response.aiter_lines():
                        if chunk.startswith("0:"):
                            try: full_response += json.loads(chunk[2:])
                            except: continue
                        elif chunk.startswith(("e:", "d:")):
                            try: usage_info = json.loads(chunk[2:]).get("usage", {})
                            except: continue

            tool_calls = None
            content_response = full_response
            if "<tool_call>" in full_response and "</tool_call>" in full_response:
                tool_call_str = full_response.split("<tool_call>")[1].split("</tool_call>")[0]
                tool_call = json.loads(tool_call_str.strip())
                tool_calls = [{"id": generate_random_id("call_"), "type": "function", "function": {"name": tool_call["name"], "arguments": json.dumps(tool_call["parameters"])}}]
                content_response = None
            
            return JSONResponse(content={
                "id": chat_id, "object": "chat.completion", "created": int(time.time()), "model": model_id,
                "choices": [{"index": 0, "message": {"role": "assistant", "content": content_response, "tool_calls": tool_calls}, "finish_reason": "stop" if not tool_calls else "tool_calls"}],
                "usage": {"prompt_tokens": usage_info.get("promptTokens", 0), "completion_tokens": usage_info.get("completionTokens", 0), "total_tokens": usage_info.get("promptTokens", 0) + usage_info.get("completionTokens", 0)}
            })
        except httpx.HTTPStatusError as e:
            return JSONResponse(status_code=e.response.status_code, content={"error": {"message": f"Upstream API error. Details: {e.response.text}", "type": "upstream_error"}})

@app.post("/v1/images/generations", tags=["Images"])
async def generate_images(request: ImageGenerationRequest):
    """Handles image generation requests."""
    results = []
    try:
        async with httpx.AsyncClient(timeout=120) as client:
            for _ in range(request.n):
                model = request.model or "default"
                if model in ["gpt-image-1", "dall-e-3", "dall-e-2", "nextlm-image-1"]:
                    headers = {'Content-Type': 'application/json', 'User-Agent': 'Mozilla/5.0', 'Referer': 'https://www.chatwithmono.xyz/'}
                    payload = {"prompt": request.prompt, "model": model}
                    resp = await client.post(IMAGE_GEN_API_URL, headers=headers, json=payload)
                    resp.raise_for_status()
                    data = resp.json()
                    b64_image = data.get("image")
                    if not b64_image: return JSONResponse(status_code=502, content={"error": "Missing base64 image in response"})
                    
                    image_url = f"data:image/png;base64,{b64_image}"
                    if SNAPZION_API_KEY:
                        upload_headers = {"Authorization": SNAPZION_API_KEY}
                        upload_files = {'file': ('image.png', base64.b64decode(b64_image), 'image/png')}
                        upload_resp = await client.post(SNAPZION_UPLOAD_URL, headers=upload_headers, files=upload_files)
                        if upload_resp.status_code == 200:
                            image_url = upload_resp.json().get("url", image_url)
                    
                    results.append({"url": image_url, "b64_json": b64_image, "revised_prompt": data.get("revised_prompt")})
                else:
                    params = {"prompt": request.prompt, "aspect_ratio": request.aspect_ratio, "link": "typegpt.net"}
                    resp = await client.get(IMAGE_API_URL, params=params)
                    resp.raise_for_status()
                    data = resp.json()
                    results.append({"url": data.get("image_link"), "b64_json": data.get("base64_output")})
    except httpx.HTTPStatusError as e:
        return JSONResponse(status_code=502, content={"error": f"Image generation failed. Upstream error: {e.response.status_code}", "details": e.response.text})
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": "An internal error occurred.", "details": str(e)})
    return {"created": int(time.time()), "data": results}

@app.post("/v1/ocr", response_model=OcrResponse, tags=["OCR"])
async def perform_ocr(request: OcrRequest):
    """
    Performs Optical Character Recognition (OCR) on an image using the Florence-2 model.
    Provide an image via a URL or a base64 encoded string.
    """
    if not ocr_client:
        raise HTTPException(status_code=503, detail="OCR service is not available. Gradio client failed to initialize.")

    image_path, temp_file_path = None, None
    try:
        if request.image_url:
            image_path = request.image_url
        elif request.image_b64:
            image_bytes = base64.b64decode(request.image_b64)
            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_file:
                temp_file.write(image_bytes)
                temp_file_path = temp_file.name
            image_path = temp_file_path

        prediction = ocr_client.predict(image=handle_file(image_path), task_prompt="OCR", api_name="/process_image")
        
        if not prediction or not isinstance(prediction, tuple):
             raise HTTPException(status_code=502, detail="Invalid response from OCR service.")
        
        raw_result = prediction[0]
        ocr_text = raw_result.get("OCR", "")
        return OcrResponse(ocr_text=ocr_text, raw_response=raw_result)
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"An error occurred during OCR processing: {str(e)}")
    finally:
        if temp_file_path:
            os.unlink(temp_file_path)

@app.post("/v1/moderations", tags=["Moderation"])
async def create_moderation(request: ModerationRequest):
    """Handles moderation requests, conforming to the OpenAI API specification."""
    input_texts = [request.input] if isinstance(request.input, str) else request.input
    if not input_texts:
         return JSONResponse(status_code=400, content={"error": {"message": "Request must have at least one input string."}})
    headers = {'Content-Type': 'application/json', 'User-Agent': 'Mozilla/5.0', 'Referer': 'https://www.chatwithmono.xyz/'}
    results = []
    try:
        async with httpx.AsyncClient(timeout=30) as client:
            for text_input in input_texts:
                resp = await client.post(MODERATION_API_URL, headers=headers, json={"text": text_input})
                resp.raise_for_status()
                upstream_data = resp.json()
                upstream_categories = upstream_data.get("categories", {})
                openai_categories = {
                    "hate": upstream_categories.get("hate", False), "hate/threatening": False, "harassment": False, "harassment/threatening": False,
                    "self-harm": upstream_categories.get("self-harm", False), "self-harm/intent": False, "self-harm/instructions": False,
                    "sexual": upstream_categories.get("sexual", False), "sexual/minors": False,
                    "violence": upstream_categories.get("violence", False), "violence/graphic": False,
                }
                result_item = {
                    "flagged": upstream_data.get("overall_sentiment") == "flagged",
                    "categories": openai_categories,
                    "category_scores": {k: 1.0 if v else 0.0 for k, v in openai_categories.items()},
                }
                if reason := upstream_data.get("reason"):
                    result_item["reason"] = reason
                results.append(result_item)
    except httpx.HTTPStatusError as e:
        return JSONResponse(status_code=502, content={"error": {"message": f"Moderation failed. Upstream error: {e.response.status_code}", "details": e.response.text}})
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": {"message": "An internal error occurred during moderation.", "details": str(e)}})
    
    return JSONResponse(content={"id": generate_random_id("modr-"), "model": request.model, "results": results})


# --- Main Execution ---
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)