0.2.5.0

33 files changed, 993 insertions, 565 deletions

diff --git a/g4f/Provider/Bing.py b/g4f/Provider/Bing.py
index 786fec49..69c32775 100644
--- a/g4f/Provider/Bing.py
+++ b/g4f/Provider/Bing.py
@@ -12,7 +12,7 @@ from aiohttp import ClientSession, ClientTimeout, BaseConnector, WSMsgType
 from ..typing import AsyncResult, Messages, ImageType, Cookies
 from ..image import ImageRequest
 from ..errors import ResponseStatusError
-from .base_provider import AsyncGeneratorProvider
+from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
 from .helper import get_connector, get_random_hex
 from .bing.upload_image import upload_image
 from .bing.conversation import Conversation, create_conversation, delete_conversation
@@ -26,8 +26,9 @@ class Tones:
     creative = "Creative"
     balanced = "Balanced"
     precise = "Precise"
+    copilot = "Balanced"
 
-class Bing(AsyncGeneratorProvider):
+class Bing(AsyncGeneratorProvider, ProviderModelMixin):
     """
     Bing provider for generating responses using the Bing API.
     """
@@ -35,18 +36,22 @@ class Bing(AsyncGeneratorProvider):
     working = True
     supports_message_history = True
     supports_gpt_4 = True
+    default_model = "balanced"
+    models = [key for key in Tones.__dict__ if not key.startswith("__")]
         
-    @staticmethod
+    @classmethod
     def create_async_generator(
+        cls,
         model: str,
         messages: Messages,
         proxy: str = None,
         timeout: int = 900,
         cookies: Cookies = None,
         connector: BaseConnector = None,
-        tone: str = Tones.balanced,
+        tone: str = None,
         image: ImageType = None,
         web_search: bool = False,
+        context: str = None,
         **kwargs
     ) -> AsyncResult:
         """
@@ -62,13 +67,12 @@ class Bing(AsyncGeneratorProvider):
         :param web_search: Flag to enable or disable web search.
         :return: An asynchronous result object.
         """
-        if len(messages) < 2:
-            prompt = messages[0]["content"]
-            context = None
-        else:
-            prompt = messages[-1]["content"]
-            context = create_context(messages[:-1])
-
+        prompt = messages[-1]["content"]
+        if context is None:
+            context = create_context(messages[:-1]) if len(messages) > 1 else None
+        if tone is None:
+            tone = tone if model.startswith("gpt-4") else model
+        tone = cls.get_model("" if tone is None else tone.lower())
         gpt4_turbo = True if model.startswith("gpt-4-turbo") else False
 
         return stream_generate(
@@ -86,7 +90,9 @@ def create_context(messages: Messages) -> str:
     :return: A string representing the context created from the messages.
     """
     return "".join(
-        f"[{message['role']}]" + ("(#message)" if message['role'] != "system" else "(#additional_instructions)") + f"\n{message['content']}"
+        f"[{message['role']}]" + ("(#message)"
+        if message['role'] != "system"
+        else "(#additional_instructions)") + f"\n{message['content']}"
         for message in messages
     ) + "\n\n"
 
@@ -122,7 +128,7 @@ class Defaults:
         "ActionRequest","Chat",
         "ConfirmationCard", "Context",
         "InternalSearchQuery", #"InternalSearchResult",
-        "Disengaged", #"InternalLoaderMessage",
+        #"Disengaged", "InternalLoaderMessage",
         "Progress", "RenderCardRequest",
         "RenderContentRequest", "AdsQuery",
         "SemanticSerp", "GenerateContentQuery",
@@ -131,53 +137,93 @@ class Defaults:
     ]
 
     sliceIds = {
-        "Balanced": [
+        "balanced": [
             "supllmnfe","archnewtf",
             "stpstream", "stpsig", "vnextvoicecf", "scmcbase", "cmcpupsalltf", "sydtransctrl",
             "thdnsrch", "220dcl1s0", "0215wcrwips0", "0305hrthrots0", "0130gpt4t",
             "bingfc", "0225unsticky1", "0228scss0",
             "defquerycf", "defcontrol", "3022tphpv"
         ],
-        "Creative": [
+        "creative": [
             "bgstream", "fltltst2c",
             "stpstream", "stpsig", "vnextvoicecf", "cmcpupsalltf", "sydtransctrl",
             "0301techgnd", "220dcl1bt15", "0215wcrwip", "0305hrthrot", "0130gpt4t",
             "bingfccf", "0225unsticky1", "0228scss0",
             "3022tpvs0"
         ],
-        "Precise": [
+        "precise": [
             "bgstream", "fltltst2c",
             "stpstream", "stpsig", "vnextvoicecf", "cmcpupsalltf", "sydtransctrl",
             "0301techgnd", "220dcl1bt15", "0215wcrwip", "0305hrthrot", "0130gpt4t",
             "bingfccf", "0225unsticky1", "0228scss0",
             "defquerycf", "3022tpvs0"
         ],
+        "copilot": []
     }
 
     optionsSets = {
-        "Balanced": [
-             "nlu_direct_response_filter", "deepleo",
-            "disable_emoji_spoken_text", "responsible_ai_policy_235",
-            "enablemm", "dv3sugg", "autosave",
-            "iyxapbing", "iycapbing",
-            "galileo", "saharagenconv5", "gldcl1p",
-            "gpt4tmncnp"
-        ],
-        "Creative": [
+        "balanced": {
+            "default": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg", "autosave",
+                "iyxapbing", "iycapbing",
+                "galileo", "saharagenconv5", "gldcl1p",
+                "gpt4tmncnp"
+            ],
+            "nosearch": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg", "autosave",
+                "iyxapbing", "iycapbing",
+                "galileo", "sunoupsell", "base64filter", "uprv4p1upd",
+                "hourthrot", "noctprf", "gndlogcf", "nosearchall"
+            ]
+        },
+        "creative": {
+            "default": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg",
+                "iyxapbing", "iycapbing",
+                "h3imaginative", "techinstgnd", "hourthrot", "clgalileo", "gencontentv3",
+                "gpt4tmncnp"
+            ],
+            "nosearch": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg", "autosave",
+                "iyxapbing", "iycapbing",
+                "h3imaginative", "sunoupsell", "base64filter", "uprv4p1upd",
+                "hourthrot", "noctprf", "gndlogcf", "nosearchall",
+                "clgalileo", "nocache", "up4rp14bstcst"
+            ]
+        },
+        "precise": {
+            "default": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg",
+                "iyxapbing", "iycapbing",
+                "h3precise", "techinstgnd", "hourthrot", "techinstgnd", "hourthrot",
+                "clgalileo", "gencontentv3"
+            ],
+            "nosearch": [
+                "nlu_direct_response_filter", "deepleo",
+                "disable_emoji_spoken_text", "responsible_ai_policy_235",
+                "enablemm", "dv3sugg", "autosave",
+                "iyxapbing", "iycapbing",
+                "h3precise", "sunoupsell", "base64filter", "uprv4p1upd",
+                "hourthrot", "noctprf", "gndlogcf", "nosearchall",
+                "clgalileo", "nocache", "up4rp14bstcst"
+            ]
+        },
+        "copilot": [
             "nlu_direct_response_filter", "deepleo",
             "disable_emoji_spoken_text", "responsible_ai_policy_235",
             "enablemm", "dv3sugg",
             "iyxapbing", "iycapbing",
-            "h3imaginative", "techinstgnd", "hourthrot", "clgalileo", "gencontentv3",
-            "gpt4tmncnp"
-        ],
-        "Precise": [
-            "nlu_direct_response_filter", "deepleo",
-            "disable_emoji_spoken_text", "responsible_ai_policy_235",
-            "enablemm", "dv3sugg",
-            "iyxapbing", "iycapbing",
-            "h3precise", "techinstgnd", "hourthrot", "techinstgnd", "hourthrot",
-            "clgalileo", "gencontentv3"
+            "h3precise", "clgalileo", "gencontentv3", "prjupy"
         ],
     }
 
@@ -232,7 +278,8 @@ def create_message(
     context: str = None,
     image_request: ImageRequest = None,
     web_search: bool = False,
-    gpt4_turbo: bool = False
+    gpt4_turbo: bool = False,
+    new_conversation: bool = True
 ) -> str:
     """
     Creates a message for the Bing API with specified parameters.
@@ -247,7 +294,12 @@ def create_message(
     :return: A formatted string message for the Bing API.
     """
 
-    options_sets = []
+    options_sets = Defaults.optionsSets[tone]
+    if not web_search and "nosearch" in options_sets:
+        options_sets = options_sets["nosearch"]
+    elif "default" in options_sets:
+        options_sets = options_sets["default"]
+    options_sets = options_sets.copy()
     if gpt4_turbo:
         options_sets.append("dlgpt4t")
 
@@ -255,16 +307,16 @@ def create_message(
     struct = {
         "arguments":[{
             "source": "cib",
-            "optionsSets": [*Defaults.optionsSets[tone], *options_sets],
+            "optionsSets": options_sets,
             "allowedMessageTypes": Defaults.allowedMessageTypes,
             "sliceIds": Defaults.sliceIds[tone],
             "verbosity": "verbose",
-            "scenario": "SERP",
+            "scenario": "CopilotMicrosoftCom", # "SERP",
             "plugins": [{"id": "c310c353-b9f0-4d76-ab0d-1dd5e979cf68", "category": 1}] if web_search else [],
             "traceId": get_random_hex(40),
             "conversationHistoryOptionsSets": ["autosave","savemem","uprofupd","uprofgen"],
             "gptId": "copilot",
-            "isStartOfSession": True,
+            "isStartOfSession": new_conversation,
             "requestId": request_id,
             "message":{
                 **Defaults.location,
@@ -277,8 +329,7 @@ def create_message(
                 "requestId": request_id,
                 "messageId": request_id
             },
-            "tone": tone,
-            "extraExtensionParameters": {"gpt-creator-persona": {"personaId": "copilot"}},
+            "tone": getattr(Tones, tone),
             "spokenTextMode": "None",
             "conversationId": conversation.conversationId,
             "participant": {"id": conversation.clientId}
@@ -298,7 +349,7 @@ def create_message(
         struct['arguments'][0]['previousMessages'] = [{
             "author": "user",
             "description": context,
-            "contextType": "WebPage",
+            "contextType": "ClientApp",
             "messageType": "Context",
             "messageId": "discover-web--page-ping-mriduna-----"
         }]
@@ -317,8 +368,9 @@ async def stream_generate(
     gpt4_turbo: bool = False,
     timeout: int = 900,
     conversation: Conversation = None,
+    return_conversation: bool = False,
     raise_apology: bool = False,
-    max_retries: int = 5,
+    max_retries: int = None,
     sleep_retry: int = 15,
     **kwargs
 ):
@@ -336,13 +388,20 @@ async def stream_generate(
     :return: An asynchronous generator yielding responses.
     """
     headers = create_headers(cookies)
+    new_conversation = conversation is None
+    max_retries = (5 if new_conversation else 0) if max_retries is None else max_retries
     async with ClientSession(
         timeout=ClientTimeout(total=timeout), connector=connector
     ) as session:
-        while conversation is None:
+        first = True
+        while first or conversation is None:
+            first = False
             do_read = True
             try:
-                conversation = await create_conversation(session, headers)
+                if conversation is None:
+                    conversation = await create_conversation(session, headers, tone)
+                if return_conversation:
+                    yield conversation
             except ResponseStatusError as e:
                 max_retries -= 1
                 if max_retries < 1:
@@ -353,8 +412,10 @@ async def stream_generate(
                 await asyncio.sleep(sleep_retry)
                 continue
 
-            image_request = await upload_image(session, image, tone, headers) if image else None
+            image_request = await upload_image(session, image, getattr(Tones, tone), headers) if image else None
             async with session.ws_connect(
+                'wss://s.copilot.microsoft.com/sydney/ChatHub'
+                if tone == "copilot" else
                 'wss://sydney.bing.com/sydney/ChatHub',
                 autoping=False,
                 params={'sec_access_token': conversation.conversationSignature},
@@ -363,7 +424,12 @@ async def stream_generate(
                 await wss.send_str(format_message({'protocol': 'json', 'version': 1}))
                 await wss.send_str(format_message({"type": 6}))
                 await wss.receive(timeout=timeout)
-                await wss.send_str(create_message(conversation, prompt, tone, context, image_request, web_search, gpt4_turbo))
+                await wss.send_str(create_message(
+                    conversation, prompt, tone,
+                    context if new_conversation else None,
+                    image_request, web_search, gpt4_turbo,
+                    new_conversation
+                ))
                 response_txt = ''
                 returned_text = ''
                 message_id = None
@@ -399,14 +465,15 @@ async def stream_generate(
                                     image_client = BingCreateImages(cookies, proxy)
                                     image_response = await image_client.create_async(prompt)
                                 except Exception as e:
-                                    response_txt += f"\nhttps://www.bing.com/images/create?q={parse.quote(prompt)}"
-                                do_read = False
+                                    if debug.logging:
+                                        print(f"Bing: Failed to create images: {e}")
+                                    image_response = f"\nhttps://www.bing.com/images/create?q={parse.quote(prompt)}"
                             if response_txt.startswith(returned_text):
                                 new = response_txt[len(returned_text):]
-                                if new != "\n":
+                                if new not in ("", "\n"):
                                     yield new
                                     returned_text = response_txt
-                            if image_response:
+                            if image_response is not None:
                                 yield image_response
                         elif response.get('type') == 2:
                             result = response['item']['result']
diff --git a/g4f/Provider/You.py b/g4f/Provider/You.py
index 85b60452..9b040367 100644
--- a/g4f/Provider/You.py
+++ b/g4f/Provider/You.py
@@ -4,14 +4,18 @@ import re
 import json
 import base64
 import uuid
-from asyncio import get_running_loop
-from aiohttp import ClientSession, FormData, BaseConnector, CookieJar
+try:
+    from ..requests.curl_cffi import FormData
+    has_curl_cffi = True
+except ImportError:
+    has_curl_cffi = False
 
 from ..typing import AsyncResult, Messages, ImageType, Cookies
 from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
-from .helper import format_prompt, get_connector
+from .helper import format_prompt
 from ..image import to_bytes, ImageResponse
-from ..requests import WebDriver, raise_for_status, get_args_from_browser
+from ..requests import StreamSession, raise_for_status
+from ..errors import MissingRequirementsError
 
 class You(AsyncGeneratorProvider, ProviderModelMixin):
     url = "https://you.com"
@@ -33,8 +37,6 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
     model_aliases = {
         "claude-v2": "claude-2"
     }
-    _args: dict = None
-    _cookie_jar: CookieJar = None
     _cookies = None
     _cookies_used = 0
 
@@ -45,19 +47,12 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
         messages: Messages,
         image: ImageType = None,
         image_name: str = None,
-        connector: BaseConnector = None,
-        webdriver: WebDriver = None,
         proxy: str = None,
         chat_mode: str = "default",
         **kwargs,
     ) -> AsyncResult:
-        if cls._args is None:
-            cls._args = get_args_from_browser(cls.url, webdriver, proxy)
-            cls._cookie_jar = CookieJar(loop=get_running_loop())
-        else:
-            if "cookies" in cls._args:
-                del cls._args["cookies"]
-            cls._cookie_jar._loop = get_running_loop()
+        if not has_curl_cffi:
+            raise MissingRequirementsError('Install "curl_cffi" package')
         if image is not None:
             chat_mode = "agent"
         elif not model or model == cls.default_model:
@@ -67,10 +62,9 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
         else:
             chat_mode = "custom"
             model = cls.get_model(model)
-        async with ClientSession(
-            connector=get_connector(connector, proxy),
-            cookie_jar=cls._cookie_jar,
-            **cls._args
+        async with StreamSession(
+            proxy=proxy,
+            impersonate="chrome"
         ) as session:
             cookies = await cls.get_cookies(session) if chat_mode != "default" else None
             upload = json.dumps([await cls.upload_file(session, cookies, to_bytes(image), image_name)]) if image else ""
@@ -82,8 +76,8 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
             #     and idx < len(questions)
             # ]
             headers = {
-                "accept": "text/event-stream",
-                "referer": f"{cls.url}/search?fromSearchBar=true&tbm=youchat",
+                "Accept": "text/event-stream",
+                "Referer": f"{cls.url}/search?fromSearchBar=true&tbm=youchat",
             }
             data = {
                 "userFiles": upload,
@@ -106,12 +100,12 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
                 cookies=cookies
             ) as response:
                 await raise_for_status(response)
-                async for line in response.content:
+                async for line in response.iter_lines():
                     if line.startswith(b'event: '):
-                        event = line[7:-1].decode()
+                        event = line[7:].decode()
                     elif line.startswith(b'data: '):
                         if event in ["youChatUpdate", "youChatToken"]:
-                            data = json.loads(line[6:-1])
+                            data = json.loads(line[6:])
                         if event == "youChatToken" and event in data:
                             yield data[event]
                         elif event == "youChatUpdate" and "t" in data:
@@ -122,7 +116,7 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
                                 yield data["t"]                         
 
     @classmethod
-    async def upload_file(cls, client: ClientSession, cookies: Cookies, file: bytes, filename: str = None) -> dict:
+    async def upload_file(cls, client: StreamSession, cookies: Cookies, file: bytes, filename: str = None) -> dict:
         async with client.get(
             f"{cls.url}/api/get_nonce",
             cookies=cookies,
@@ -146,7 +140,7 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
         return result
 
     @classmethod
-    async def get_cookies(cls, client: ClientSession) -> Cookies:
+    async def get_cookies(cls, client: StreamSession) -> Cookies:
         if not cls._cookies or cls._cookies_used >= 5:
             cls._cookies = await cls.create_cookies(client)
             cls._cookies_used = 0
@@ -173,7 +167,7 @@ class You(AsyncGeneratorProvider, ProviderModelMixin):
         return f"Basic {auth}"
 
     @classmethod
-    async def create_cookies(cls, client: ClientSession) -> Cookies:
+    async def create_cookies(cls, client: StreamSession) -> Cookies:
         user_uuid = str(uuid.uuid4())
         async with client.post(
             "https://web.stytch.com/sdk/v1/passwords",
diff --git a/g4f/Provider/bing/conversation.py b/g4f/Provider/bing/conversation.py
index da842808..886efa68 100644
--- a/g4f/Provider/bing/conversation.py
+++ b/g4f/Provider/bing/conversation.py
@@ -20,7 +20,7 @@ class Conversation:
         self.clientId = clientId
         self.conversationSignature = conversationSignature
 
-async def create_conversation(session: ClientSession, headers: dict) -> Conversation:
+async def create_conversation(session: ClientSession, headers: dict, tone: str) -> Conversation:
     """
     Create a new conversation asynchronously.
 
@@ -31,7 +31,10 @@ async def create_conversation(session: ClientSession, headers: dict) -> Conversa
     Returns:
     Conversation: An instance representing the created conversation.
     """
-    url = "https://www.bing.com/turing/conversation/create?bundleVersion=1.1626.1"
+    if tone == "copilot":
+        url = "https://copilot.microsoft.com/turing/conversation/create?bundleVersion=1.1634.3-nodesign2"
+    else:
+        url = "https://www.bing.com/turing/conversation/create?bundleVersion=1.1626.1"
     async with session.get(url, headers=headers) as response:
         await raise_for_status(response, "Failed to create conversation")
         data = await response.json()
diff --git a/g4f/Provider/needs_auth/OpenaiChat.py b/g4f/Provider/needs_auth/OpenaiChat.py
index 3d19e003..6601f500 100644
--- a/g4f/Provider/needs_auth/OpenaiChat.py
+++ b/g4f/Provider/needs_auth/OpenaiChat.py
@@ -15,6 +15,12 @@ except ImportError:
     has_arkose_generator = False
 
 try:
+    import webview
+    has_webview = True
+except ImportError:
+    has_webview = False
+
+try:
     from selenium.webdriver.common.by import By
     from selenium.webdriver.support.ui import WebDriverWait
     from selenium.webdriver.support import expected_conditions as EC
@@ -25,10 +31,10 @@ from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin
 from ..helper import get_cookies
 from ...webdriver import get_browser
 from ...typing import AsyncResult, Messages, Cookies, ImageType, Union, AsyncIterator
-from ...requests import get_args_from_browser
+from ...requests import get_args_from_browser, raise_for_status
 from ...requests.aiohttp import StreamSession
 from ...image import to_image, to_bytes, ImageResponse, ImageRequest
-from ...errors import MissingRequirementsError, MissingAuthError
+from ...errors import MissingRequirementsError, MissingAuthError, ProviderNotWorkingError
 from ... import debug
 
 class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
@@ -134,7 +140,8 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
         }
         # Post the image data to the service and get the image data
         async with session.post(f"{cls.url}/backend-api/files", json=data, headers=headers) as response:
-            response.raise_for_status()
+            cls._update_request_args()
+            await raise_for_status(response)
             image_data = {
                 **data,
                 **await response.json(),
@@ -152,14 +159,15 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                 "x-ms-blob-type": "BlockBlob"
             }
         ) as response:
-            response.raise_for_status()
+            await raise_for_status(response)
         # Post the file ID to the service and get the download URL
         async with session.post(
             f"{cls.url}/backend-api/files/{image_data['file_id']}/uploaded",
             json={},
             headers=headers
         ) as response:
-            response.raise_for_status()
+            cls._update_request_args(session)
+            await raise_for_status(response)
             image_data["download_url"] = (await response.json())["download_url"]
         return ImageRequest(image_data)
 
@@ -178,7 +186,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
         if not cls.default_model:
             async with session.get(f"{cls.url}/backend-api/models", headers=headers) as response:
                 cls._update_request_args(session)
-                response.raise_for_status()
+                await raise_for_status(response)
                 data = await response.json()
                 if "categories" in data:
                     cls.default_model = data["categories"][-1]["default_model"]
@@ -261,7 +269,8 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
         file_id = first_part["asset_pointer"].split("file-service://", 1)[1]
         try:
             async with session.get(f"{cls.url}/backend-api/files/{file_id}/download", headers=headers) as response:
-                response.raise_for_status()
+                cls._update_request_args(session)
+                await raise_for_status(response)
                 download_url = (await response.json())["download_url"]
                 return ImageResponse(download_url, prompt)
         except Exception as e:
@@ -288,6 +297,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
             json={"is_visible": False},
             headers=headers
         ) as response:
+            cls._update_request_args(session)
             ...
 
     @classmethod
@@ -337,31 +347,32 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
         if parent_id is None:
             parent_id = str(uuid.uuid4())
 
-        # Read api_key from arguments
-        api_key = kwargs["access_token"] if "access_token" in kwargs else api_key
-
         async with StreamSession(
             proxies={"https": proxy},
             impersonate="chrome",
             timeout=timeout
         ) as session:
-            # Read api_key and cookies from cache / browser config
+            api_key = kwargs["access_token"] if "access_token" in kwargs else api_key
             if cls._headers is None or cls._expires is None or time.time() > cls._expires:
-                if api_key is None:
-                    # Read api_key from cookies
+                if cls._headers is None:
                     cookies = get_cookies("chat.openai.com", False) if cookies is None else cookies
                     api_key = cookies["access_token"] if "access_token" in cookies else api_key
-                cls._create_request_args(cookies)
+                if api_key is None:
+                    try:
+                        await cls.webview_access_token() if has_webview else None
+                    except Exception as e:
+                        if debug.logging:
+                            print(f"Use webview failed: {e}")
             else:
                 api_key = cls._api_key if api_key is None else api_key
-            # Read api_key with session cookies
-            #if api_key is None and cookies:
-            #    api_key = await cls.fetch_access_token(session, cls._headers)
-            # Load default model
-            if cls.default_model is None and api_key is not None:
+
+            if api_key is not None:
+                cls._create_request_args(cookies)
+                cls._set_api_key(api_key)
+
+            if cls.default_model is None and cls._headers is not None:
                 try:
                     if not model:
-                        cls._set_api_key(api_key)
                         cls.default_model = cls.get_model(await cls.get_default_model(session, cls._headers))
                     else:
                         cls.default_model = cls.get_model(model)
@@ -369,8 +380,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                     if debug.logging:
                         print("OpenaiChat: Load default_model failed")
                         print(f"{e.__class__.__name__}: {e}")
-            # Browse api_key and default model
-            if api_key is None or cls.default_model is None:
+            if cls.default_model is None:
                 login_url = os.environ.get("G4F_LOGIN_URL")
                 if login_url:
                     yield f"Please login: [ChatGPT]({login_url})\n\n"
@@ -379,20 +389,21 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                 except MissingRequirementsError:
                     raise MissingAuthError(f'Missing "access_token". Add a "api_key" please')
                 cls.default_model = cls.get_model(await cls.get_default_model(session, cls._headers))
-            else:
-                cls._set_api_key(api_key)
 
             async with session.post(
                 f"{cls.url}/backend-api/sentinel/chat-requirements",
                 json={"conversation_mode_kind": "primary_assistant"},
                 headers=cls._headers
             ) as response:
-                response.raise_for_status()
+                cls._update_request_args(session)
+                await raise_for_status(response)
                 data = await response.json()
+                blob = data["arkose"]["dx"]
                 need_arkose = data["arkose"]["required"]
                 chat_token = data["token"]
 
             if need_arkose and not has_arkose_generator:
+                raise ProviderNotWorkingError("OpenAI Plus Subscriber are not working")
                 raise MissingRequirementsError('Install "py-arkose-generator" package')
 
             try:
@@ -407,6 +418,7 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
             while fields.finish_reason is None:
                 conversation_id = conversation_id if fields.conversation_id is None else fields.conversation_id
                 parent_id = parent_id if fields.message_id is None else fields.message_id
+                websocket_request_id = str(uuid.uuid4())
                 data = {
                     "action": action,
                     "conversation_mode": {"kind": "primary_assistant"},
@@ -416,25 +428,29 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                     "parent_message_id": parent_id,
                     "model": model,
                     "history_and_training_disabled": history_disabled and not auto_continue,
+                    "websocket_request_id": websocket_request_id
                 }
                 if action != "continue":
                     messages = messages if conversation_id is None else [messages[-1]]
-                    data["messages"] = cls.create_messages(messages, image_request)                
+                    data["messages"] = cls.create_messages(messages, image_request)
+                headers = {
+                    "Accept": "text/event-stream",
+                    "OpenAI-Sentinel-Chat-Requirements-Token": chat_token,
+                    **cls._headers
+                }
+                if need_arkose:
+                    raise ProviderNotWorkingError("OpenAI Plus Subscriber are not working")
+                    headers["OpenAI-Sentinel-Arkose-Token"] = await cls.get_arkose_token(session, cls._headers, blob)
+                    headers["OpenAI-Sentinel-Chat-Requirements-Token"] = chat_token
 
                 async with session.post(
                     f"{cls.url}/backend-api/conversation",
                     json=data,
-                    headers={
-                        "Accept": "text/event-stream",
-                        **({"OpenAI-Sentinel-Arkose-Token": await cls.get_arkose_token(session)} if need_arkose else {}),
-                        "OpenAI-Sentinel-Chat-Requirements-Token": chat_token,
-                        **cls._headers
-                    }
+                    headers=headers
                 ) as response:
                     cls._update_request_args(session)
-                    if not response.ok:
-                        raise RuntimeError(f"Response {response.status}: {await response.text()}")
-                    async for chunk in cls.iter_messages_chunk(response.iter_lines(), session, fields):
+                    await raise_for_status(response)
+                    async for chunk in cls.iter_messages_chunk(response.iter_lines(), session, fields, websocket_request_id):
                         if response_fields:
                             response_fields = False
                             yield fields
@@ -447,21 +463,35 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
                 await cls.delete_conversation(session, cls._headers, fields.conversation_id)
 
     @staticmethod
-    async def iter_messages_ws(ws: ClientWebSocketResponse, conversation_id: str) -> AsyncIterator:
+    async def iter_messages_ws(ws: ClientWebSocketResponse, conversation_id: str, is_curl: bool) -> AsyncIterator:
         while True:
-            message = await ws.receive_json()
+            if is_curl:
+                message = json.loads(ws.recv()[0])
+            else:
+                message = await ws.receive_json()
             if message["conversation_id"] == conversation_id:
                 yield base64.b64decode(message["body"])
 
     @classmethod
-    async def iter_messages_chunk(cls, messages: AsyncIterator, session: StreamSession, fields: ResponseFields) -> AsyncIterator:
+    async def iter_messages_chunk(
+        cls,
+        messages: AsyncIterator,
+        session: StreamSession,
+        fields: ResponseFields
+    ) -> AsyncIterator:
         last_message: int = 0
         async for message in messages:
             if message.startswith(b'{"wss_url":'):
                 message = json.loads(message)
-                async with session.ws_connect(message["wss_url"]) as ws:
-                    async for chunk in cls.iter_messages_chunk(cls.iter_messages_ws(ws, message["conversation_id"]), session, fields):
+                ws = await session.ws_connect(message["wss_url"])
+                try:
+                    async for chunk in cls.iter_messages_chunk(
+                        cls.iter_messages_ws(ws, message["conversation_id"], hasattr(ws, "recv")),
+                        session, fields
+                    ):
                         yield chunk
+                finally:
+                    await ws.aclose()
                 break
             async for chunk in cls.iter_messages_line(session, message, fields):
                 if fields.finish_reason is not None:
@@ -514,6 +544,43 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
             fields.finish_reason = line["message"]["metadata"]["finish_details"]["type"]
 
     @classmethod
+    async def webview_access_token(cls) -> str:
+        window = webview.create_window("OpenAI Chat", cls.url)
+        await asyncio.sleep(3)
+        prompt_input = None
+        while not prompt_input:
+            try:
+                await asyncio.sleep(1)
+                prompt_input = window.dom.get_element("#prompt-textarea")
+            except:
+                ...
+        window.evaluate_js("""
+this._fetch = this.fetch;
+this.fetch = async (url, options) => {
+    const response = await this._fetch(url, options);
+    if (url == "https://chat.openai.com/backend-api/conversation") {
+        this._headers = options.headers;
+        return response;
+    }
+    return response;
+};
+""")
+        window.evaluate_js("""
+            document.querySelector('.from-token-main-surface-secondary').click();
+        """)
+        headers = None
+        while headers is None:
+            headers = window.evaluate_js("this._headers")
+            await asyncio.sleep(1)
+        headers["User-Agent"] = window.evaluate_js("this.navigator.userAgent")
+        cookies = [list(*cookie.items()) for cookie in window.get_cookies()]
+        window.destroy()
+        cls._cookies = dict([(name, cookie.value) for name, cookie in cookies])
+        cls._headers = headers
+        cls._expires = int(time.time()) + 60 * 60 * 4
+        cls._update_cookie_header()
+
+    @classmethod
     def browse_access_token(cls, proxy: str = None, timeout: int = 1200) -> None:
         """
         Browse to obtain an access token.
@@ -542,10 +609,10 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
             cls._update_cookie_header()
             cls._set_api_key(access_token)
         finally:
-            driver.close() 
+            driver.close()
 
     @classmethod
-    async def get_arkose_token(cls, session: StreamSession) -> str:
+    async def get_arkose_token(cls, session: StreamSession, headers: dict, blob: str) -> str:
         """
         Obtain an Arkose token for the session.
 
@@ -559,16 +626,15 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
             RuntimeError: If unable to retrieve the token.
         """
         config = {
-            "pkey": "3D86FBBA-9D22-402A-B512-3420086BA6CC",
+            "pkey": "35536E1E-65B4-4D96-9D97-6ADB7EFF8147",
             "surl": "https://tcr9i.chat.openai.com",
-            "headers": {
-                "User-Agent": 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/107.0.0.0 Safari/537.36'
-            },
+            "headers": headers,
             "site": cls.url,
+            "data": {"blob": blob}
         }
         args_for_request = get_values_for_request(config)
         async with session.post(**args_for_request) as response:
-            response.raise_for_status()
+            await raise_for_status(response)
             decoded_json = await response.json()
             if "token" in decoded_json:
                 return decoded_json["token"]
@@ -591,7 +657,9 @@ class OpenaiChat(AsyncGeneratorProvider, ProviderModelMixin):
 
     @classmethod
     def _create_request_args(cls, cookies: Union[Cookies, None]):
-        cls._headers = {}
+        cls._headers = {
+            "User-Agent": 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/107.0.0.0 Safari/537.36'
+        }
         cls._cookies = {} if cookies is None else cookies
         cls._update_cookie_header()
 
diff --git a/g4f/__init__.py b/g4f/__init__.py
index 441225f1..dce116c4 100644
--- a/g4f/__init__.py
+++ b/g4f/__init__.py
@@ -69,11 +69,14 @@ def get_model_and_provider(model    : Union[Model, str],
     if isinstance(model, Model):
         model = model.name
 
-    if ignored and isinstance(provider, BaseRetryProvider):
-        provider.providers = [p for p in provider.providers if p.__name__ not in ignored]
-
     if not ignore_working and not provider.working:
         raise ProviderNotWorkingError(f'{provider.__name__} is not working')
+
+    if not ignore_working and isinstance(provider, BaseRetryProvider):
+        provider.providers = [p for p in provider.providers if p.working]
+
+    if ignored and isinstance(provider, BaseRetryProvider):
+        provider.providers = [p for p in provider.providers if p.__name__ not in ignored]
     
     if not ignore_stream and not provider.supports_stream and stream:
         raise StreamNotSupportedError(f'{provider.__name__} does not support "stream" argument')
diff --git a/g4f/debug.py b/g4f/debug.py
index 69f7d55c..3f08e9e8 100644
--- a/g4f/debug.py
+++ b/g4f/debug.py
@@ -3,4 +3,5 @@ from .providers.types import ProviderType
 logging: bool = False
 version_check: bool = True
 last_provider: ProviderType = None
-last_model: str = None
\ No newline at end of file
+last_model: str = None
+version: str = None
\ No newline at end of file
diff --git a/g4f/gui/client/html/index.html b/g4f/gui/client/index.html
index 46a9c541..9ce6b66a 100644
--- a/g4f/gui/client/html/index.html
+++ b/g4f/gui/client/index.html
@@ -1,5 +1,5 @@
 <!DOCTYPE html>
-<html lang="en">
+<html lang="en" data-framework="javascript">
 
 <head>
     <meta charset="UTF-8">
@@ -10,14 +10,14 @@
     <meta property="og:image" content="https://openai.com/content/images/2022/11/ChatGPT.jpg">
     <meta property="og:description" content="A conversational AI system that listens, learns, and challenges">
     <meta property="og:url" content="https://g4f.ai">
-    <link rel="stylesheet" href="/assets/css/style.css">
-    <link rel="apple-touch-icon" sizes="180x180" href="/assets/img/apple-touch-icon.png">
-    <link rel="icon" type="image/png" sizes="32x32" href="/assets/img/favicon-32x32.png">
-    <link rel="icon" type="image/png" sizes="16x16" href="/assets/img/favicon-16x16.png">
-    <link rel="manifest" href="/assets/img/site.webmanifest">
-    <script src="/assets/js/icons.js"></script>
-    <script src="/assets/js/highlightjs-copy.min.js"></script>
-    <script src="/assets/js/chat.v1.js" defer></script>
+    <link rel="stylesheet" href="/static/css/style.css">
+    <link rel="apple-touch-icon" sizes="180x180" href="/static/img/apple-touch-icon.png">
+    <link rel="icon" type="image/png" sizes="32x32" href="/static/img/favicon-32x32.png">
+    <link rel="icon" type="image/png" sizes="16x16" href="/static/img/favicon-16x16.png">
+    <link rel="manifest" href="/static/img/site.webmanifest">
+    <script src="/static/js/icons.js"></script>
+    <script src="/static/js/highlightjs-copy.min.js"></script>
+    <script src="/static/js/chat.v1.js" defer></script>
     <script src="https://cdn.jsdelivr.net/npm/markdown-it@13.0.1/dist/markdown-it.min.js"></script>
     <link rel="stylesheet"
         href="//cdn.jsdelivr.net/gh/highlightjs/cdn-release@11.7.0/build/styles/base16/dracula.min.css">
@@ -38,8 +38,8 @@
     </script>
     <script src="https://unpkg.com/gpt-tokenizer/dist/cl100k_base.js" async></script>
     <script>
-        const user_image = '<img src="/assets/img/user.png" alt="your avatar">';
-        const gpt_image = '<img src="/assets/img/gpt.png" alt="your avatar">';
+        const user_image = '<img src="/static/img/user.png" alt="your avatar">';
+        const gpt_image = '<img src="/static/img/gpt.png" alt="your avatar">';
     </script>
     <style>
         .hljs {
@@ -74,8 +74,8 @@
             background: #8b3dff;
         }
     </style>
-    <script src="/assets/js/highlight.min.js"></script>
-    <script>window.conversation_id = `{{chat_id}}`</script>
+    <script src="/static/js/highlight.min.js"></script>
+    <script>window.conversation_id = "{{chat_id}}"</script>
     <title>g4f - gui</title>
 </head>
 
@@ -94,11 +94,10 @@
                     <i class="fa-regular fa-trash"></i>
                     <span>Clear Conversations</span>
                 </button>
-                <div class="info">
-                    <i class="fa-brands fa-telegram"></i>
-                    <span class="convo-title">tele ~ <a href="https://t.me/g4f_official">@g4f_official</a>
-                    </span>
-                </div>
+                <button onclick="save_storage()">
+                    <i class="fa-solid fa-download"></i>
+                    <a href="" onclick="return false;">Export Conversations</a>
+                </button>
                 <div class="info">
                     <i class="fa-brands fa-github"></i>
                     <span class="convo-title">github ~ <a href="https://github.com/xtekky/gpt4free">@gpt4free</a>
@@ -161,6 +160,7 @@
                         <option value="gemini-pro">gemini-pro</option>
                         <option value="">----</option>
                     </select>
+                    <select name="model2" id="model2" class="hidden"></select>
                 </div>
                 <div class="field">
                     <select name="jailbreak" id="jailbreak" style="display: none;">
diff --git a/g4f/gui/client/css/style.css b/g4f/gui/client/static/css/style.css
index 17f3e4b3..ba3f1187 100644
--- a/g4f/gui/client/css/style.css
+++ b/g4f/gui/client/static/css/style.css
@@ -15,7 +15,7 @@
     margin: auto;
     display: flex;
     flex-direction: column;
-    gap: 16px;
+    gap: var(--inner-gap);
     max-width: 200px;
     padding: var(--section-gap);
     overflow: none;
@@ -106,6 +106,10 @@ body {
     border: 1px solid var(--blur-border);
 }
 
+.hidden {
+    display: none;
+}
+
 .conversations {
     max-width: 260px;
     padding: var(--section-gap);
@@ -179,7 +183,8 @@ body {
 .conversations {
     display: flex;
     flex-direction: column;
-    gap: 16px;
+    gap: var(--inner-gap);
+    padding: var(--inner-gap);
 }
 
 .conversations .title {
@@ -569,7 +574,7 @@ label[for="camera"] {
     height: fit-content;
     display: flex;
     align-items: center;
-    gap: 16px;
+    gap: var(--inner-gap);
 }
 
 .field .about {
@@ -653,10 +658,15 @@ select {
     font-size: 14px;
 }
 
+.bottom_buttons button a {
+    color: var(--colour-3);
+    font-weight: 500;
+}
+
 .conversations .top {
     display: flex;
     flex-direction: column;
-    gap: 16px;
+    gap: var(--inner-gap);
     overflow: auto;
 }
 
diff --git a/g4f/gui/client/img/android-chrome-192x192.png b/g4f/gui/client/static/img/android-chrome-192x192.png
index 3c32aceb..3c32aceb 100644
--- a/g4f/gui/client/img/android-chrome-192x192.png
+++ b/g4f/gui/client/static/img/android-chrome-192x192.png
diff --git a/g4f/gui/client/img/android-chrome-512x512.png b/g4f/gui/client/static/img/android-chrome-512x512.png
index ae601c93..ae601c93 100644
--- a/g4f/gui/client/img/android-chrome-512x512.png
+++ b/g4f/gui/client/static/img/android-chrome-512x512.png
diff --git a/g4f/gui/client/img/apple-touch-icon.png b/g4f/gui/client/static/img/apple-touch-icon.png
index 1143d19a..1143d19a 100644
--- a/g4f/gui/client/img/apple-touch-icon.png
+++ b/g4f/gui/client/static/img/apple-touch-icon.png
diff --git a/g4f/gui/client/img/favicon-16x16.png b/g4f/gui/client/static/img/favicon-16x16.png
index 6e934fb8..6e934fb8 100644
--- a/g4f/gui/client/img/favicon-16x16.png
+++ b/g4f/gui/client/static/img/favicon-16x16.png
diff --git a/g4f/gui/client/img/favicon-32x32.png b/g4f/gui/client/static/img/favicon-32x32.png
index efc095b5..efc095b5 100644
--- a/g4f/gui/client/img/favicon-32x32.png
+++ b/g4f/gui/client/static/img/favicon-32x32.png
diff --git a/g4f/gui/client/img/gpt.png b/g4f/gui/client/static/img/gpt.png
index 60e24da0..60e24da0 100644
--- a/g4f/gui/client/img/gpt.png
+++ b/g4f/gui/client/static/img/gpt.png
diff --git a/g4f/gui/client/img/site.webmanifest b/g4f/gui/client/static/img/site.webmanifest
index f8eab4d7..f8eab4d7 100644
--- a/g4f/gui/client/img/site.webmanifest
+++ b/g4f/gui/client/static/img/site.webmanifest
diff --git a/g4f/gui/client/img/user.png b/g4f/gui/client/static/img/user.png
index d1908e1d..d1908e1d 100644
--- a/g4f/gui/client/img/user.png
+++ b/g4f/gui/client/static/img/user.png
diff --git a/g4f/gui/client/js/chat.v1.js b/g4f/gui/client/static/js/chat.v1.js
index 8774fbf1..5440fc4a 100644
--- a/g4f/gui/client/js/chat.v1.js
+++ b/g4f/gui/client/static/js/chat.v1.js
@@ -1,5 +1,4 @@
 const colorThemes       = document.querySelectorAll('[name="theme"]');
-const markdown          = window.markdownit();
 const message_box       = document.getElementById(`messages`);
 const messageInput      = document.getElementById(`message-input`);
 const box_conversations = document.querySelector(`.top`);
@@ -12,12 +11,15 @@ const imageInput        = document.getElementById("image");
 const cameraInput       = document.getElementById("camera");
 const fileInput         = document.getElementById("file");
 const inputCount        = document.getElementById("input-count")
+const providerSelect    = document.getElementById("provider");
 const modelSelect       = document.getElementById("model");
+const modelProvider     = document.getElementById("model2");
 const systemPrompt      = document.getElementById("systemPrompt")
+const jailbreak         = document.getElementById("jailbreak");
 
 let prompt_lock = false;
 
-hljs.addPlugin(new CopyButtonPlugin());
+const options = ["switch", "model", "model2", "jailbreak", "patch", "provider", "history"];
 
 messageInput.addEventListener("blur", () => {
     window.scrollTo(0, 0);
@@ -34,15 +36,17 @@ appStorage = window.localStorage || {
     length: 0
 }
 
+const markdown = window.markdownit();
 const markdown_render = (content) => {
     return markdown.render(content
-        .replaceAll(/<!--.+-->/gm, "")
+        .replaceAll(/<!-- generated images start -->[\s\S]+<!-- generated images end -->/gm, "")
         .replaceAll(/<img data-prompt="[^>]+">/gm, "")
     )
         .replaceAll("<a href=", '<a target="_blank" href=')
         .replaceAll('<code>', '<code class="language-plaintext">')
 }
 
+hljs.addPlugin(new CopyButtonPlugin());
 let typesetPromise = Promise.resolve();
 const highlight = (container) => {
     container.querySelectorAll('code:not(.hljs').forEach((el) => {
@@ -90,48 +94,48 @@ const handle_ask = async () => {
     window.scrollTo(0, 0);
 
     message = messageInput.value
-    if (message.length > 0) {
-        messageInput.value = "";
-        prompt_lock = true;
-        count_input()
-        await add_conversation(window.conversation_id, message);
-        if ("text" in fileInput.dataset) {
-            message += '\n```' + fileInput.dataset.type + '\n'; 
-            message += fileInput.dataset.text;
-            message += '\n```'
-        }
-        let message_index = await add_message(window.conversation_id, "user", message);
-        window.token = message_id();
-
-        if (imageInput.dataset.src) URL.revokeObjectURL(imageInput.dataset.src);
-        const input = imageInput && imageInput.files.length > 0 ? imageInput : cameraInput
-        if (input.files.length > 0) imageInput.dataset.src = URL.createObjectURL(input.files[0]);
-        else delete imageInput.dataset.src
-
-        model = modelSelect.options[modelSelect.selectedIndex].value
-        message_box.innerHTML += `
-            <div class="message" data-index="${message_index}">
-                <div class="user">
-                    ${user_image}
-                    <i class="fa-solid fa-xmark"></i>
-                    <i class="fa-regular fa-phone-arrow-up-right"></i>
-                </div>
-                <div class="content" id="user_${token}"> 
-                    <div class="content_inner">
-                    ${markdown_render(message)}
-                    ${imageInput.dataset.src
-                        ? '<img src="' + imageInput.dataset.src + '" alt="Image upload">'
-                        : ''
-                    }
-                    </div>
-                    <div class="count">${count_words_and_tokens(message, model)}</div>
+    if (message.length <= 0) {
+        return;
+    }
+    messageInput.value = "";
+    prompt_lock = true;
+    count_input()
+    await add_conversation(window.conversation_id, message);
+
+    if ("text" in fileInput.dataset) {
+        message += '\n```' + fileInput.dataset.type + '\n'; 
+        message += fileInput.dataset.text;
+        message += '\n```'
+    }
+    let message_index = await add_message(window.conversation_id, "user", message);
+    window.token = message_id();
+
+    if (imageInput.dataset.src) URL.revokeObjectURL(imageInput.dataset.src);
+    const input = imageInput && imageInput.files.length > 0 ? imageInput : cameraInput
+    if (input.files.length > 0) imageInput.dataset.src = URL.createObjectURL(input.files[0]);
+    else delete imageInput.dataset.src
+
+    message_box.innerHTML += `
+        <div class="message" data-index="${message_index}">
+            <div class="user">
+                ${user_image}
+                <i class="fa-solid fa-xmark"></i>
+                <i class="fa-regular fa-phone-arrow-up-right"></i>
+            </div>
+            <div class="content" id="user_${token}"> 
+                <div class="content_inner">
+                ${markdown_render(message)}
+                ${imageInput.dataset.src
+                    ? '<img src="' + imageInput.dataset.src + '" alt="Image upload">'
+                    : ''
+                }
                 </div>
+                <div class="count">${count_words_and_tokens(message, get_selected_model())}</div>
             </div>
-        `;
-        await register_remove_message();
-        highlight(message_box);
-        await ask_gpt();
-    }
+        </div>
+    `;
+    highlight(message_box);
+    await ask_gpt();
 };
 
 const remove_cancel_button = async () => {
@@ -143,7 +147,7 @@ const remove_cancel_button = async () => {
     }, 300);
 };
 
-const prepare_messages = (messages, filter_last_message = true) => {
+const prepare_messages = (messages, filter_last_message=true) => {
     // Removes none user messages at end
     if (filter_last_message) {
         let last_message;
@@ -193,20 +197,54 @@ const prepare_messages = (messages, filter_last_message = true) => {
     return new_messages;
 }
 
+async function add_message_chunk(message) {
+    if (message.type == "conversation") {
+        console.info("Conversation used:", message.conversation)
+    } else if (message.type == "provider") {
+        window.provider_result = message.provider;
+        window.content.querySelector('.provider').innerHTML = `
+            <a href="${message.provider.url}" target="_blank">
+                ${message.provider.name}
+            </a>
+            ${message.provider.model ? ' with ' + message.provider.model : ''}
+        `
+    } else if (message.type == "message") {
+        console.error(messag.message)
+        return;
+    } else if (message.type == "error") {
+        console.error(message.error);
+        window.content_inner.innerHTML += `<p><strong>An error occured:</strong> ${message.error}</p>`;
+    } else if (message.type == "content") {
+        window.text += message.content;
+        html = markdown_render(window.text);
+        let lastElement, lastIndex = null;
+        for (element of ['</p>', '</code></pre>', '</p>\n</li>\n</ol>', '</li>\n</ol>', '</li>\n</ul>']) {
+            const index = html.lastIndexOf(element)
+            if (index - element.length > lastIndex) {
+                lastElement = element;
+                lastIndex = index;
+            }
+        }
+        if (lastIndex) {
+            html = html.substring(0, lastIndex) + '<span id="cursor"></span>' + lastElement;
+        }
+        window.content_inner.innerHTML = html;
+        window.content_count.innerText = count_words_and_tokens(text, window.provider_result?.model);
+        highlight(window.content_inner);
+    }
+
+    window.scrollTo(0, 0);
+    if (message_box.scrollTop >= message_box.scrollHeight - message_box.clientHeight - 100) {
+        message_box.scrollTo({ top: message_box.scrollHeight, behavior: "auto" });
+    }
+}
+
 const ask_gpt = async () => {
     regenerate.classList.add(`regenerate-hidden`);
     messages = await get_messages(window.conversation_id);
     total_messages = messages.length;
-
     messages = prepare_messages(messages);
 
-    window.scrollTo(0, 0);
-    window.controller = new AbortController();
-
-    jailbreak    = document.getElementById("jailbreak");
-    provider     = document.getElementById("provider");
-    window.text  = '';
-
     stop_generating.classList.remove(`stop_generating-hidden`);
 
     message_box.scrollTop = message_box.scrollHeight;
@@ -229,103 +267,31 @@ const ask_gpt = async () => {
             </div>
         </div>
     `;
-    content = document.getElementById(`gpt_${window.token}`);
-    content_inner = content.querySelector('.content_inner');
-    content_count = content.querySelector('.count');
+
+    window.controller = new AbortController();
+    window.text  = "";
+    window.error = null;
+    window.provider_result = null;
+
+    window.content = document.getElementById(`gpt_${window.token}`);
+    window.content_inner = content.querySelector('.content_inner');
+    window.content_count = content.querySelector('.count');
 
     message_box.scrollTop = message_box.scrollHeight;
     window.scrollTo(0, 0);
-
-    error = provider_result = null;
     try {
-        let body = JSON.stringify({
+        const input = imageInput && imageInput.files.length > 0 ? imageInput : cameraInput;
+        const file = input && input.files.length > 0 ? input.files[0] : null;
+        await api("conversation", {
             id: window.token,
             conversation_id: window.conversation_id,
-            model: modelSelect.options[modelSelect.selectedIndex].value,
-            jailbreak: jailbreak.options[jailbreak.selectedIndex].value,
-            web_search: document.getElementById(`switch`).checked,
-            provider: provider.options[provider.selectedIndex].value,
-            patch_provider: document.getElementById('patch')?.checked,
+            model: get_selected_model(),
+            jailbreak: jailbreak?.options[jailbreak.selectedIndex].value,
+            web_search: document.getElementById("switch").checked,
+            provider: providerSelect.options[providerSelect.selectedIndex].value,
+            patch_provider: document.getElementById("patch")?.checked,
             messages: messages
-        });
-        const headers = {
-            accept: 'text/event-stream'
-        }
-        const input = imageInput && imageInput.files.length > 0 ? imageInput : cameraInput
-        if (input && input.files.length > 0) {
-            const formData = new FormData();
-            formData.append('image', input.files[0]);
-            formData.append('json', body);
-            body = formData;
-        } else {
-            headers['content-type'] = 'application/json';
-        }
-
-        const response = await fetch(`/backend-api/v2/conversation`, {
-            method: 'POST',
-            signal: window.controller.signal,
-            headers: headers,
-            body: body
-        });
-        const reader = response.body.pipeThrough(new TextDecoderStream()).getReader();
-        let buffer = ""
-        while (true) {
-            const { value, done } = await reader.read();
-            if (done) break;
-            for (const line of value.split("\n")) {
-                if (!line) {
-                    continue;
-                }
-                let message;
-                try {
-                    message = JSON.parse(buffer + line);
-                    buffer = "";
-                } catch {
-                    buffer += line
-                    continue;
-                }
-                if (message.type == "content") {
-                    text += message.content;
-                } else if (message.type == "provider") {
-                    provider_result = message.provider
-                    content.querySelector('.provider').innerHTML = `
-                        <a href="${provider_result.url}" target="_blank">
-                            ${provider_result.name}
-                        </a>
-                        ${provider_result.model ? ' with ' + provider_result.model : ''}
-                    `
-                } else if (message.type == "error") {
-                    error = message.error;
-                } else if (messag.type == "message") {
-                    console.error(messag.message)
-                }
-            }
-            if (error) {
-                console.error(error);
-                content_inner.innerHTML += `<p><strong>An error occured:</strong> ${error}</p>`;
-            } else {
-                html = markdown_render(text);
-                let lastElement, lastIndex = null;
-                for (element of ['</p>', '</code></pre>', '</p>\n</li>\n</ol>', '</li>\n</ol>', '</li>\n</ul>']) {
-                    const index = html.lastIndexOf(element)
-                    if (index - element.length > lastIndex) {
-                        lastElement = element;
-                        lastIndex = index;
-                    }
-                }
-                if (lastIndex) {
-                    html = html.substring(0, lastIndex) + '<span id="cursor"></span>' + lastElement;
-                }
-                content_inner.innerHTML = html;
-                content_count.innerText = count_words_and_tokens(text, provider_result?.model);
-                highlight(content_inner);
-            }
-
-            window.scrollTo(0, 0);
-            if (message_box.scrollTop >= message_box.scrollHeight - message_box.clientHeight - 100) {
-                message_box.scrollTo({ top: message_box.scrollHeight, behavior: "auto" });
-            }
-        }
+        }, file);
         if (!error) {
             html = markdown_render(text);
             content_inner.innerHTML = html;
@@ -350,7 +316,7 @@ const ask_gpt = async () => {
         await add_message(window.conversation_id, "assistant", text, provider_result);
         await load_conversation(window.conversation_id);
     } else {
-        let cursorDiv = document.getElementById(`cursor`);
+        let cursorDiv = document.getElementById("cursor");
         if (cursorDiv) cursorDiv.parentNode.removeChild(cursorDiv);
     }
     window.scrollTo(0, 0);
@@ -439,7 +405,7 @@ const new_conversation = async () => {
     say_hello();
 };
 
-const load_conversation = async (conversation_id, scroll = true) => {
+const load_conversation = async (conversation_id, scroll=true) => {
     let conversation = await get_conversation(conversation_id);
     let messages = conversation?.items || [];
 
@@ -454,7 +420,6 @@ const load_conversation = async (conversation_id, scroll = true) => {
         last_model = item.provider?.model;
         let next_i = parseInt(i) + 1;
         let next_provider = item.provider ? item.provider : (messages.length > next_i ? messages[next_i].provider : null);
-
         let provider_link = item.provider?.name ? `<a href="${item.provider.url}" target="_blank">${item.provider.name}</a>` : "";
         let provider = provider_link ? `
             <div class="provider">
@@ -491,7 +456,6 @@ const load_conversation = async (conversation_id, scroll = true) => {
     }
 
     message_box.innerHTML = elements;
-
     register_remove_message();
     highlight(message_box);
 
@@ -543,7 +507,9 @@ async function add_conversation(conversation_id, content) {
 }
 
 async function save_system_message() {
-    if (!window.conversation_id) return;
+    if (!window.conversation_id) {
+        return;
+    }
     const conversation = await get_conversation(window.conversation_id);
     conversation.system = systemPrompt?.value;
     await save_conversation(window.conversation_id, conversation);
@@ -580,7 +546,6 @@ const remove_message = async (conversation_id, index) => {
 
 const add_message = async (conversation_id, role, content, provider) => {
     const conversation = await get_conversation(conversation_id);
-
     conversation.items.push({
         role: role,
         content: content,
@@ -662,15 +627,14 @@ sidebar_button.addEventListener("click", (event) => {
         sidebar.classList.add("shown");
         sidebar_button.classList.add("rotated");
     }
-
     window.scrollTo(0, 0);
 });
 
-const register_settings_localstorage = async () => {
-    for (id of ["switch", "model", "jailbreak", "patch", "provider", "history"]) {
+const register_settings_storage = async () => {
+    options.forEach((id) => {
         element = document.getElementById(id);
         if (!element) {
-            continue;
+            return;
         }
         element.addEventListener('change', async (event) => {
             switch (event.target.type) {
@@ -684,14 +648,14 @@ const register_settings_localstorage = async () => {
                     console.warn("Unresolved element type");
             }
         });
-    }
+    });
 }
 
-const load_settings_localstorage = async () => {
-    for (id of ["switch", "model", "jailbreak", "patch", "provider", "history"]) {
+const load_settings_storage = async () => {
+    options.forEach((id) => {
         element = document.getElementById(id);
-        if (!element || !(value = appStorage.getItem(element.id))) {
-            continue;
+        if (!element || !(value = appStorage.getItem(id))) {
+            return;
         }
         if (value) {
             switch (element.type) {
@@ -705,7 +669,7 @@ const load_settings_localstorage = async () => {
                     console.warn("Unresolved element type");
             }
         }
-    }
+    });
 }
 
 const say_hello = async () => {
@@ -780,13 +744,16 @@ function count_words_and_tokens(text, model) {
 }
 
 let countFocus = messageInput;
+let timeoutId;
 const count_input = async () => {
-    if (countFocus.value) {
-        model = modelSelect.options[modelSelect.selectedIndex].value;
-        inputCount.innerText = count_words_and_tokens(countFocus.value, model);
-    } else {
-        inputCount.innerHTML = "&nbsp;"
-    }
+    if (timeoutId) clearTimeout(timeoutId);
+    timeoutId = setTimeout(() => {
+        if (countFocus.value) {
+            inputCount.innerText = count_words_and_tokens(countFocus.value, get_selected_model());
+        } else {
+            inputCount.innerHTML = "&nbsp;"
+        }
+    }, 100);
 };
 messageInput.addEventListener("keyup", count_input);
 systemPrompt.addEventListener("keyup", count_input);
@@ -799,9 +766,21 @@ systemPrompt.addEventListener("blur", function() {
     count_input();
 });
 
-window.onload = async () => {
-    setTheme();
+window.addEventListener('load', async function() {
+    await on_load();
+    if (window.conversation_id == "{{chat_id}}") {
+        window.conversation_id = uuid();
+    } else {
+        await on_api();
+    }
+});
+
+window.addEventListener('pywebviewready', async function() {
+    await on_api();
+});
 
+async function on_load() {
+    setTheme();
     count_input();
 
     if (/\/chat\/.+/.test(window.location.href)) {
@@ -809,9 +788,10 @@ window.onload = async () => {
     } else {
         say_hello()
     }
-
     load_conversations();
+}
 
+async function on_api() {
     messageInput.addEventListener("keydown", async (evt) => {
         if (prompt_lock) return;
 
@@ -824,46 +804,17 @@ window.onload = async () => {
             messageInput.style.height = messageInput.scrollHeight  + "px";
         }
     });
-    
     sendButton.addEventListener(`click`, async () => {
         console.log("clicked send");
         if (prompt_lock) return;
         await handle_ask();
     });
-
     messageInput.focus();
 
-    register_settings_localstorage();
-};
-
-(async () => {
-    response = await fetch('/backend-api/v2/models')
-    models = await response.json()
-
-    for (model of models) {
-        let option = document.createElement('option');
-        option.value = option.text = model;
-        modelSelect.appendChild(option);
-    }
+    register_settings_storage();
 
-    response = await fetch('/backend-api/v2/providers')
-    providers = await response.json()
-    select = document.getElementById('provider');
-
-    for (provider of providers) {
-        let option = document.createElement('option');
-        option.value = option.text = provider;
-        select.appendChild(option);
-    }
-
-    await load_settings_localstorage()
-})();
-
-(async () => {
-    response = await fetch('/backend-api/v2/version')
-    versions = await response.json()
-    
-    document.title = 'g4f - gui - ' + versions["version"];
+    versions = await api("version");
+    document.title = 'g4f - ' + versions["version"];
     let text = "version ~ "
     if (versions["version"] != versions["latest_version"]) {
         let release_url = 'https://github.com/xtekky/gpt4free/releases/tag/' + versions["latest_version"];
@@ -873,7 +824,24 @@ window.onload = async () => {
         text += versions["version"];
     }
     document.getElementById("version_text").innerHTML = text
-})()
+
+    models = await api("models");
+    models.forEach((model) => {
+        let option = document.createElement("option");
+        option.value = option.text = model;
+        modelSelect.appendChild(option);
+    });
+
+    providers = await api("providers")
+    providers.forEach((provider) => {
+        let option = document.createElement("option");
+        option.value = option.text = provider;
+        providerSelect.appendChild(option);
+    })
+
+    await load_provider_models(appStorage.getItem("provider"));
+    load_settings_storage()
+}
 
 for (const el of [imageInput, cameraInput]) {
     el.addEventListener('click', async () => {
@@ -889,6 +857,7 @@ fileInput.addEventListener('click', async (event) => {
     fileInput.value = '';
     delete fileInput.dataset.text;
 });
+
 fileInput.addEventListener('change', async (event) => {
     if (fileInput.files.length) {
         type = fileInput.files[0].type;
@@ -903,8 +872,21 @@ fileInput.addEventListener('change', async (event) => {
         }
         fileInput.dataset.type = type
         const reader = new FileReader();
-        reader.addEventListener('load', (event) => {
+        reader.addEventListener('load', async (event) => {
             fileInput.dataset.text = event.target.result;
+            if (type == "json") {
+                const data = JSON.parse(fileInput.dataset.text);
+                if ("g4f" in data.options) {
+                    Object.keys(data).forEach(key => {
+                        if (key != "options" && !localStorage.getItem(key)) {
+                            appStorage.setItem(key, JSON.stringify(data[key]));
+                        } 
+                    });
+                    delete fileInput.dataset.text;
+                    await load_conversations();
+                    fileInput.value = "";
+                }
+            }
         });
         reader.readAsText(fileInput.files[0]);
     } else {
@@ -914,4 +896,126 @@ fileInput.addEventListener('change', async (event) => {
 
 systemPrompt?.addEventListener("blur", async () => {
     await save_system_message();
-});
\ No newline at end of file
+});
+
+function get_selected_model() {
+    if (modelProvider.selectedIndex >= 0) {
+        return modelProvider.options[modelProvider.selectedIndex].value;
+    } else if (modelSelect.selectedIndex >= 0) {
+        return modelSelect.options[modelSelect.selectedIndex].value;
+    }
+}
+
+async function api(ressource, args=null, file=null) {
+    if (window?.pywebview) {
+        if (args) {
+            if (ressource == "models") {
+                ressource = "provider_models";
+            }
+            return pywebview.api["get_" + ressource](args);
+        }
+        return pywebview.api["get_" + ressource]();
+    }
+    if (ressource == "models" && args) {
+        ressource = `${ressource}/${args}`;
+    }
+    const url = `/backend-api/v2/${ressource}`;
+    if (ressource == "conversation") {
+        const body = JSON.stringify(args);
+        const headers = {
+            accept: 'text/event-stream'
+        }
+        if (file) {
+            const formData = new FormData();
+            formData.append('file', file);
+            formData.append('json', body);
+            body = formData;
+        } else {
+            headers['content-type'] = 'application/json';
+        }
+        response = await fetch(url, {
+            method: 'POST',
+            signal: window.controller.signal,
+            headers: headers,
+            body: body
+        });
+        return read_response(response);
+    }
+    response = await fetch(url);
+    return await response.json();
+}
+
+async function read_response(response) {
+    const reader = response.body.pipeThrough(new TextDecoderStream()).getReader();
+    let buffer = ""
+    while (true) {
+        const { value, done } = await reader.read();
+        if (done) {
+            break;
+        }
+        for (const line of value.split("\n")) {
+            if (!line) {
+                continue;
+            }
+            try {
+                add_message_chunk(JSON.parse(buffer + line))
+                buffer = "";
+            } catch {
+                buffer += line
+            }
+        }
+    }
+}
+
+async function load_provider_models(providerIndex=null) {
+    if (!providerIndex) {
+        providerIndex = providerSelect.selectedIndex;
+    }
+    const provider = providerSelect.options[providerIndex].value;
+    if (!provider) {
+        return;
+    }
+    const models = await api('models', provider);
+    modelProvider.innerHTML = '';
+    if (models.length > 0) {
+        modelSelect.classList.add("hidden");
+        modelProvider.classList.remove("hidden");
+        models.forEach((model) => {
+            let option = document.createElement('option');
+            option.value = option.text = model.model;
+            option.selected = model.default;
+            modelProvider.appendChild(option);
+        });
+    } else {
+        modelProvider.classList.add("hidden");
+        modelSelect.classList.remove("hidden");
+    }
+};
+providerSelect.addEventListener("change", () => load_provider_models());
+
+function save_storage() {
+    let filename = new Date().toLocaleString()
+    filename += ".json"
+    let data = {"options": {"g4f": ""}};
+    for (let i = 0; i < appStorage.length; i++){
+        let key = appStorage.key(i);
+        let item = appStorage.getItem(key);
+        if (key.startsWith("conversation:")) {
+            data[key] = JSON.parse(item);
+        } else {
+            data["options"][key] = item;
+        }
+    }
+    data = JSON.stringify(data, null, 4);
+    const blob = new Blob([data], {type: 'text/csv'});
+    if(window.navigator.msSaveOrOpenBlob) {
+        window.navigator.msSaveBlob(blob, filename);
+    } else{
+        const elem = window.document.createElement('a');
+        elem.href = window.URL.createObjectURL(blob);
+        elem.download = filename;        
+        document.body.appendChild(elem);
+        elem.click();        
+        document.body.removeChild(elem);
+    }
+}
\ No newline at end of file
diff --git a/g4f/gui/client/js/highlight.min.js b/g4f/gui/client/static/js/highlight.min.js
index d410b45b..d410b45b 100644
--- a/g4f/gui/client/js/highlight.min.js
+++ b/g4f/gui/client/static/js/highlight.min.js
diff --git a/g4f/gui/client/js/highlightjs-copy.min.js b/g4f/gui/client/static/js/highlightjs-copy.min.js
index ac11d33e..ac11d33e 100644
--- a/g4f/gui/client/js/highlightjs-copy.min.js
+++ b/g4f/gui/client/static/js/highlightjs-copy.min.js
diff --git a/g4f/gui/client/js/icons.js b/g4f/gui/client/static/js/icons.js
index 84fed38d..84fed38d 100644
--- a/g4f/gui/client/js/icons.js
+++ b/g4f/gui/client/static/js/icons.js
diff --git a/g4f/gui/server/api.py b/g4f/gui/server/api.py
new file mode 100644
index 00000000..966319e4
--- /dev/null
+++ b/g4f/gui/server/api.py
@@ -0,0 +1,185 @@
+import logging
+import json
+from typing import Iterator
+
+try:
+    import webview
+except ImportError:
+    ...
+
+from g4f import version, models
+from g4f import get_last_provider, ChatCompletion
+from g4f.errors import VersionNotFoundError
+from g4f.Provider import ProviderType, __providers__, __map__
+from g4f.providers.base_provider import ProviderModelMixin
+from g4f.Provider.bing.create_images import patch_provider
+from g4f.Provider.Bing import Conversation
+
+conversations: dict[str, Conversation] = {}
+
+class Api():
+
+    def get_models(self) -> list[str]:
+        """
+        Return a list of all models.
+
+        Fetches and returns a list of all available models in the system.
+
+        Returns:
+            List[str]: A list of model names.
+        """
+        return models._all_models
+
+    def get_provider_models(self, provider: str) -> list[dict]:
+        if provider in __map__:
+            provider: ProviderType = __map__[provider]
+            if issubclass(provider, ProviderModelMixin):
+                return [{"model": model, "default": model == provider.default_model} for model in provider.get_models()]
+            elif provider.supports_gpt_35_turbo or provider.supports_gpt_4:
+                return [
+                    *([{"model": "gpt-4", "default": not provider.supports_gpt_4}] if provider.supports_gpt_4 else []),
+                    *([{"model": "gpt-3.5-turbo", "default": not provider.supports_gpt_4}] if provider.supports_gpt_35_turbo else [])
+                ]
+            else:
+                return [];
+
+    def get_providers(self) -> list[str]:
+        """
+        Return a list of all working providers.
+        """
+        return [provider.__name__ for provider in __providers__ if provider.working]
+
+    def get_version(self):
+        """
+        Returns the current and latest version of the application.
+
+        Returns:
+            dict: A dictionary containing the current and latest version.
+        """
+        try:
+            current_version = version.utils.current_version
+        except VersionNotFoundError:
+            current_version = None
+        return {
+            "version": current_version,
+            "latest_version": version.utils.latest_version,
+        }
+
+    def generate_title(self):
+        """
+        Generates and returns a title based on the request data.
+
+        Returns:
+            dict: A dictionary with the generated title.
+        """
+        return {'title': ''}
+
+    def get_conversation(self, options: dict, **kwargs) -> Iterator:
+        window = webview.active_window()
+        for message in self._create_response_stream(
+            self._prepare_conversation_kwargs(options, kwargs),
+            options.get("conversation_id")
+        ):
+            window.evaluate_js(f"this.add_message_chunk({json.dumps(message)})")
+
+    def _prepare_conversation_kwargs(self, json_data: dict, kwargs: dict):
+        """
+        Prepares arguments for chat completion based on the request data.
+
+        Reads the request and prepares the necessary arguments for handling 
+        a chat completion request.
+
+        Returns:
+            dict: Arguments prepared for chat completion.
+        """ 
+        provider = json_data.get('provider', None)
+        if "image" in kwargs and provider is None:
+            provider = "Bing"
+        if provider == 'OpenaiChat':
+            kwargs['auto_continue'] = True
+
+        messages = json_data['messages']
+        if json_data.get('web_search'):
+            if provider == "Bing":
+                kwargs['web_search'] = True
+            else:
+                from .internet import get_search_message
+                messages[-1]["content"] = get_search_message(messages[-1]["content"])
+
+        conversation_id = json_data.get("conversation_id")
+        if conversation_id and conversation_id in conversations:
+            kwargs["conversation"] = conversations[conversation_id]
+
+        model = json_data.get('model')
+        model = model if model else models.default
+        patch = patch_provider if json_data.get('patch_provider') else None
+
+        return {
+            "model": model,
+            "provider": provider,
+            "messages": messages,
+            "stream": True,
+            "ignore_stream": True,
+            "patch_provider": patch,
+            "return_conversation": True,
+            **kwargs
+        }
+
+    def _create_response_stream(self, kwargs, conversation_id: str) -> Iterator:
+        """
+        Creates and returns a streaming response for the conversation.
+
+        Args:
+            kwargs (dict): Arguments for creating the chat completion.
+
+        Yields:
+            str: JSON formatted response chunks for the stream.
+
+        Raises:
+            Exception: If an error occurs during the streaming process.
+        """
+        try:
+            first = True
+            for chunk in ChatCompletion.create(**kwargs):
+                if first:
+                    first = False
+                    yield self._format_json("provider", get_last_provider(True))
+                if isinstance(chunk, Conversation):
+                    conversations[conversation_id] = chunk
+                    yield self._format_json("conversation", conversation_id)
+                elif isinstance(chunk, Exception):
+                    logging.exception(chunk)
+                    yield self._format_json("message", get_error_message(chunk))
+                else:
+                    yield self._format_json("content", chunk)
+        except Exception as e:
+            logging.exception(e)
+            yield self._format_json('error', get_error_message(e))
+
+    def _format_json(self, response_type: str, content):
+        """
+        Formats and returns a JSON response.
+
+        Args:
+            response_type (str): The type of the response.
+            content: The content to be included in the response.
+
+        Returns:
+            str: A JSON formatted string.
+        """
+        return {
+            'type': response_type,
+            response_type: content
+        }
+    
+def get_error_message(exception: Exception) -> str:
+    """
+    Generates a formatted error message from an exception.
+
+    Args:
+        exception (Exception): The exception to format.
+
+    Returns:
+        str: A formatted error message string.
+    """
+    return f"{get_last_provider().__name__}: {type(exception).__name__}: {exception}"
\ No newline at end of file
diff --git a/g4f/gui/server/app.py b/g4f/gui/server/app.py
index 2b55698c..869d3880 100644
--- a/g4f/gui/server/app.py
+++ b/g4f/gui/server/app.py
@@ -1,3 +1,9 @@
+import sys, os
 from flask import Flask
 
-app = Flask(__name__, template_folder='./../client/html')
\ No newline at end of file
+if getattr(sys, 'frozen', False):
+    template_folder = os.path.join(sys._MEIPASS, "client")
+else:
+    template_folder = "../client"
+
+app = Flask(__name__, template_folder=template_folder, static_folder=f"{template_folder}/static")
\ No newline at end of file
diff --git a/g4f/gui/server/backend.py b/g4f/gui/server/backend.py
index 454ed1c6..fb8404d4 100644
--- a/g4f/gui/server/backend.py
+++ b/g4f/gui/server/backend.py
@@ -1,15 +1,9 @@
-import logging
 import json
 from flask import request, Flask
-from typing import Generator
-from g4f import version, models
-from g4f import get_last_provider, ChatCompletion
 from g4f.image import is_allowed_extension, to_image
-from g4f.errors import VersionNotFoundError
-from g4f.Provider import __providers__
-from g4f.Provider.bing.create_images import patch_provider
+from .api import Api
 
-class Backend_Api:
+class Backend_Api(Api):    
     """
     Handles various endpoints in a Flask application for backend operations.
 
@@ -33,6 +27,10 @@ class Backend_Api:
                 'function': self.get_models,
                 'methods': ['GET']
             },
+            '/backend-api/v2/models/<provider>': {
+                'function': self.get_provider_models,
+                'methods': ['GET']
+            },
             '/backend-api/v2/providers': {
                 'function': self.get_providers,
                 'methods': ['GET']
@@ -54,7 +52,7 @@ class Backend_Api:
                 'methods': ['POST']
             }
         }
-    
+
     def handle_error(self):
         """
         Initialize the backend API with the given Flask application.
@@ -64,49 +62,7 @@ class Backend_Api:
         """
         print(request.json)
         return 'ok', 200
-    
-    def get_models(self):
-        """
-        Return a list of all models.
-
-        Fetches and returns a list of all available models in the system.
-
-        Returns:
-            List[str]: A list of model names.
-        """
-        return models._all_models
-    
-    def get_providers(self):
-        """
-        Return a list of all working providers.
-        """
-        return [provider.__name__ for provider in __providers__ if provider.working]
-        
-    def get_version(self):
-        """
-        Returns the current and latest version of the application.
-
-        Returns:
-            dict: A dictionary containing the current and latest version.
-        """
-        try:
-            current_version = version.utils.current_version
-        except VersionNotFoundError:
-            current_version = None
-        return {
-            "version": current_version,
-            "latest_version": version.utils.latest_version,
-        }
-    
-    def generate_title(self):
-        """
-        Generates and returns a title based on the request data.
 
-        Returns:
-            dict: A dictionary with the generated title.
-        """
-        return {'title': ''}
-    
     def handle_conversation(self):
         """
         Handles conversation requests and streams responses back.
@@ -114,26 +70,10 @@ class Backend_Api:
         Returns:
             Response: A Flask response object for streaming.
         """
-        kwargs = self._prepare_conversation_kwargs()
-
-        return self.app.response_class(
-            self._create_response_stream(kwargs),
-            mimetype='text/event-stream'
-        )
-    
-    def _prepare_conversation_kwargs(self):
-        """
-        Prepares arguments for chat completion based on the request data.
-
-        Reads the request and prepares the necessary arguments for handling 
-        a chat completion request.
-
-        Returns:
-            dict: Arguments prepared for chat completion.
-        """
+        
         kwargs = {}
-        if "image" in request.files:
-            file = request.files['image']
+        if "file" in request.files:
+            file = request.files['file']
             if file.filename != '' and is_allowed_extension(file.filename):
                 kwargs['image'] = to_image(file.stream, file.filename.endswith('.svg'))
                 kwargs['image_name'] = file.filename
@@ -141,66 +81,20 @@ class Backend_Api:
             json_data = json.loads(request.form['json'])
         else:
             json_data = request.json
-            
-        provider = json_data.get('provider', '').replace('g4f.Provider.', '')
-        provider = provider if provider and provider != "Auto" else None
-
-        if "image" in kwargs and not provider:
-            provider = "Bing"
-        if provider == 'OpenaiChat':
-            kwargs['auto_continue'] = True
-
-        messages = json_data['messages']
-        if json_data.get('web_search'):
-            if provider == "Bing":
-                kwargs['web_search'] = True
-            else:
-                # ResourceWarning: unclosed event loop
-                from .internet import get_search_message
-                messages[-1]["content"] = get_search_message(messages[-1]["content"])
-
-        model = json_data.get('model')
-        model = model if model else models.default
-        patch = patch_provider if json_data.get('patch_provider') else None
-
-        return {
-            "model": model,
-            "provider": provider,
-            "messages": messages,
-            "stream": True,
-            "ignore_stream": True,
-            "patch_provider": patch,
-            **kwargs
-        }
 
-    def _create_response_stream(self, kwargs) -> Generator[str, None, None]:
-        """
-        Creates and returns a streaming response for the conversation.
+        kwargs = self._prepare_conversation_kwargs(json_data, kwargs)
 
-        Args:
-            kwargs (dict): Arguments for creating the chat completion.
+        return self.app.response_class(
+            self._create_response_stream(kwargs, json_data.get("conversation_id")),
+            mimetype='text/event-stream'
+        )
 
-        Yields:
-            str: JSON formatted response chunks for the stream.
+    def get_provider_models(self, provider: str):
+        models = super().get_provider_models(provider)
+        if models is None:
+            return 404, "Provider not found"
+        return models
 
-        Raises:
-            Exception: If an error occurs during the streaming process.
-        """
-        try:
-            first = True
-            for chunk in ChatCompletion.create(**kwargs):
-                if first:
-                    first = False
-                    yield self._format_json('provider', get_last_provider(True))
-                if isinstance(chunk, Exception):
-                    logging.exception(chunk)
-                    yield self._format_json('message', get_error_message(chunk))
-                else:
-                    yield self._format_json('content', str(chunk))
-        except Exception as e:
-            logging.exception(e)
-            yield self._format_json('error', get_error_message(e))
-            
     def _format_json(self, response_type: str, content) -> str:
         """
         Formats and returns a JSON response.
@@ -212,19 +106,4 @@ class Backend_Api:
         Returns:
             str: A JSON formatted string.
         """
-        return json.dumps({
-            'type': response_type,
-            response_type: content
-        }) + "\n"
-    
-def get_error_message(exception: Exception) -> str:
-    """
-    Generates a formatted error message from an exception.
-
-    Args:
-        exception (Exception): The exception to format.
-
-    Returns:
-        str: A formatted error message string.
-    """
-    return f"{get_last_provider().__name__}: {type(exception).__name__}: {exception}"
\ No newline at end of file
+        return json.dumps(super()._format_json(response_type, content)) + "\n"
\ No newline at end of file
diff --git a/g4f/gui/server/website.py b/g4f/gui/server/website.py
index 2705664d..4e611177 100644
--- a/g4f/gui/server/website.py
+++ b/g4f/gui/server/website.py
@@ -1,6 +1,5 @@
-from flask import render_template, send_file, redirect
-from time import time
-from os import urandom
+import uuid
+from flask import render_template, redirect
 
 class Website:
     def __init__(self, app) -> None:
@@ -18,23 +17,12 @@ class Website:
                 'function': self._chat,
                 'methods': ['GET', 'POST']
             },
-            '/assets/<folder>/<file>': {
-                'function': self._assets,
-                'methods': ['GET', 'POST']
-            }
         }
 
     def _chat(self, conversation_id):
         if '-' not in conversation_id:
             return redirect('/chat')
-
-        return render_template('index.html', chat_id = conversation_id)
+        return render_template('index.html', chat_id=conversation_id)
 
     def _index(self):
-        return render_template('index.html', chat_id = f'{urandom(4).hex()}-{urandom(2).hex()}-{urandom(2).hex()}-{urandom(2).hex()}-{hex(int(time() * 1000))[2:]}')
-
-    def _assets(self, folder: str, file: str):
-        try:
-            return send_file(f"./../client/{folder}/{file}", as_attachment=False)
-        except:
-            return "File not found", 404
\ No newline at end of file
+        return render_template('index.html', chat_id=str(uuid.uuid4()))
\ No newline at end of file
diff --git a/g4f/gui/webview.py b/g4f/gui/webview.py
index 5a4263dc..a8e745f3 100644
--- a/g4f/gui/webview.py
+++ b/g4f/gui/webview.py
@@ -1,24 +1,37 @@
 import webview
-from functools import partial
-from platformdirs import user_config_dir
+try:
+    from platformdirs import user_config_dir
+    has_platformdirs = True
+except ImportError:
+    has_platformdirs = False
 
-from g4f.gui import run_gui
 from g4f.gui.run import gui_parser
+from g4f.gui.server.api import Api
 import g4f.version
 import g4f.debug
 
-def run_webview(host: str = "0.0.0.0", port: int = 8080, debug: bool = True):
-    webview.create_window(f"g4f - {g4f.version.utils.current_version}", f"http://{host}:{port}/")
-    if debug:
-        g4f.debug.logging = True
+def run_webview(
+    debug: bool = False,
+    storage_path: str = None
+):
+    webview.create_window(
+        f"g4f - {g4f.version.utils.current_version}",
+        "client/index.html",
+        text_select=True,
+        js_api=Api(),
+    )
+    if has_platformdirs and storage_path is None:
+        storage_path = user_config_dir("g4f-webview")
     webview.start(
-        partial(run_gui, host, port),
         private_mode=False,
-        storage_path=user_config_dir("g4f-webview"),
-        debug=debug
+        storage_path=storage_path,
+        debug=debug,
+        ssl=True
     )
 
 if __name__ == "__main__":
     parser = gui_parser()
     args = parser.parse_args()
-    run_webview(args.host, args.port, args.debug)
\ No newline at end of file
+    if args.debug:
+        g4f.debug.logging = True
+    run_webview(args.debug)
\ No newline at end of file
diff --git a/g4f/gui/webview.spec b/g4f/gui/webview.spec
new file mode 100644
index 00000000..360e264e
--- /dev/null
+++ b/g4f/gui/webview.spec
@@ -0,0 +1,45 @@
+# -*- mode: python ; coding: utf-8 -*-
+
+
+block_cipher = None
+
+
+a = Analysis(
+    ['webview.py'],
+    pathex=[],
+    binaries=[],
+    datas=[],
+    hiddenimports=[],
+    hookspath=[],
+    hooksconfig={},
+    runtime_hooks=[],
+    excludes=[],
+    win_no_prefer_redirects=False,
+    win_private_assemblies=False,
+    cipher=block_cipher,
+    noarchive=False,
+)
+pyz = PYZ(a.pure, a.zipped_data, cipher=block_cipher)
+
+exe = EXE(
+    pyz,
+    a.scripts,
+    a.binaries,
+    Tree('client', prefix='client'),
+    a.zipfiles,
+    a.datas,
+    [],
+    name='webview',
+    debug=False,
+    bootloader_ignore_signals=False,
+    strip=False,
+    upx=True,
+    upx_exclude=[],
+    runtime_tmpdir=None,
+    console=False,
+    disable_windowed_traceback=False,
+    argv_emulation=False,
+    target_arch=None,
+    codesign_identity=None,
+    entitlements_file=None,
+)
diff --git a/g4f/providers/base_provider.py b/g4f/providers/base_provider.py
index 32a0c01b..c8397193 100644
--- a/g4f/providers/base_provider.py
+++ b/g4f/providers/base_provider.py
@@ -70,7 +70,14 @@ class AbstractProvider(BaseProvider):
             loop.run_in_executor(executor, create_func),
             timeout=kwargs.get("timeout")
         )
-    
+
+    def get_parameters(cls) -> dict:
+        return signature(
+            cls.create_async_generator if issubclass(cls, AsyncGeneratorProvider) else
+            cls.create_async if issubclass(cls, AsyncProvider) else
+            cls.create_completion
+        ).parameters
+
     @classmethod
     @property
     def params(cls) -> str:
@@ -83,17 +90,12 @@ class AbstractProvider(BaseProvider):
         Returns:
             str: A string listing the supported parameters.
         """
-        sig = signature(
-            cls.create_async_generator if issubclass(cls, AsyncGeneratorProvider) else
-            cls.create_async if issubclass(cls, AsyncProvider) else
-            cls.create_completion
-        )
 
         def get_type_name(annotation: type) -> str:
             return annotation.__name__ if hasattr(annotation, "__name__") else str(annotation)
 
         args = ""
-        for name, param in sig.parameters.items():
+        for name, param in cls.get_parameters():
             if name in ("self", "kwargs") or (name == "stream" and not cls.supports_stream):
                 continue
             args += f"\n    {name}"
diff --git a/g4f/requests/__init__.py b/g4f/requests/__init__.py
index f2946fc1..e65de99a 100644
--- a/g4f/requests/__init__.py
+++ b/g4f/requests/__init__.py
@@ -1,22 +1,48 @@
 from __future__ import annotations
 
-from typing import Union
-from aiohttp import ClientResponse
-from requests import Response as RequestsResponse
-
 try:
     from curl_cffi.requests import Session, Response
-    from .curl_cffi import StreamResponse, StreamSession
+    from .curl_cffi import StreamResponse, StreamSession, FormData
     has_curl_cffi = True
 except ImportError:
     from typing import Type as Session, Type as Response
-    from .aiohttp import StreamResponse, StreamSession
+    from .aiohttp import StreamResponse, StreamSession, FormData
     has_curl_cffi = False
+try:
+    import webview
+    import asyncio
+    has_webview = True
+except ImportError:
+    has_webview = False
 
+from .raise_for_status import raise_for_status
 from ..webdriver import WebDriver, WebDriverSession
 from ..webdriver import bypass_cloudflare, get_driver_cookies
-from ..errors import MissingRequirementsError, RateLimitError, ResponseStatusError
-from .defaults import DEFAULT_HEADERS
+from ..errors import MissingRequirementsError
+from .defaults import DEFAULT_HEADERS, WEBVIEW_HAEDERS
+
+async def get_args_from_webview(url: str) -> dict:
+    if not has_webview:
+        raise MissingRequirementsError('Install "webview" package')
+    window = webview.create_window("", url, hidden=True)
+    await asyncio.sleep(2)
+    body = None
+    while body is None:
+        try:
+            await asyncio.sleep(1)
+            body = window.dom.get_element("body:not(.no-js)")
+        except:
+            ...
+    headers = {
+        **WEBVIEW_HAEDERS,
+        "User-Agent": window.evaluate_js("this.navigator.userAgent"),
+        "Accept-Language": window.evaluate_js("this.navigator.language"),
+        "Referer": window.real_url
+    }
+    cookies = [list(*cookie.items()) for cookie in window.get_cookies()]
+    cookies = dict([(name, cookie.value) for name, cookie in cookies])
+    window.destroy()
+    return {"headers": headers, "cookies": cookies}
 
 def get_args_from_browser(
     url: str,
@@ -79,24 +105,4 @@ def get_session_from_browser(url: str, webdriver: WebDriver = None, proxy: str =
         proxies={"https": proxy, "http": proxy},
         timeout=timeout,
         impersonate="chrome"
-    )
-
-async def raise_for_status_async(response: Union[StreamResponse, ClientResponse], message: str = None):
-    if response.status in (429, 402):
-        raise RateLimitError(f"Response {response.status}: Rate limit reached")
-    message = await response.text() if not response.ok and message is None else message
-    if response.status == 403 and "<title>Just a moment...</title>" in message:
-        raise ResponseStatusError(f"Response {response.status}: Cloudflare detected")
-    elif not response.ok:
-        raise ResponseStatusError(f"Response {response.status}: {message}")
-
-def raise_for_status(response: Union[StreamResponse, ClientResponse, Response, RequestsResponse], message: str = None):
-    if isinstance(response, StreamSession) or isinstance(response, ClientResponse):
-        return raise_for_status_async(response, message)
-
-    if response.status_code in (429, 402):
-        raise RateLimitError(f"Response {response.status_code}: Rate limit reached")
-    elif response.status_code == 403 and "<title>Just a moment...</title>" in response.text:
-        raise ResponseStatusError(f"Response {response.status_code}: Cloudflare detected")
-    elif not response.ok:
-        raise ResponseStatusError(f"Response {response.status_code}: {response.text if message is None else message}")
\ No newline at end of file
+    )
\ No newline at end of file
diff --git a/g4f/requests/aiohttp.py b/g4f/requests/aiohttp.py
index 6979b20a..16b052eb 100644
--- a/g4f/requests/aiohttp.py
+++ b/g4f/requests/aiohttp.py
@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from aiohttp import ClientSession, ClientResponse, ClientTimeout, BaseConnector
+from aiohttp import ClientSession, ClientResponse, ClientTimeout, BaseConnector, FormData
 from typing import AsyncIterator, Any, Optional
 
 from .defaults import DEFAULT_HEADERS
diff --git a/g4f/requests/curl_cffi.py b/g4f/requests/curl_cffi.py
index cfcdd63b..91142365 100644
--- a/g4f/requests/curl_cffi.py
+++ b/g4f/requests/curl_cffi.py
@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from curl_cffi.requests import AsyncSession, Response
+from curl_cffi.requests import AsyncSession, Response, CurlMime
 from typing import AsyncGenerator, Any
 from functools import partialmethod
 import json
@@ -65,6 +65,8 @@ class StreamSession(AsyncSession):
     def request(
         self, method: str, url: str, **kwargs
     ) -> StreamResponse:
+        if isinstance(kwargs.get("data"), CurlMime):
+            kwargs["multipart"] = kwargs.pop("data")
         """Create and return a StreamResponse object for the given HTTP request."""
         return StreamResponse(super().request(method, url, stream=True, **kwargs))
 
@@ -75,3 +77,7 @@ class StreamSession(AsyncSession):
     put = partialmethod(request, "PUT")
     patch = partialmethod(request, "PATCH")
     delete = partialmethod(request, "DELETE")
+
+class FormData(CurlMime):
+    def add_field(self, name, data=None, content_type: str = None, filename: str = None) -> None:
+        self.addpart(name, content_type=content_type, filename=filename, data=data)
\ No newline at end of file
diff --git a/g4f/requests/defaults.py b/g4f/requests/defaults.py
index 2457f046..3183eb5a 100644
--- a/g4f/requests/defaults.py
+++ b/g4f/requests/defaults.py
@@ -16,4 +16,14 @@ DEFAULT_HEADERS = {
     "referer": "",
     "accept-encoding": "gzip, deflate, br",
     "accept-language": "en-US",
+}
+WEBVIEW_HAEDERS = {
+    "Accept": "*/*",
+    "Accept-Encoding": "gzip, deflate, br",
+    "Accept-Language": "",
+    "Referer": "",
+    "Sec-Fetch-Dest": "empty",
+    "Sec-Fetch-Mode": "cors",
+    "Sec-Fetch-Site": "same-origin",
+    "User-Agent": "",
 }
\ No newline at end of file
diff --git a/g4f/requests/raise_for_status.py b/g4f/requests/raise_for_status.py
new file mode 100644
index 00000000..9e8e141c
--- /dev/null
+++ b/g4f/requests/raise_for_status.py
@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+from typing import Union
+from aiohttp import ClientResponse
+from requests import Response as RequestsResponse
+
+from ..errors import ResponseStatusError, RateLimitError
+from . import Response, StreamResponse
+
+class CloudflareError(ResponseStatusError):
+    ...
+
+def is_cloudflare(text: str) -> bool:
+    return '<div id="cf-please-wait">' in text or "<title>Just a moment...</title>" in text
+
+async def raise_for_status_async(response: Union[StreamResponse, ClientResponse], message: str = None):
+    if response.status in (429, 402):
+        raise RateLimitError(f"Response {response.status}: Rate limit reached")
+    message = await response.text() if not response.ok and message is None else message
+    if response.status == 403 and is_cloudflare(message):
+        raise CloudflareError(f"Response {response.status}: Cloudflare detected")
+    elif not response.ok:
+        raise ResponseStatusError(f"Response {response.status}: {message}")
+
+def raise_for_status(response: Union[Response, StreamResponse, ClientResponse, RequestsResponse], message: str = None):
+    if hasattr(response, "status"):
+        return raise_for_status_async(response, message)
+
+    if response.status_code in (429, 402):
+        raise RateLimitError(f"Response {response.status_code}: Rate limit reached")
+    elif response.status_code == 403 and is_cloudflare(response.text):
+        raise CloudflareError(f"Response {response.status_code}: Cloudflare detected")
+    elif not response.ok:
+        raise ResponseStatusError(f"Response {response.status_code}: {response.text if message is None else message}")
\ No newline at end of file
diff --git a/g4f/version.py b/g4f/version.py
index 0d85a7f5..eda2b8fe 100644
--- a/g4f/version.py
+++ b/g4f/version.py
@@ -6,6 +6,7 @@ from functools import cached_property
 from importlib.metadata import version as get_package_version, PackageNotFoundError
 from subprocess import check_output, CalledProcessError, PIPE
 from .errors import VersionNotFoundError
+from . import debug
 
 PACKAGE_NAME = "g4f"
 GITHUB_REPOSITORY = "xtekky/gpt4free"
@@ -64,6 +65,9 @@ class VersionUtils:
             VersionNotFoundError: If the version cannot be determined from the package manager, 
                                   Docker environment, or git repository.
         """
+        if debug.version:
+            return debug.version
+
         # Read from package manager
         try:
             return get_package_version(PACKAGE_NAME)