webscout 8.2.7__py3-none-any.whl → 8.2.8__py3-none-any.whl

webscout/Provider/Groq.py

@@ -0,0 +1,801 @@
+from typing import Any, AsyncGenerator, Dict, Optional, Callable, List, Union
+
+import httpx
+import json
+
+# Import curl_cffi for improved request handling
+from curl_cffi.requests import Session
+from curl_cffi import CurlError
+
+from webscout.AIutel import Optimizers
+from webscout.AIutel import Conversation
+from webscout.AIutel import AwesomePrompts, sanitize_stream # Import sanitize_stream
+from webscout.AIbase import Provider, AsyncProvider
+from webscout import exceptions
+
+class GROQ(Provider):
+    """
+    A class to interact with the GROQ AI API.
+    """
+
+    # Default models list (will be updated dynamically)
+    AVAILABLE_MODELS = [
+        "distil-whisper-large-v3-en",
+        "gemma2-9b-it",
+        "llama-3.3-70b-versatile",
+        "llama-3.1-8b-instant",
+        "llama-guard-3-8b",
+        "llama3-70b-8192",
+        "llama3-8b-8192",
+        "whisper-large-v3",
+        "whisper-large-v3-turbo",
+        "meta-llama/llama-4-scout-17b-16e-instruct",
+        "meta-llama/llama-4-maverick-17b-128e-instruct",
+        "playai-tts",
+        "playai-tts-arabic",
+        "qwen-qwq-32b",
+        "mistral-saba-24b",
+        "qwen-2.5-coder-32b",
+        "qwen-2.5-32b",
+        "deepseek-r1-distill-qwen-32b",
+        "deepseek-r1-distill-llama-70b",
+        "llama-3.3-70b-specdec",
+        "llama-3.2-1b-preview",
+        "llama-3.2-3b-preview",
+        "llama-3.2-11b-vision-preview",
+        "llama-3.2-90b-vision-preview",
+        "mixtral-8x7b-32768"
+    ]
+    
+    @classmethod
+    def get_models(cls, api_key: str = None):
+        """Fetch available models from Groq API.
+        
+        Args:
+            api_key (str, optional): Groq API key. If not provided, returns default models.
+            
+        Returns:
+            list: List of available model IDs
+        """
+        if not api_key:
+            return cls.AVAILABLE_MODELS
+            
+        try:
+            # Use a temporary curl_cffi session for this class method
+            temp_session = Session()
+            headers = {
+                "Content-Type": "application/json",
+                "Authorization": f"Bearer {api_key}",
+            }
+            
+            response = temp_session.get(
+                "https://api.groq.com/openai/v1/models",
+                headers=headers,
+                impersonate="chrome110"  # Use impersonate for fetching
+            )
+            
+            if response.status_code != 200:
+                return cls.AVAILABLE_MODELS
+                
+            data = response.json()
+            if "data" in data and isinstance(data["data"], list):
+                return [model["id"] for model in data["data"]]
+            return cls.AVAILABLE_MODELS
+            
+        except (CurlError, Exception):
+            # Fallback to default models list if fetching fails
+            return cls.AVAILABLE_MODELS
+
+    def __init__(
+        self,
+        api_key: str,
+        is_conversation: bool = True,
+        max_tokens: int = 600,
+        temperature: float = 1,
+        presence_penalty: int = 0,
+        frequency_penalty: int = 0,
+        top_p: float = 1,
+        model: str = "mixtral-8x7b-32768",
+        timeout: int = 30,
+        intro: str = None,
+        filepath: str = None,
+        update_file: bool = True,
+        proxies: dict = {},
+        history_offset: int = 10250,
+        act: str = None,
+        system_prompt: Optional[str] = None,
+    ):
+        """Instantiates GROQ
+
+        Args:
+            api_key (key): GROQ's API key.
+            is_conversation (bool, optional): Flag for chatting conversationally. Defaults to True.
+            max_tokens (int, optional): Maximum number of tokens to be generated upon completion. Defaults to 600.
+            temperature (float, optional): Charge of the generated text's randomness. Defaults to 1.
+            presence_penalty (int, optional): Chances of topic being repeated. Defaults to 0.
+            frequency_penalty (int, optional): Chances of word being repeated. Defaults to 0.
+            top_p (float, optional): Sampling threshold during inference time. Defaults to 0.999.
+            model (str, optional): LLM model name. Defaults to "mixtral-8x7b-32768".
+            timeout (int, optional): Http request timeout. Defaults to 30.
+            intro (str, optional): Conversation introductory prompt. Defaults to None.
+            filepath (str, optional): Path to file containing conversation history. Defaults to None.
+            update_file (bool, optional): Add new prompts and responses to the file. Defaults to True.
+            proxies (dict, optional): Http request proxies. Defaults to {}.
+            history_offset (int, optional): Limit conversation history to this number of last texts. Defaults to 10250.
+            act (str|int, optional): Awesome prompt key or index. (Used as intro). Defaults to None.
+            system_prompt (str, optional): System prompt to guide the conversation. Defaults to None.
+        """
+        # Update available models from API
+        self.update_available_models(api_key)
+        
+        # Validate model after updating available models
+        if model not in self.AVAILABLE_MODELS:
+            raise ValueError(f"Invalid model: {model}. Choose from: {self.AVAILABLE_MODELS}")
+
+        # Initialize curl_cffi Session
+        self.session = Session()
+        self.is_conversation = is_conversation
+        self.max_tokens_to_sample = max_tokens
+        self.api_key = api_key
+        self.model = model
+        self.temperature = temperature
+        self.presence_penalty = presence_penalty
+        self.frequency_penalty = frequency_penalty
+        self.top_p = top_p
+        self.chat_endpoint = "https://api.groq.com/openai/v1/chat/completions" 
+        self.stream_chunk_size = 64
+        self.timeout = timeout
+        self.last_response = {}
+        self.system_prompt = system_prompt
+        self.available_functions: Dict[str, Callable] = {}  # Store available functions
+        self.headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.api_key}",
+        }
+
+        self.__available_optimizers = (
+            method
+            for method in dir(Optimizers)
+            if callable(getattr(Optimizers, method)) and not method.startswith("__")
+        )
+        
+        # Update curl_cffi session headers
+        self.session.headers.update(self.headers)
+        
+        # Set up conversation
+        Conversation.intro = (
+            AwesomePrompts().get_act(
+                act, raise_not_found=True, default=None, case_insensitive=True
+            )
+            if act
+            else intro or Conversation.intro
+        )
+        self.conversation = Conversation(
+            is_conversation, self.max_tokens_to_sample, filepath, update_file
+        )
+        self.conversation.history_offset = history_offset
+        
+        # Set proxies for curl_cffi session
+        self.session.proxies = proxies
+    
+    @staticmethod
+    def _groq_extractor(chunk: Union[str, Dict[str, Any]]) -> Optional[Dict]:
+        """Extracts the 'delta' object from Groq stream JSON chunks."""
+        if isinstance(chunk, dict):
+            # Return the delta object itself, or None if not found
+            return chunk.get("choices", [{}])[0].get("delta")
+        return None
+
+    @classmethod
+    def update_available_models(cls, api_key=None):
+        """Update the available models list from Groq API"""
+        try:
+            models = cls.get_models(api_key)
+            if models and len(models) > 0:
+                cls.AVAILABLE_MODELS = models
+        except Exception:
+            # Fallback to default models list if fetching fails
+            pass
+
+    def add_function(self, function_name: str, function: Callable):
+        """Add a function to the available functions dictionary.
+
+        Args:
+            function_name (str): The name of the function to be used in the prompt.
+            function (Callable): The function itself.
+        """
+        self.available_functions[function_name] = function
+
+    def ask(
+        self,
+        prompt: str,
+        stream: bool = False,
+        raw: bool = False,
+        optimizer: str = None,
+        conversationally: bool = False,
+        tools: Optional[List[Dict[str, Any]]] = None,  # Add tools parameter
+    ) -> dict:
+        """Chat with AI
+
+        Args:
+            prompt (str): Prompt to be send.
+            stream (bool, optional): Flag for streaming response. Defaults to False.
+            raw (bool, optional): Stream back raw response as received. Defaults to False.
+            optimizer (str, optional): Prompt optimizer name - `[code, shell_command]`. Defaults to None.
+            conversationally (bool, optional): Chat conversationally when using optimizer. Defaults to False.
+            tools (List[Dict[str, Any]], optional): List of tool definitions. See example in class docstring. Defaults to None.
+
+        Returns:
+           dict : {}
+        """
+        conversation_prompt = self.conversation.gen_complete_prompt(prompt)
+        if optimizer:
+            if optimizer in self.__available_optimizers:
+                conversation_prompt = getattr(Optimizers, optimizer)(
+                    conversation_prompt if conversationally else prompt
+                )
+            else:
+                raise Exception(
+                    f"Optimizer is not one of {self.__available_optimizers}"
+                )
+
+        messages = [{"content": conversation_prompt, "role": "user"}]
+        if self.system_prompt:
+            messages.insert(0, {"role": "system", "content": self.system_prompt})
+
+        self.session.headers.update(self.headers)
+        payload = {
+            "frequency_penalty": self.frequency_penalty,
+            "messages": messages,
+            "model": self.model,
+            "presence_penalty": self.presence_penalty,
+            "stream": stream,
+            "temperature": self.temperature,
+            "top_p": self.top_p,
+            "tools": tools  # Include tools in the payload
+        }
+
+        def for_stream():
+            try:
+                response = self.session.post(
+                    self.chat_endpoint, 
+                    json=payload, 
+                    stream=True, 
+                    timeout=self.timeout,
+                    impersonate="chrome110"  # Use impersonate for better compatibility
+                )
+                if not response.status_code == 200:
+                    raise exceptions.FailedToGenerateResponseError(
+                        # Removed response.reason_phrase
+                        f"Failed to generate response - ({response.status_code}) - {response.text}"
+                    )
+
+                streaming_text = ""
+                # Use sanitize_stream
+                processed_stream = sanitize_stream(
+                    data=response.iter_content(chunk_size=None), # Pass byte iterator
+                    intro_value="data:",
+                    to_json=True,     # Stream sends JSON
+                    content_extractor=self._groq_extractor, # Use the delta extractor
+                    yield_raw_on_error=False # Skip non-JSON lines or lines where extractor fails
+                )
+
+                for delta in processed_stream:
+                    # delta is the extracted 'delta' object or None
+                    if delta and isinstance(delta, dict):
+                        content = delta.get("content")
+                        if content:
+                            streaming_text += content
+                            resp = {"text": content} # Yield only the new chunk text
+                            self.last_response = {"choices": [{"delta": {"content": streaming_text}}]} # Update last_response structure
+                            yield resp if not raw else content # Yield dict or raw string chunk
+                        # Note: Tool calls in streaming delta are less common in OpenAI format, usually in final message
+
+            except CurlError as e:
+                raise exceptions.FailedToGenerateResponseError(f"CurlError: {str(e)}")
+            except Exception as e:
+                raise exceptions.FailedToGenerateResponseError(f"Error: {str(e)}")
+
+            # Handle tool calls if any
+            if 'tool_calls' in self.last_response.get('choices', [{}])[0].get('message', {}):
+                tool_calls = self.last_response['choices'][0]['message']['tool_calls']
+                for tool_call in tool_calls:
+                    function_name = tool_call.get('function', {}).get('name')
+                    arguments = json.loads(tool_call.get('function', {}).get('arguments', "{}"))
+                    if function_name in self.available_functions:
+                        tool_response = self.available_functions[function_name](**arguments)
+                        messages.append({
+                            "tool_call_id": tool_call['id'],
+                            "role": "tool",
+                            "name": function_name,
+                            "content": tool_response
+                        })
+                        payload['messages'] = messages
+                        # Make a second call to get the final response
+                        try:
+                            second_response = self.session.post(
+                                self.chat_endpoint, 
+                                json=payload, 
+                                timeout=self.timeout,
+                                impersonate="chrome110"  # Use impersonate for better compatibility
+                            )
+                            if second_response.status_code == 200:
+                                self.last_response = second_response.json()
+                            else:
+                                raise exceptions.FailedToGenerateResponseError(
+                                    f"Failed to execute tool - {second_response.text}"
+                                )
+                        except CurlError as e:
+                            raise exceptions.FailedToGenerateResponseError(f"CurlError during tool execution: {str(e)}")
+                        except Exception as e:
+                            raise exceptions.FailedToGenerateResponseError(f"Error during tool execution: {str(e)}")
+
+            self.conversation.update_chat_history(
+                prompt, self.get_message(self.last_response)
+            )
+
+        def for_non_stream():
+            try:
+                response = self.session.post(
+                    self.chat_endpoint, 
+                    json=payload, 
+                    stream=False, 
+                    timeout=self.timeout,
+                    impersonate="chrome110"  # Use impersonate for better compatibility
+                )
+                if (
+                    not response.status_code == 200
+                ):
+                    raise exceptions.FailedToGenerateResponseError(
+                         # Removed response.reason_phrase
+                        f"Failed to generate response - ({response.status_code}) - {response.text}"
+                    )
+                
+                response_text = response.text # Get raw text
+
+                # Use sanitize_stream to parse the non-streaming JSON response
+                processed_stream = sanitize_stream(
+                    data=response_text,
+                    to_json=True, # Parse the whole text as JSON
+                    intro_value=None,
+                    # Extractor for non-stream structure (returns the whole parsed dict)
+                    content_extractor=lambda chunk: chunk if isinstance(chunk, dict) else None,
+                    yield_raw_on_error=False
+                )
+                
+                # Extract the single result (the parsed JSON dictionary)
+                resp = next(processed_stream, None)
+                if resp is None:
+                    raise exceptions.FailedToGenerateResponseError("Failed to parse non-stream JSON response")
+
+            except CurlError as e:
+                raise exceptions.FailedToGenerateResponseError(f"CurlError: {str(e)}")
+            except Exception as e:
+                # Catch the original AttributeError here if it happens before the raise
+                if isinstance(e, AttributeError) and 'reason_phrase' in str(e):
+                     raise exceptions.FailedToGenerateResponseError(
+                        f"Failed to generate response - ({response.status_code}) - {response.text}"
+                    )
+                raise exceptions.FailedToGenerateResponseError(f"Error: {str(e)}")
+
+            # Handle tool calls if any
+            if 'tool_calls' in resp.get('choices', [{}])[0].get('message', {}):
+                tool_calls = resp['choices'][0]['message']['tool_calls']
+                for tool_call in tool_calls:
+                    function_name = tool_call.get('function', {}).get('name')
+                    arguments = json.loads(tool_call.get('function', {}).get('arguments', "{}"))
+                    if function_name in self.available_functions:
+                        tool_response = self.available_functions[function_name](**arguments)
+                        messages.append({
+                            "tool_call_id": tool_call['id'],
+                            "role": "tool",
+                            "name": function_name,
+                            "content": tool_response
+                        })
+                        payload['messages'] = messages
+                        # Make a second call to get the final response
+                        try:
+                            second_response = self.session.post(
+                                self.chat_endpoint, 
+                                json=payload, 
+                                timeout=self.timeout,
+                                impersonate="chrome110"  # Use impersonate for better compatibility
+                            )
+                            if second_response.status_code == 200:
+                                resp = second_response.json()
+                            else:
+                                raise exceptions.FailedToGenerateResponseError(
+                                    f"Failed to execute tool - {second_response.text}"
+                                )
+                        except CurlError as e:
+                            raise exceptions.FailedToGenerateResponseError(f"CurlError during tool execution: {str(e)}")
+                        except Exception as e:
+                            raise exceptions.FailedToGenerateResponseError(f"Error during tool execution: {str(e)}")
+
+            self.last_response.update(resp)
+            self.conversation.update_chat_history(
+                prompt, self.get_message(self.last_response)
+            )
+            return resp
+
+        return for_stream() if stream else for_non_stream()
+
+    def chat(
+        self,
+        prompt: str,
+        stream: bool = False,
+        optimizer: str = None,
+        conversationally: bool = False,
+        tools: Optional[List[Dict[str, Any]]] = None,
+    ) -> str:
+        """Generate response `str`
+        Args:
+            prompt (str): Prompt to be send.
+            stream (bool, optional): Flag for streaming response. Defaults to False.
+            optimizer (str, optional): Prompt optimizer name - `[code, shell_command]`. Defaults to None.
+            conversationally (bool, optional): Chat conversationally when using optimizer. Defaults to False.
+            tools (List[Dict[str, Any]], optional): List of tool definitions. See example in class docstring. Defaults to None.
+        Returns:
+            str: Response generated
+        """
+
+        def for_stream():
+            for response in self.ask(
+                prompt, True, optimizer=optimizer, conversationally=conversationally, tools=tools
+            ):
+                yield self.get_message(response)
+
+        def for_non_stream():
+            return self.get_message(
+                self.ask(
+                    prompt,
+                    False,
+                    optimizer=optimizer,
+                    conversationally=conversationally,
+                    tools=tools
+                )
+            )
+
+        return for_stream() if stream else for_non_stream()
+
+    def get_message(self, response: dict) -> str:
+        """Retrieves message only from response
+
+        Args:
+            response (dict): Response generated by `self.ask`
+
+        Returns:
+            str: Message extracted
+        """
+        assert isinstance(response, dict), "Response should be of dict data-type only"
+        try:
+            # Check delta first for streaming
+            if response.get("choices") and response["choices"][0].get("delta") and response["choices"][0]["delta"].get("content"):
+                return response["choices"][0]["delta"]["content"]
+            # Check message content for non-streaming or final message
+            if response.get("choices") and response["choices"][0].get("message") and response["choices"][0]["message"].get("content"):
+                return response["choices"][0]["message"]["content"]
+        except (KeyError, IndexError, TypeError):
+            # Handle cases where the structure might be different or content is null/missing
+            pass
+        return "" # Return empty string if no content found
+
+
+class AsyncGROQ(AsyncProvider):
+    """
+    An asynchronous class to interact with the GROQ AI API.
+    """
+
+    # Use the same model list as the synchronous class
+    AVAILABLE_MODELS = GROQ.AVAILABLE_MODELS
+
+    def __init__(
+        self,
+        api_key: str,
+        is_conversation: bool = True,
+        max_tokens: int = 600,
+        temperature: float = 1,
+        presence_penalty: int = 0,
+        frequency_penalty: int = 0,
+        top_p: float = 1,
+        model: str = "mixtral-8x7b-32768",
+        timeout: int = 30,
+        intro: str = None,
+        filepath: str = None,
+        update_file: bool = True,
+        proxies: dict = {},
+        history_offset: int = 10250,
+        act: str = None,
+        system_prompt: Optional[str] = None,
+    ):
+        """Instantiates AsyncGROQ
+
+        Args:
+            api_key (key): GROQ's API key.
+            is_conversation (bool, optional): Flag for chatting conversationally. Defaults to True.
+            max_tokens (int, optional): Maximum number of tokens to be generated upon completion. Defaults to 600.
+            temperature (float, optional): Charge of the generated text's randomness. Defaults to 1.
+            presence_penalty (int, optional): Chances of topic being repeated. Defaults to 0.
+            frequency_penalty (int, optional): Chances of word being repeated. Defaults to 0.
+            top_p (float, optional): Sampling threshold during inference time. Defaults to 0.999.
+            model (str, optional): LLM model name. Defaults to "gpt-3.5-turbo".
+            timeout (int, optional): Http request timeout. Defaults to 30.
+            intro (str, optional): Conversation introductory prompt. Defaults to None.
+            filepath (str, optional): Path to file containing conversation history. Defaults to None.
+            update_file (bool, optional): Add new prompts and responses to the file. Defaults to True.
+            proxies (dict, optional): Http request proxies. Defaults to {}.
+            history_offset (int, optional): Limit conversation history to this number of last texts. Defaults to 10250.
+            act (str|int, optional): Awesome prompt key or index. (Used as intro). Defaults to None.
+            system_prompt (str, optional): System prompt to guide the conversation. Defaults to None.
+        """
+        # Update available models from API
+        GROQ.update_available_models(api_key)
+        
+        # Validate model after updating available models
+        if model not in self.AVAILABLE_MODELS:
+            raise ValueError(f"Invalid model: {model}. Choose from: {self.AVAILABLE_MODELS}")
+
+        self.is_conversation = is_conversation
+        self.max_tokens_to_sample = max_tokens
+        self.api_key = api_key
+        self.model = model
+        self.temperature = temperature
+        self.presence_penalty = presence_penalty
+        self.frequency_penalty = frequency_penalty
+        self.top_p = top_p
+        self.chat_endpoint = "https://api.groq.com/openai/v1/chat/completions"
+        self.stream_chunk_size = 64
+        self.timeout = timeout
+        self.last_response = {}
+        self.system_prompt = system_prompt
+        self.available_functions: Dict[str, Callable] = {}  # Store available functions
+        self.headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {self.api_key}",
+        }
+
+        self.__available_optimizers = (
+            method
+            for method in dir(Optimizers)
+            if callable(getattr(Optimizers, method)) and not method.startswith("__")
+        )
+        Conversation.intro = (
+            AwesomePrompts().get_act(
+                act, raise_not_found=True, default=None, case_insensitive=True
+            )
+            if act
+            else intro or Conversation.intro
+        )
+        self.conversation = Conversation(
+            is_conversation, self.max_tokens_to_sample, filepath, update_file
+        )
+        self.conversation.history_offset = history_offset
+        self.session = httpx.AsyncClient(headers=self.headers, proxies=proxies)
+
+    def add_function(self, function_name: str, function: Callable):
+        """Add a function to the available functions dictionary.
+
+        Args:
+            function_name (str): The name of the function to be used in the prompt.
+            function (Callable): The function itself.
+        """
+        self.available_functions[function_name] = function
+
+    async def ask(
+        self,
+        prompt: str,
+        stream: bool = False,
+        raw: bool = False,
+        optimizer: str = None,
+        conversationally: bool = False,
+        tools: Optional[List[Dict[str, Any]]] = None,
+    ) -> Union[dict, AsyncGenerator]:
+        """Chat with AI asynchronously.
+
+        Args:
+            prompt (str): Prompt to be send.
+            stream (bool, optional): Flag for streaming response. Defaults to False.
+            raw (bool, optional): Stream back raw response as received. Defaults to False.
+            optimizer (str, optional): Prompt optimizer name - `[code, shell_command]`. Defaults to None.
+            conversationally (bool, optional): Chat conversationally when using optimizer. Defaults to False.
+            tools (List[Dict[str, Any]], optional): List of tool definitions. See example in class docstring. Defaults to None.
+        Returns:
+           dict|AsyncGenerator : ai content
+        """
+        conversation_prompt = self.conversation.gen_complete_prompt(prompt)
+        if optimizer:
+            if optimizer in self.__available_optimizers:
+                conversation_prompt = getattr(Optimizers, optimizer)(
+                    conversation_prompt if conversationally else prompt
+                )
+            else:
+                raise Exception(
+                    f"Optimizer is not one of {self.__available_optimizers}"
+                )
+
+        messages = [{"content": conversation_prompt, "role": "user"}]
+        if self.system_prompt:
+            messages.insert(0, {"role": "system", "content": self.system_prompt})
+
+        payload = {
+            "frequency_penalty": self.frequency_penalty,
+            "messages": messages,
+            "model": self.model,
+            "presence_penalty": self.presence_penalty,
+            "stream": stream,
+            "temperature": self.temperature,
+            "top_p": self.top_p,
+            "tools": tools
+        }
+
+        async def for_stream():
+            async with self.session.stream(
+                "POST", self.chat_endpoint, json=payload, timeout=self.timeout
+            ) as response:
+                if not response.is_success:
+                    raise exceptions.FailedToGenerateResponseError(
+                        # Removed response.reason_phrase (not available in httpx response)
+                        f"Failed to generate response - ({response.status_code})"
+                    )
+
+                message_load = ""
+                intro_value = "data:"
+                async for value in response.aiter_lines():
+                    try:
+                        if value.startswith(intro_value):
+                            value = value[len(intro_value) :]
+                        resp = json.loads(value)
+                        incomplete_message = await self.get_message(resp)
+                        if incomplete_message:
+                            message_load += incomplete_message
+                            resp["choices"][0]["delta"]["content"] = message_load
+                            self.last_response.update(resp)
+                            yield value if raw else resp
+                        elif raw:
+                            yield value
+                    except json.decoder.JSONDecodeError:
+                        pass
+
+                # Handle tool calls if any (in streaming mode)
+                if 'tool_calls' in self.last_response.get('choices', [{}])[0].get('message', {}):
+                    tool_calls = self.last_response['choices'][0]['message']['tool_calls']
+                    for tool_call in tool_calls:
+                        function_name = tool_call.get('function', {}).get('name')
+                        arguments = json.loads(tool_call.get('function', {}).get('arguments', "{}"))
+                        if function_name in self.available_functions:
+                            tool_response = self.available_functions[function_name](**arguments)
+                            messages.append({
+                                "tool_call_id": tool_call['id'],
+                                "role": "tool",
+                                "name": function_name,
+                                "content": tool_response
+                            })
+                            payload['messages'] = messages
+                            # Make a second call to get the final response
+                            second_response = await self.session.post(
+                                self.chat_endpoint, json=payload, timeout=self.timeout
+                            )
+                            if second_response.is_success:
+                                self.last_response = second_response.json()
+                            else:
+                                raise exceptions.FailedToGenerateResponseError(
+                                    f"Failed to execute tool - {second_response.text}"
+                                )
+
+            self.conversation.update_chat_history(
+                prompt, await self.get_message(self.last_response)
+            )
+
+        async def for_non_stream():
+            response = await self.session.post(
+                self.chat_endpoint, json=payload, timeout=self.timeout
+            )
+            if not response.is_success:
+                raise exceptions.FailedToGenerateResponseError(
+                    # Removed response.reason_phrase (not available in httpx response)
+                    f"Failed to generate response - ({response.status_code})"
+                )
+            resp = response.json()
+
+            # Handle tool calls if any (in non-streaming mode)
+            if 'tool_calls' in resp.get('choices', [{}])[0].get('message', {}):
+                tool_calls = resp['choices'][0]['message']['tool_calls']
+                for tool_call in tool_calls:
+                    function_name = tool_call.get('function', {}).get('name')
+                    arguments = json.loads(tool_call.get('function', {}).get('arguments', "{}"))
+                    if function_name in self.available_functions:
+                        tool_response = self.available_functions[function_name](**arguments)
+                        messages.append({
+                            "tool_call_id": tool_call['id'],
+                            "role": "tool",
+                            "name": function_name,
+                            "content": tool_response
+                        })
+                        payload['messages'] = messages
+                        # Make a second call to get the final response
+                        second_response = await self.session.post(
+                            self.chat_endpoint, json=payload, timeout=self.timeout
+                        )
+                        if second_response.is_success:
+                            resp = second_response.json()
+                        else:
+                            raise exceptions.FailedToGenerateResponseError(
+                                f"Failed to execute tool - {second_response.text}"
+                            )
+
+            self.last_response.update(resp)
+            self.conversation.update_chat_history(
+                prompt, await self.get_message(self.last_response)
+            )
+            return resp
+
+        return for_stream() if stream else await for_non_stream()
+
+    async def chat(
+        self,
+        prompt: str,
+        stream: bool = False,
+        optimizer: str = None,
+        conversationally: bool = False,
+        tools: Optional[List[Dict[str, Any]]] = None,
+    ) -> Union[str, AsyncGenerator]:
+        """Generate response `str` asynchronously.
+        Args:
+            prompt (str): Prompt to be send.
+            stream (bool, optional): Flag for streaming response. Defaults to False.
+            optimizer (str, optional): Prompt optimizer name - `[code, shell_command]`. Defaults to None.
+            conversationally (bool, optional): Chat conversationally when using optimizer. Defaults to False.
+            tools (List[Dict[str, Any]], optional): List of tool definitions. See example in class docstring. Defaults to None.
+        Returns:
+            str|AsyncGenerator: Response generated
+        """
+
+        async def for_stream():
+            async_ask = await self.ask(
+                prompt, True, optimizer=optimizer, conversationally=conversationally, tools=tools
+            )
+            async for response in async_ask:
+                yield await self.get_message(response)
+
+        async def for_non_stream():
+            return await self.get_message(
+                await self.ask(
+                    prompt,
+                    False,
+                    optimizer=optimizer,
+                    conversationally=conversationally,
+                    tools=tools
+                )
+            )
+
+        return for_stream() if stream else await for_non_stream()
+
+    async def get_message(self, response: dict) -> str:
+        """Retrieves message only from response
+
+        Args:
+            response (dict): Response generated by `self.ask`
+
+        Returns:
+            str: Message extracted
+        """
+        assert isinstance(response, dict), "Response should be of dict data-type only"
+        try:
+            # Check delta first for streaming
+            if response.get("choices") and response["choices"][0].get("delta") and response["choices"][0]["delta"].get("content"):
+                return response["choices"][0]["delta"]["content"]
+            # Check message content for non-streaming or final message
+            if response.get("choices") and response["choices"][0].get("message") and response["choices"][0]["message"].get("content"):
+                return response["choices"][0]["message"]["content"]
+        except (KeyError, IndexError, TypeError):
+             # Handle cases where the structure might be different or content is null/missing
+            pass
+        return "" # Return empty string if no content found
+        
+if __name__ == "__main__":
+    # Example usage
+    api_key = "gsk_*******************************"
+    groq = GROQ(api_key=api_key, model="compound-beta")
+    prompt = "What is the capital of France?"
+    response = groq.chat(prompt)
+    print(response)