agno 0.1.2__py3-none-any.whl → 2.3.13__py3-none-any.whl

agno/models/litellm/chat.py

@@ -0,0 +1,503 @@
+import json
+from dataclasses import dataclass
+from os import getenv
+from typing import Any, AsyncIterator, Dict, Iterator, List, Optional, Type, Union
+
+from pydantic import BaseModel
+
+from agno.models.base import Model
+from agno.models.message import Message
+from agno.models.metrics import Metrics
+from agno.models.response import ModelResponse
+from agno.run.agent import RunOutput
+from agno.tools.function import Function
+from agno.utils.log import log_debug, log_error, log_warning
+from agno.utils.openai import _format_file_for_message, audio_to_message, images_to_message
+from agno.utils.tokens import count_schema_tokens
+
+try:
+    import litellm
+    from litellm import validate_environment
+except ImportError:
+    raise ImportError("`litellm` not installed. Please install it via `pip install litellm`")
+
+
+@dataclass
+class LiteLLM(Model):
+    """
+    A class for interacting with LiteLLM Python SDK.
+
+    LiteLLM allows you to use a unified interface for various LLM providers.
+    For more information, see: https://docs.litellm.ai/docs/
+    """
+
+    id: str = "gpt-4o"
+    name: str = "LiteLLM"
+    provider: str = "LiteLLM"
+
+    api_key: Optional[str] = None
+    api_base: Optional[str] = None
+    max_tokens: Optional[int] = None
+    temperature: float = 0.7
+    top_p: float = 1.0
+    metadata: Optional[Dict[str, Any]] = None
+    extra_headers: Optional[Dict[str, Any]] = None
+    extra_query: Optional[Dict[str, Any]] = None
+    extra_body: Optional[Dict[str, Any]] = None
+    request_params: Optional[Dict[str, Any]] = None
+
+    client: Optional[Any] = None
+
+    def __post_init__(self):
+        """Initialize the model after the dataclass initialization."""
+        super().__post_init__()
+
+        # Set up API key from environment variable if not already set
+        if not self.client and not self.api_key:
+            self.api_key = getenv("LITELLM_API_KEY")
+            if not self.api_key:
+                # Check for other present valid keys, e.g. OPENAI_API_KEY if self.id is an OpenAI model
+                env_validation = validate_environment(model=self.id, api_base=self.api_base)
+                if not env_validation.get("keys_in_environment"):
+                    log_error(
+                        "LITELLM_API_KEY not set. Please set the LITELLM_API_KEY or other valid environment variables."
+                    )
+
+    def get_client(self) -> Any:
+        """
+        Returns a LiteLLM client.
+
+        Returns:
+            Any: An instance of the LiteLLM client.
+        """
+        if self.client is not None:
+            return self.client
+
+        self.client = litellm
+        return self.client
+
+    def _format_messages(self, messages: List[Message], compress_tool_results: bool = False) -> List[Dict[str, Any]]:
+        """Format messages for LiteLLM API."""
+        formatted_messages = []
+        for m in messages:
+            # Use compressed content for tool messages if compression is active
+            if m.role == "tool":
+                content = m.get_content(use_compressed_content=compress_tool_results)
+            else:
+                content = m.content if m.content is not None else ""
+
+            msg = {"role": m.role, "content": content}
+
+            # Handle media
+            if (m.images is not None and len(m.images) > 0) or (m.audio is not None and len(m.audio) > 0):
+                if isinstance(m.content, str):
+                    content_list = [{"type": "text", "text": m.content}]
+                    if m.images is not None:
+                        content_list.extend(images_to_message(images=m.images))
+                    if m.audio is not None:
+                        content_list.extend(audio_to_message(audio=m.audio))
+                    msg["content"] = content_list
+
+            if m.videos is not None and len(m.videos) > 0:
+                log_warning("Video input is currently unsupported by LLM providers.")
+
+            # Handle files
+            if m.files is not None:
+                if isinstance(msg["content"], str):
+                    content_list = [{"type": "text", "text": msg["content"]}]
+                else:
+                    content_list = msg["content"] if isinstance(msg["content"], list) else []
+                for file in m.files:
+                    file_part = _format_file_for_message(file)
+                    if file_part:
+                        content_list.append(file_part)
+                msg["content"] = content_list
+
+            # Handle tool calls in assistant messages
+            if m.role == "assistant" and m.tool_calls:
+                msg["tool_calls"] = [
+                    {
+                        "id": tc.get("id", f"call_{i}"),
+                        "type": "function",
+                        "function": {"name": tc["function"]["name"], "arguments": tc["function"]["arguments"]},
+                    }
+                    for i, tc in enumerate(m.tool_calls)
+                ]
+
+            # Handle tool responses
+            if m.role == "tool":
+                msg["tool_call_id"] = m.tool_call_id or ""
+                msg["name"] = m.name or ""
+
+                if m.audio is not None and len(m.audio) > 0:
+                    log_warning("Audio input is currently unsupported.")
+
+                if m.images is not None and len(m.images) > 0:
+                    log_warning("Image input is currently unsupported.")
+
+                if m.videos is not None and len(m.videos) > 0:
+                    log_warning("Video input is currently unsupported.")
+            formatted_messages.append(msg)
+
+        return formatted_messages
+
+    def get_request_params(self, tools: Optional[List[Dict[str, Any]]] = None) -> Dict[str, Any]:
+        """
+        Returns keyword arguments for API requests.
+
+        Returns:
+            Dict[str, Any]: The API kwargs for the model.
+        """
+        base_params: Dict[str, Any] = {
+            "model": self.id,
+            "temperature": self.temperature,
+            "top_p": self.top_p,
+        }
+
+        if self.max_tokens:
+            base_params["max_tokens"] = self.max_tokens
+        if self.api_key:
+            base_params["api_key"] = self.api_key
+        if self.api_base:
+            base_params["api_base"] = self.api_base
+        if self.extra_headers:
+            base_params["extra_headers"] = self.extra_headers
+        if self.extra_query:
+            base_params["extra_query"] = self.extra_query
+        if tools:
+            base_params["tools"] = tools
+            base_params["tool_choice"] = "auto"
+
+        # Handle metadata via extra_body as per LiteLLM docs
+        if self.metadata:
+            if self.extra_body:
+                base_params["extra_body"] = {**self.extra_body, "metadata": self.metadata}
+            else:
+                base_params["extra_body"] = {"metadata": self.metadata}
+        elif self.extra_body:
+            base_params["extra_body"] = self.extra_body
+
+        # Add additional request params if provided
+        request_params: Dict[str, Any] = {k: v for k, v in base_params.items() if v is not None}
+        if self.request_params:
+            request_params.update(self.request_params)
+
+        if request_params:
+            log_debug(f"Calling {self.provider} with request parameters: {request_params}", log_level=2)
+        return request_params
+
+    def invoke(
+        self,
+        messages: List[Message],
+        assistant_message: Message,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+        tools: Optional[List[Dict[str, Any]]] = None,
+        tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
+        run_response: Optional[RunOutput] = None,
+        compress_tool_results: bool = False,
+    ) -> ModelResponse:
+        """Sends a chat completion request to the LiteLLM API."""
+        completion_kwargs = self.get_request_params(tools=tools)
+        completion_kwargs["messages"] = self._format_messages(messages, compress_tool_results)
+
+        if run_response and run_response.metrics:
+            run_response.metrics.set_time_to_first_token()
+
+        assistant_message.metrics.start_timer()
+
+        provider_response = self.get_client().completion(**completion_kwargs)
+
+        assistant_message.metrics.stop_timer()
+
+        model_response = self._parse_provider_response(provider_response, response_format=response_format)
+        return model_response
+
+    def invoke_stream(
+        self,
+        messages: List[Message],
+        assistant_message: Message,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+        tools: Optional[List[Dict[str, Any]]] = None,
+        tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
+        run_response: Optional[RunOutput] = None,
+        compress_tool_results: bool = False,
+    ) -> Iterator[ModelResponse]:
+        """Sends a streaming chat completion request to the LiteLLM API."""
+        completion_kwargs = self.get_request_params(tools=tools)
+        completion_kwargs["messages"] = self._format_messages(messages, compress_tool_results)
+        completion_kwargs["stream"] = True
+        completion_kwargs["stream_options"] = {"include_usage": True}
+
+        if run_response and run_response.metrics:
+            run_response.metrics.set_time_to_first_token()
+
+        assistant_message.metrics.start_timer()
+
+        for chunk in self.get_client().completion(**completion_kwargs):
+            yield self._parse_provider_response_delta(chunk)
+
+        assistant_message.metrics.stop_timer()
+
+    async def ainvoke(
+        self,
+        messages: List[Message],
+        assistant_message: Message,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+        tools: Optional[List[Dict[str, Any]]] = None,
+        tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
+        run_response: Optional[RunOutput] = None,
+        compress_tool_results: bool = False,
+    ) -> ModelResponse:
+        """Sends an asynchronous chat completion request to the LiteLLM API."""
+        completion_kwargs = self.get_request_params(tools=tools)
+        completion_kwargs["messages"] = self._format_messages(messages, compress_tool_results)
+
+        if run_response and run_response.metrics:
+            run_response.metrics.set_time_to_first_token()
+
+        assistant_message.metrics.start_timer()
+
+        provider_response = await self.get_client().acompletion(**completion_kwargs)
+
+        assistant_message.metrics.stop_timer()
+
+        model_response = self._parse_provider_response(provider_response, response_format=response_format)
+        return model_response
+
+    async def ainvoke_stream(
+        self,
+        messages: List[Message],
+        assistant_message: Message,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+        tools: Optional[List[Dict[str, Any]]] = None,
+        tool_choice: Optional[Union[str, Dict[str, Any]]] = None,
+        run_response: Optional[RunOutput] = None,
+        compress_tool_results: bool = False,
+    ) -> AsyncIterator[ModelResponse]:
+        """Sends an asynchronous streaming chat request to the LiteLLM API."""
+        completion_kwargs = self.get_request_params(tools=tools)
+        completion_kwargs["messages"] = self._format_messages(messages, compress_tool_results)
+        completion_kwargs["stream"] = True
+        completion_kwargs["stream_options"] = {"include_usage": True}
+
+        if run_response and run_response.metrics:
+            run_response.metrics.set_time_to_first_token()
+
+        assistant_message.metrics.start_timer()
+
+        try:
+            # litellm.acompletion returns a coroutine that resolves to an async iterator
+            # We need to await it first to get the actual async iterator
+            async_stream = await self.get_client().acompletion(**completion_kwargs)
+            async for chunk in async_stream:
+                yield self._parse_provider_response_delta(chunk)
+
+            assistant_message.metrics.stop_timer()
+
+        except Exception as e:
+            log_error(f"Error in streaming response: {e}")
+            raise
+
+    def _parse_provider_response(self, response: Any, **kwargs) -> ModelResponse:
+        """Parse the provider response."""
+        model_response = ModelResponse()
+
+        response_message = response.choices[0].message
+
+        if response_message.content is not None:
+            model_response.content = response_message.content
+
+        if hasattr(response_message, "tool_calls") and response_message.tool_calls:
+            model_response.tool_calls = []
+            for tool_call in response_message.tool_calls:
+                model_response.tool_calls.append(
+                    {
+                        "id": tool_call.id,
+                        "type": "function",
+                        "function": {"name": tool_call.function.name, "arguments": tool_call.function.arguments},
+                    }
+                )
+
+        if response.usage is not None:
+            model_response.response_usage = self._get_metrics(response.usage)
+
+        return model_response
+
+    def _parse_provider_response_delta(self, response_delta: Any) -> ModelResponse:
+        """Parse the provider response delta for streaming responses."""
+        model_response = ModelResponse()
+
+        if hasattr(response_delta, "choices") and len(response_delta.choices) > 0:
+            choice_delta = response_delta.choices[0].delta
+
+            if choice_delta:
+                if hasattr(choice_delta, "content") and choice_delta.content is not None:
+                    model_response.content = choice_delta.content
+
+                if hasattr(choice_delta, "tool_calls") and choice_delta.tool_calls:
+                    processed_tool_calls = []
+                    for tool_call in choice_delta.tool_calls:
+                        # Get the actual index from the tool call, defaulting to 0 if not available
+                        actual_index = getattr(tool_call, "index", 0) if hasattr(tool_call, "index") else 0
+
+                        # Create a basic structure with the correct index
+                        tool_call_dict = {"index": actual_index, "type": "function"}
+
+                        # Extract ID if available
+                        if hasattr(tool_call, "id") and tool_call.id is not None:
+                            tool_call_dict["id"] = tool_call.id
+
+                        # Extract function data
+                        function_data = {}
+                        if hasattr(tool_call, "function"):
+                            if hasattr(tool_call.function, "name") and tool_call.function.name is not None:
+                                function_data["name"] = tool_call.function.name
+                            if hasattr(tool_call.function, "arguments") and tool_call.function.arguments is not None:
+                                function_data["arguments"] = tool_call.function.arguments
+
+                        tool_call_dict["function"] = function_data
+                        processed_tool_calls.append(tool_call_dict)
+
+                    model_response.tool_calls = processed_tool_calls
+
+        # Add usage metrics if present in streaming response
+        if hasattr(response_delta, "usage") and response_delta.usage is not None:
+            model_response.response_usage = self._get_metrics(response_delta.usage)
+
+        return model_response
+
+    @staticmethod
+    def parse_tool_calls(tool_calls_data: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+        """
+        Build tool calls from streamed tool call data.
+
+        Args:
+            tool_calls_data (List[Dict[str, Any]]): The tool call data to build from.
+
+        Returns:
+            List[Dict[str, Any]]: The built tool calls.
+        """
+        # Early return for empty list
+        if not tool_calls_data:
+            return []
+
+        # Group tool calls by index
+        tool_calls_by_index: Dict[int, Dict[str, Any]] = {}
+
+        for tc in tool_calls_data:
+            # Get index (default to 0)
+            index = tc.get("index", 0)
+            if not isinstance(index, int):
+                index = 0
+
+            # Initialize if first time seeing this index
+            if index not in tool_calls_by_index:
+                tool_calls_by_index[index] = {"id": None, "type": "function", "function": {"name": "", "arguments": ""}}
+
+            # Update with new information
+            if tc.get("id") is not None:
+                tool_calls_by_index[index]["id"] = tc["id"]
+
+            if tc.get("type") is not None:
+                tool_calls_by_index[index]["type"] = tc["type"]
+
+            # Update function information
+            function_data = tc.get("function", {})
+            if not isinstance(function_data, dict):
+                function_data = {}
+
+            # Update function name if provided
+            if function_data.get("name") is not None:
+                name = function_data.get("name", "")
+                if isinstance(tool_calls_by_index[index]["function"], dict):
+                    # type: ignore
+                    tool_calls_by_index[index]["function"]["name"] = name
+
+            # Update function arguments if provided
+            if function_data.get("arguments") is not None:
+                args = function_data.get("arguments", "")
+                if isinstance(tool_calls_by_index[index]["function"], dict):
+                    current_args = tool_calls_by_index[index]["function"].get("arguments", "")  # type: ignore
+                    if isinstance(current_args, str) and isinstance(args, str):
+                        # type: ignore
+                        tool_calls_by_index[index]["function"]["arguments"] = current_args + args
+
+        # Process arguments - Ensure they're valid JSON for the Message.log() method
+        result = []
+        for tc in tool_calls_by_index.values():
+            # Make a safe copy to avoid modifying the original
+            tc_copy = {
+                "id": tc.get("id"),
+                "type": tc.get("type", "function"),
+                "function": {"name": "", "arguments": ""},
+            }
+
+            # Safely copy function data
+            if isinstance(tc.get("function"), dict):
+                func_dict = tc.get("function", {})
+                tc_copy["function"]["name"] = func_dict.get("name", "")
+
+                # Process arguments
+                args = func_dict.get("arguments", "")
+                if args and isinstance(args, str):
+                    try:
+                        # Check if arguments are already valid JSON
+                        parsed = json.loads(args)
+                        # If it's not a dict, convert to a JSON string of a dict
+                        if not isinstance(parsed, dict):
+                            tc_copy["function"]["arguments"] = json.dumps({"value": parsed})
+                        else:
+                            tc_copy["function"]["arguments"] = args
+                    except json.JSONDecodeError:
+                        # If not valid JSON, make it a JSON dict
+                        tc_copy["function"]["arguments"] = json.dumps({"text": args})
+
+            result.append(tc_copy)
+
+        return result
+
+    def _get_metrics(self, response_usage: Any) -> Metrics:
+        """
+        Parse the given LiteLLM usage into an Agno Metrics object.
+
+        Args:
+            response_usage: Usage data from LiteLLM
+
+        Returns:
+            Metrics: Parsed metrics data
+        """
+        metrics = Metrics()
+
+        if isinstance(response_usage, dict):
+            metrics.input_tokens = response_usage.get("prompt_tokens") or 0
+            metrics.output_tokens = response_usage.get("completion_tokens") or 0
+        else:
+            metrics.input_tokens = response_usage.prompt_tokens or 0
+            metrics.output_tokens = response_usage.completion_tokens or 0
+
+        metrics.total_tokens = metrics.input_tokens + metrics.output_tokens
+
+        return metrics
+
+    def count_tokens(
+        self,
+        messages: List[Message],
+        tools: Optional[List[Union[Function, Dict[str, Any]]]] = None,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+    ) -> int:
+        formatted_messages = self._format_messages(messages, compress_tool_results=True)
+        formatted_tools = self._format_tools(tools) if tools else None
+        tokens = litellm.token_counter(
+            model=self.id,
+            messages=formatted_messages,
+            tools=formatted_tools,  # type: ignore
+        )
+        return tokens + count_schema_tokens(response_format, self.id)
+
+    async def acount_tokens(
+        self,
+        messages: List[Message],
+        tools: Optional[List[Union[Function, Dict[str, Any]]]] = None,
+        response_format: Optional[Union[Dict, Type[BaseModel]]] = None,
+    ) -> int:
+        return self.count_tokens(messages, tools, response_format)

agno/models/litellm/litellm_openai.py

@@ -0,0 +1,42 @@
+from dataclasses import dataclass, field
+from os import getenv
+from typing import Any, Dict, Optional
+
+from agno.exceptions import ModelAuthenticationError
+from agno.models.openai.like import OpenAILike
+
+
+@dataclass
+class LiteLLMOpenAI(OpenAILike):
+    """
+    A class for interacting with LiteLLM.
+
+    Attributes:
+        id (str): The id of the LiteLLM model. Default is "gpt-4o".
+        name (str): The name of this chat model instance. Default is "LiteLLM".
+        provider (str): The provider of the model. Default is "LiteLLM".
+        base_url (str): The base url to which the requests are sent.
+    """
+
+    id: str = "gpt-4o"
+    name: str = "LiteLLM"
+    provider: str = "LiteLLM"
+
+    api_key: Optional[str] = field(default_factory=lambda: getenv("LITELLM_API_KEY"))
+    base_url: str = "http://0.0.0.0:4000"
+
+    def _get_client_params(self) -> Dict[str, Any]:
+        """
+        Returns client parameters for API requests, checking for LITELLM_API_KEY.
+
+        Returns:
+            Dict[str, Any]: A dictionary of client parameters for API requests.
+        """
+        if not self.api_key:
+            self.api_key = getenv("LITELLM_API_KEY")
+            if not self.api_key:
+                raise ModelAuthenticationError(
+                    message="LITELLM_API_KEY not set. Please set the LITELLM_API_KEY environment variable.",
+                    model_name=self.name,
+                )
+        return super()._get_client_params()

agno/models/llama_cpp/__init__.py

@@ -0,0 +1,5 @@
+from agno.models.llama_cpp.llama_cpp import LlamaCpp
+
+__all__ = [
+    "LlamaCpp",
+]

agno/models/llama_cpp/llama_cpp.py

@@ -0,0 +1,22 @@
+from dataclasses import dataclass
+
+from agno.models.openai.like import OpenAILike
+
+
+@dataclass
+class LlamaCpp(OpenAILike):
+    """
+    A class for interacting with LLMs using Llama CPP.
+
+    Attributes:
+        id (str): The id of the Llama CPP model. Default is "ggml-org/gpt-oss-20b-GGUF".
+        name (str): The name of this chat model instance. Default is "LlamaCpp".
+        provider (str): The provider of the model. Default is "LlamaCpp".
+        base_url (str): The base url to which the requests are sent.
+    """
+
+    id: str = "ggml-org/gpt-oss-20b-GGUF"
+    name: str = "LlamaCpp"
+    provider: str = "LlamaCpp"
+
+    base_url: str = "http://127.0.0.1:8080/v1"

agno/models/lmstudio/__init__.py

@@ -0,0 +1,5 @@
+from agno.models.lmstudio.lmstudio import LMStudio
+
+__all__ = [
+    "LMStudio",
+]

agno/models/lmstudio/lmstudio.py

@@ -0,0 +1,25 @@
+from dataclasses import dataclass
+
+from agno.models.openai.like import OpenAILike
+
+
+@dataclass
+class LMStudio(OpenAILike):
+    """
+    A class for interacting with LM Studio.
+
+    Attributes:
+        id (str): The id of the LM Studio model. Default is "qwen2.5-7b-instruct-1m".
+        name (str): The name of this chat model instance. Default is "LMStudio".
+        provider (str): The provider of the model. Default is "LMStudio".
+        base_url (str): The base url to which the requests are sent.
+    """
+
+    id: str = "qwen2.5-7b-instruct-1m"
+    name: str = "LMStudio"
+    provider: str = "LMStudio"
+
+    base_url: str = "http://127.0.0.1:1234/v1"
+
+    supports_native_structured_outputs: bool = False
+    supports_json_schema_outputs: bool = True