agno 2.0.4__py3-none-any.whl → 2.0.6__py3-none-any.whl

agno/models/base.py

@@ -21,7 +21,7 @@ from uuid import uuid4
 from pydantic import BaseModel
 
 from agno.exceptions import AgentRunException
-from agno.media import Audio, Image, Video
+from agno.media import Audio, File, Image, Video
 from agno.models.message import Citations, Message
 from agno.models.metrics import Metrics
 from agno.models.response import ModelResponse, ModelResponseEvent, ToolExecution
@@ -46,6 +46,7 @@ class MessageData:
     response_audio: Optional[Audio] = None
     response_image: Optional[Image] = None
     response_video: Optional[Video] = None
+    response_file: Optional[File] = None
 
     # Data from the provider that we might need on subsequent messages
     response_provider_data: Optional[Dict[str, Any]] = None
@@ -266,6 +267,11 @@ class Model(ABC):
                                 model_response.videos = []
                             model_response.videos.extend(function_call_response.videos)
 
+                        if function_call_response.files is not None:
+                            if model_response.files is None:
+                                model_response.files = []
+                            model_response.files.extend(function_call_response.files)
+
                         if (
                             function_call_response.event
                             in [
@@ -293,7 +299,7 @@ class Model(ABC):
                     messages=messages, function_call_results=function_call_results, **model_response.extra or {}
                 )
 
-                if any(msg.images or msg.videos or msg.audio for msg in function_call_results):
+                if any(msg.images or msg.videos or msg.audio or msg.files for msg in function_call_results):
                     # Handle function call media
                     self._handle_function_call_media(messages=messages, function_call_results=function_call_results)
 
@@ -402,6 +408,11 @@ class Model(ABC):
                                 model_response.videos = []
                             model_response.videos.extend(function_call_response.videos)
 
+                        if function_call_response.files is not None:
+                            if model_response.files is None:
+                                model_response.files = []
+                            model_response.files.extend(function_call_response.files)
+
                         if (
                             function_call_response.event
                             in [
@@ -428,7 +439,7 @@ class Model(ABC):
                     messages=messages, function_call_results=function_call_results, **model_response.extra or {}
                 )
 
-                if any(msg.images or msg.videos or msg.audio for msg in function_call_results):
+                if any(msg.images or msg.videos or msg.audio or msg.files for msg in function_call_results):
                     # Handle function call media
                     self._handle_function_call_media(messages=messages, function_call_results=function_call_results)
 
@@ -607,6 +618,10 @@ class Model(ABC):
             if provider_response.videos:
                 assistant_message.video_output = provider_response.videos[-1]  # Taking last (most recent) video
 
+        if provider_response.files is not None:
+            if provider_response.files:
+                assistant_message.file_output = provider_response.files[-1]  # Taking last (most recent) file
+
         if provider_response.audios is not None:
             if provider_response.audios:
                 assistant_message.audio_output = provider_response.audios[-1]  # Taking last (most recent) audio
@@ -1213,6 +1228,8 @@ class Model(ABC):
                     function_execution_result.videos = tool_result.videos
                 if tool_result.audios:
                     function_execution_result.audios = tool_result.audios
+                if tool_result.files:
+                    function_execution_result.files = tool_result.files
             else:
                 function_call_output = str(function_execution_result.result) if function_execution_result.result else ""
 
@@ -1246,6 +1263,7 @@ class Model(ABC):
             images=function_execution_result.images,
             videos=function_execution_result.videos,
             audios=function_execution_result.audios,
+            files=function_execution_result.files,
         )
 
         # Add function call to function call results
@@ -1617,6 +1635,8 @@ class Model(ABC):
                         function_execution_result.videos = tool_result.videos
                     if tool_result.audios:
                         function_execution_result.audios = tool_result.audios
+                    if tool_result.files:
+                        function_execution_result.files = tool_result.files
                 else:
                     function_call_output = str(function_call.result)
 
@@ -1649,6 +1669,7 @@ class Model(ABC):
                 images=function_execution_result.images,
                 videos=function_execution_result.videos,
                 audios=function_execution_result.audios,
+                files=function_execution_result.files,
             )
 
             # Add function call result to function call results
@@ -1698,6 +1719,7 @@ class Model(ABC):
         all_images: List[Image] = []
         all_videos: List[Video] = []
         all_audio: List[Audio] = []
+        all_files: List[File] = []
 
         for result_message in function_call_results:
             if result_message.images:
@@ -1713,15 +1735,20 @@ class Model(ABC):
                 all_audio.extend(result_message.audio)
                 result_message.audio = None
 
+            if result_message.files:
+                all_files.extend(result_message.files)
+                result_message.files = None
+
         # If we have media artifacts, add a follow-up "user" message instead of a "tool"
         # message with the media artifacts which throws error for some models
-        if all_images or all_videos or all_audio:
+        if all_images or all_videos or all_audio or all_files:
             media_message = Message(
                 role="user",
                 content="Take note of the following content",
                 images=all_images if all_images else None,
                 videos=all_videos if all_videos else None,
                 audio=all_audio if all_audio else None,
+                files=all_files if all_files else None,
             )
             messages.append(media_message)
 

agno/models/cerebras/cerebras_openai.py

@@ -1,5 +1,5 @@
 import json
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, List, Optional, Type, Union
 
@@ -18,7 +18,7 @@ class CerebrasOpenAI(OpenAILike):
 
     parallel_tool_calls: Optional[bool] = None
     base_url: str = "https://api.cerebras.ai/v1"
-    api_key: Optional[str] = getenv("CEREBRAS_API_KEY", None)
+    api_key: Optional[str] = field(default_factory=lambda: getenv("CEREBRAS_API_KEY", None))
 
     def get_request_params(
         self,

agno/models/deepinfra/deepinfra.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,7 +22,7 @@ class DeepInfra(OpenAILike):
     name: str = "DeepInfra"
     provider: str = "DeepInfra"
 
-    api_key: Optional[str] = getenv("DEEPINFRA_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("DEEPINFRA_API_KEY"))
     base_url: str = "https://api.deepinfra.com/v1/openai"
 
     supports_native_structured_outputs: bool = False

agno/models/deepseek/deepseek.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, Optional
 
@@ -23,7 +23,7 @@ class DeepSeek(OpenAILike):
     name: str = "DeepSeek"
     provider: str = "DeepSeek"
 
-    api_key: Optional[str] = getenv("DEEPSEEK_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("DEEPSEEK_API_KEY"))
     base_url: str = "https://api.deepseek.com"
 
     # Their support for structured outputs is currently broken

agno/models/fireworks/fireworks.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,5 +22,5 @@ class Fireworks(OpenAILike):
     name: str = "Fireworks"
     provider: str = "Fireworks"
 
-    api_key: Optional[str] = getenv("FIREWORKS_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("FIREWORKS_API_KEY"))
     base_url: str = "https://api.fireworks.ai/inference/v1"

agno/models/internlm/internlm.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,5 +22,5 @@ class InternLM(OpenAILike):
     name: str = "InternLM"
     provider: str = "InternLM"
 
-    api_key: Optional[str] = getenv("INTERNLM_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("INTERNLM_API_KEY"))
     base_url: Optional[str] = "https://internlm-chat.intern-ai.org.cn/puyu/api/v1/chat/completions"

agno/models/langdb/langdb.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, Optional
 
@@ -22,10 +22,10 @@ class LangDB(OpenAILike):
     name: str = "LangDB"
     provider: str = "LangDB"
 
-    api_key: Optional[str] = getenv("LANGDB_API_KEY")
-    project_id: Optional[str] = getenv("LANGDB_PROJECT_ID")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("LANGDB_API_KEY"))
+    project_id: Optional[str] = field(default_factory=lambda: getenv("LANGDB_PROJECT_ID"))
 
-    base_host_url: str = getenv("LANGDB_API_BASE_URL", "https://api.us-east-1.langdb.ai")
+    base_host_url: str = field(default_factory=lambda: getenv("LANGDB_API_BASE_URL", "https://api.us-east-1.langdb.ai"))
 
     base_url: Optional[str] = None
     label: Optional[str] = None

agno/models/litellm/litellm_openai.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -21,5 +21,5 @@ class LiteLLMOpenAI(OpenAILike):
     name: str = "LiteLLM"
     provider: str = "LiteLLM"
 
-    api_key: Optional[str] = getenv("LITELLM_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("LITELLM_API_KEY"))
     base_url: str = "http://0.0.0.0:4000"

agno/models/message.py

@@ -74,6 +74,7 @@ class Message(BaseModel):
     audio_output: Optional[Audio] = None
     image_output: Optional[Image] = None
     video_output: Optional[Video] = None
+    file_output: Optional[File] = None
 
     # The thinking content from the model
     redacted_reasoning_content: Optional[str] = None
@@ -121,6 +122,138 @@ class Message(BaseModel):
 
     @classmethod
     def from_dict(cls, data: Dict[str, Any]) -> "Message":
+        # Handle image reconstruction properly
+        if "images" in data and data["images"]:
+            reconstructed_images = []
+            for i, img_data in enumerate(data["images"]):
+                if isinstance(img_data, dict):
+                    # If content is base64, decode it back to bytes
+                    if "content" in img_data and isinstance(img_data["content"], str):
+                        reconstructed_images.append(
+                            Image.from_base64(
+                                img_data["content"],
+                                id=img_data.get("id"),
+                                mime_type=img_data.get("mime_type"),
+                                format=img_data.get("format"),
+                            )
+                        )
+                    else:
+                        # Regular image (filepath/url)
+                        reconstructed_images.append(Image(**img_data))
+                else:
+                    reconstructed_images.append(img_data)
+            data["images"] = reconstructed_images
+
+        # Handle audio reconstruction properly
+        if "audio" in data and data["audio"]:
+            reconstructed_audio = []
+            for i, aud_data in enumerate(data["audio"]):
+                if isinstance(aud_data, dict):
+                    # If content is base64, decode it back to bytes
+                    if "content" in aud_data and isinstance(aud_data["content"], str):
+                        reconstructed_audio.append(
+                            Audio.from_base64(
+                                aud_data["content"],
+                                id=aud_data.get("id"),
+                                mime_type=aud_data.get("mime_type"),
+                                transcript=aud_data.get("transcript"),
+                                expires_at=aud_data.get("expires_at"),
+                                sample_rate=aud_data.get("sample_rate", 24000),
+                                channels=aud_data.get("channels", 1),
+                            )
+                        )
+                    else:
+                        reconstructed_audio.append(Audio(**aud_data))
+                else:
+                    reconstructed_audio.append(aud_data)
+            data["audio"] = reconstructed_audio
+
+        # Handle video reconstruction properly
+        if "videos" in data and data["videos"]:
+            reconstructed_videos = []
+            for i, vid_data in enumerate(data["videos"]):
+                if isinstance(vid_data, dict):
+                    # If content is base64, decode it back to bytes
+                    if "content" in vid_data and isinstance(vid_data["content"], str):
+                        reconstructed_videos.append(
+                            Video.from_base64(
+                                vid_data["content"],
+                                id=vid_data.get("id"),
+                                mime_type=vid_data.get("mime_type"),
+                                format=vid_data.get("format"),
+                            )
+                        )
+                    else:
+                        reconstructed_videos.append(Video(**vid_data))
+                else:
+                    reconstructed_videos.append(vid_data)
+            data["videos"] = reconstructed_videos
+
+        # Handle file reconstruction properly
+        if "files" in data and data["files"]:
+            reconstructed_files = []
+            for i, file_data in enumerate(data["files"]):
+                if isinstance(file_data, dict):
+                    # If content is base64, decode it back to bytes
+                    if "content" in file_data and isinstance(file_data["content"], str):
+                        reconstructed_files.append(
+                            File.from_base64(
+                                file_data["content"],
+                                id=file_data.get("id"),
+                                mime_type=file_data.get("mime_type"),
+                                filename=file_data.get("filename"),
+                                name=file_data.get("name"),
+                                format=file_data.get("format"),
+                            )
+                        )
+                    else:
+                        reconstructed_files.append(File(**file_data))
+                else:
+                    reconstructed_files.append(file_data)
+            data["files"] = reconstructed_files
+
+        if "audio_output" in data and data["audio_output"]:
+            aud_data = data["audio_output"]
+            if isinstance(aud_data, dict):
+                if "content" in aud_data and isinstance(aud_data["content"], str):
+                    data["audio_output"] = Audio.from_base64(
+                        aud_data["content"],
+                        id=aud_data.get("id"),
+                        mime_type=aud_data.get("mime_type"),
+                        transcript=aud_data.get("transcript"),
+                        expires_at=aud_data.get("expires_at"),
+                        sample_rate=aud_data.get("sample_rate", 24000),
+                        channels=aud_data.get("channels", 1),
+                    )
+                else:
+                    data["audio_output"] = Audio(**aud_data)
+
+        if "image_output" in data and data["image_output"]:
+            img_data = data["image_output"]
+            if isinstance(img_data, dict):
+                if "content" in img_data and isinstance(img_data["content"], str):
+                    data["image_output"] = Image.from_base64(
+                        img_data["content"],
+                        id=img_data.get("id"),
+                        mime_type=img_data.get("mime_type"),
+                        format=img_data.get("format"),
+                    )
+                else:
+                    data["image_output"] = Image(**img_data)
+
+        if "video_output" in data and data["video_output"]:
+            vid_data = data["video_output"]
+            if isinstance(vid_data, dict):
+                if "content" in vid_data and isinstance(vid_data["content"], str):
+                    data["video_output"] = Video.from_base64(
+                        vid_data["content"],
+                        id=vid_data.get("id"),
+                        mime_type=vid_data.get("mime_type"),
+                        format=vid_data.get("format"),
+                    )
+                else:
+                    data["video_output"] = Video(**vid_data)
+
         return cls(**data)
 
     def to_dict(self) -> Dict[str, Any]:
@@ -152,6 +285,8 @@ class Message(BaseModel):
             message_dict["audio"] = [aud.to_dict() for aud in self.audio]
         if self.videos:
             message_dict["videos"] = [vid.to_dict() for vid in self.videos]
+        if self.files:
+            message_dict["files"] = [file.to_dict() for file in self.files]
         if self.audio_output:
             message_dict["audio_output"] = self.audio_output.to_dict()
 

agno/models/meta/llama_openai.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, Optional
 
@@ -31,7 +31,7 @@ class LlamaOpenAI(OpenAILike):
     name: str = "LlamaOpenAI"
     provider: str = "LlamaOpenAI"
 
-    api_key: Optional[str] = getenv("LLAMA_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("LLAMA_API_KEY"))
     base_url: Optional[str] = "https://api.llama.com/compat/v1/"
 
     # Request parameters

agno/models/nebius/nebius.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, Optional
 
@@ -23,7 +23,7 @@ class Nebius(OpenAILike):
     name: str = "Nebius"
     provider: str = "Nebius"
 
-    api_key: Optional[str] = getenv("NEBIUS_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("NEBIUS_API_KEY"))
     base_url: str = "https://api.studio.nebius.com/v1/"
 
     def _get_client_params(self) -> Dict[str, Any]:

agno/models/nexus/__init__.py

@@ -0,0 +1,3 @@
+from agno.models.nexus.nexus import Nexus
+
+__all__ = ["Nexus"]

agno/models/nexus/nexus.py

@@ -0,0 +1,25 @@
+from dataclasses import dataclass
+
+from agno.models.openai.like import OpenAILike
+
+
+@dataclass
+class Nexus(OpenAILike):
+    """
+    A class for interacting with Nvidia models.
+
+    Attributes:
+        id (str): The id of the Nexus model to use. Default is "nvidia/llama-3.1-nemotron-70b-instruct".
+        name (str): The name of this chat model instance. Default is "Nexus"
+        provider (str): The provider of the model. Default is "Nexus".
+        api_key (str): The api key to authorize request to Nexus.
+        base_url (str): The base url to which the requests are sent.
+    """
+
+    id: str = "openai/gpt-4"
+    name: str = "Nexus"
+    provider: str = "Nexus"
+
+    base_url: str = "http://localhost:8000/llm/v1/"
+
+    supports_native_structured_outputs: bool = False

agno/models/nvidia/nvidia.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,7 +22,7 @@ class Nvidia(OpenAILike):
     name: str = "Nvidia"
     provider: str = "Nvidia"
 
-    api_key: Optional[str] = getenv("NVIDIA_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("NVIDIA_API_KEY"))
     base_url: str = "https://integrate.api.nvidia.com/v1"
 
     supports_native_structured_outputs: bool = False

agno/models/openai/responses.py

@@ -1088,4 +1088,10 @@ class OpenAIResponses(Model):
         metrics.output_tokens = response_usage.output_tokens or 0
         metrics.total_tokens = response_usage.total_tokens or 0
 
+        if input_tokens_details := response_usage.input_tokens_details:
+            metrics.cache_read_tokens = input_tokens_details.cached_tokens
+
+        if output_tokens_details := response_usage.output_tokens_details:
+            metrics.reasoning_tokens = output_tokens_details.reasoning_tokens
+
         return metrics

agno/models/openrouter/openrouter.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -23,6 +23,6 @@ class OpenRouter(OpenAILike):
     name: str = "OpenRouter"
     provider: str = "OpenRouter"
 
-    api_key: Optional[str] = getenv("OPENROUTER_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("OPENROUTER_API_KEY"))
     base_url: str = "https://openrouter.ai/api/v1"
     max_tokens: int = 1024

agno/models/perplexity/perplexity.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, List, Optional, Type, Union
 
@@ -42,7 +42,7 @@ class Perplexity(OpenAILike):
     name: str = "Perplexity"
     provider: str = "Perplexity"
 
-    api_key: Optional[str] = getenv("PERPLEXITY_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("PERPLEXITY_API_KEY"))
     base_url: str = "https://api.perplexity.ai/"
     max_tokens: int = 1024
     top_k: Optional[float] = None

agno/models/portkey/portkey.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, Optional, cast
 
@@ -30,8 +30,8 @@ class Portkey(OpenAILike):
     name: str = "Portkey"
     provider: str = "Portkey"
 
-    portkey_api_key: Optional[str] = getenv("PORTKEY_API_KEY")
-    virtual_key: Optional[str] = getenv("PORTKEY_VIRTUAL_KEY")
+    portkey_api_key: Optional[str] = field(default_factory=lambda: getenv("PORTKEY_API_KEY"))
+    virtual_key: Optional[str] = field(default_factory=lambda: getenv("PORTKEY_VIRTUAL_KEY"))
     config: Optional[Dict[str, Any]] = None
     base_url: str = PORTKEY_GATEWAY_URL
 

agno/models/response.py

@@ -3,7 +3,7 @@ from enum import Enum
 from time import time
 from typing import Any, Dict, List, Optional
 
-from agno.media import Audio, Image, Video
+from agno.media import Audio, File, Image, Video
 from agno.models.message import Citations
 from agno.models.metrics import Metrics
 from agno.tools.function import UserInputField
@@ -98,6 +98,7 @@ class ModelResponse:
     images: Optional[List[Image]] = None
     videos: Optional[List[Video]] = None
     audios: Optional[List[Audio]] = None
+    files: Optional[List[File]] = None
 
     # Model tool calls
     tool_calls: List[Dict[str, Any]] = field(default_factory=list)

agno/models/sambanova/sambanova.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,7 +22,7 @@ class Sambanova(OpenAILike):
     name: str = "Sambanova"
     provider: str = "Sambanova"
 
-    api_key: Optional[str] = getenv("SAMBANOVA_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("SAMBANOVA_API_KEY"))
     base_url: str = "https://api.sambanova.ai/v1"
 
     supports_native_structured_outputs: bool = False

agno/models/together/together.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -21,5 +21,5 @@ class Together(OpenAILike):
     id: str = "mistralai/Mixtral-8x7B-Instruct-v0.1"
     name: str = "Together"
     provider: str = "Together"
-    api_key: Optional[str] = getenv("TOGETHER_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("TOGETHER_API_KEY"))
     base_url: str = "https://api.together.xyz/v1"

agno/models/vercel/v0.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Optional
 
@@ -22,5 +22,5 @@ class V0(OpenAILike):
     name: str = "v0"
     provider: str = "Vercel"
 
-    api_key: Optional[str] = getenv("V0_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("V0_API_KEY"))
     base_url: str = "https://api.v0.dev/v1/"

agno/models/xai/xai.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from os import getenv
 from typing import Any, Dict, List, Optional, Type, Union
 
@@ -34,7 +34,7 @@ class xAI(OpenAILike):
     name: str = "xAI"
     provider: str = "xAI"
 
-    api_key: Optional[str] = getenv("XAI_API_KEY")
+    api_key: Optional[str] = field(default_factory=lambda: getenv("XAI_API_KEY"))
     base_url: str = "https://api.x.ai/v1"
 
     search_parameters: Optional[Dict[str, Any]] = None