nvidia-nat 1.3.0rc2__py3-none-any.whl → 1.3.0rc4__py3-none-any.whl

nat/front_ends/fastapi/message_handler.py

@@ -25,6 +25,7 @@ from pydantic import ValidationError
 from starlette.websockets import WebSocketDisconnect
 
 from nat.authentication.interfaces import FlowHandlerBase
+from nat.data_models.api_server import ChatRequest
 from nat.data_models.api_server import ChatResponse
 from nat.data_models.api_server import ChatResponseChunk
 from nat.data_models.api_server import Error
@@ -33,6 +34,8 @@ from nat.data_models.api_server import ResponsePayloadOutput
 from nat.data_models.api_server import ResponseSerializable
 from nat.data_models.api_server import SystemResponseContent
 from nat.data_models.api_server import TextContent
+from nat.data_models.api_server import UserMessageContentRoleType
+from nat.data_models.api_server import UserMessages
 from nat.data_models.api_server import WebSocketMessageStatus
 from nat.data_models.api_server import WebSocketMessageType
 from nat.data_models.api_server import WebSocketSystemInteractionMessage
@@ -64,12 +67,12 @@ class WebSocketMessageHandler:
         self._running_workflow_task: asyncio.Task | None = None
         self._message_parent_id: str = "default_id"
         self._conversation_id: str | None = None
-        self._workflow_schema_type: str = None
-        self._user_interaction_response: asyncio.Future[HumanResponse] | None = None
+        self._workflow_schema_type: str | None = None
+        self._user_interaction_response: asyncio.Future[TextContent] | None = None
 
         self._flow_handler: FlowHandlerBase | None = None
 
-        self._schema_output_mapping: dict[str, type[BaseModel] | None] = {
+        self._schema_output_mapping: dict[str, type[BaseModel] | type[None]] = {
             WorkflowSchemaType.GENERATE: self._session_manager.workflow.single_output_schema,
             WorkflowSchemaType.CHAT: ChatResponse,
             WorkflowSchemaType.CHAT_STREAM: ChatResponseChunk,
@@ -114,36 +117,58 @@ class WebSocketMessageHandler:
                     pass
 
                 elif (isinstance(validated_message, WebSocketUserInteractionResponseMessage)):
-                    user_content = await self.process_user_message_content(validated_message)
+                    user_content = await self._process_websocket_user_interaction_response_message(validated_message)
+                    assert self._user_interaction_response is not None
                     self._user_interaction_response.set_result(user_content)
             except (asyncio.CancelledError, WebSocketDisconnect):
                 # TODO: Handle the disconnect
                 break
 
-    async def process_user_message_content(
-            self, user_content: WebSocketUserMessage | WebSocketUserInteractionResponseMessage) -> BaseModel | None:
+    def _extract_last_user_message_content(self, messages: list[UserMessages]) -> TextContent:
         """
-        Processes the contents of a user message.
+        Extracts the last user's TextContent from a list of messages.
 
-        :param user_content: Incoming content data model.
-        :return: A validated Pydantic user content model or None if not found.
-        """
+        Args:
+            messages: List of UserMessages.
 
-        for user_message in user_content.content.messages[::-1]:
-            if (user_message.role == "user"):
+        Returns:
+            TextContent object from the last user message.
 
+        Raises:
+            ValueError: If no user text content is found.
+        """
+        for user_message in messages[::-1]:
+            if user_message.role == UserMessageContentRoleType.USER:
                 for attachment in user_message.content:
-
                     if isinstance(attachment, TextContent):
                         return attachment
+        raise ValueError("No user text content found in messages.")
+
+    async def _process_websocket_user_interaction_response_message(
+            self, user_content: WebSocketUserInteractionResponseMessage) -> TextContent:
+        """
+        Processes a WebSocketUserInteractionResponseMessage.
+        """
+        return self._extract_last_user_message_content(user_content.content.messages)
 
-        return None
+    async def _process_websocket_user_message(self, user_content: WebSocketUserMessage) -> ChatRequest | str:
+        """
+        Processes a WebSocketUserMessage based on schema type.
+        """
+        if self._workflow_schema_type in [WorkflowSchemaType.CHAT, WorkflowSchemaType.CHAT_STREAM]:
+            return ChatRequest(**user_content.content.model_dump(include={"messages"}))
+
+        elif self._workflow_schema_type in [WorkflowSchemaType.GENERATE, WorkflowSchemaType.GENERATE_STREAM]:
+            return self._extract_last_user_message_content(user_content.content.messages).text
+
+        raise ValueError("Unsupported workflow schema type for WebSocketUserMessage")
 
     async def process_workflow_request(self, user_message_as_validated_type: WebSocketUserMessage) -> None:
         """
         Process user messages and routes them appropriately.
 
-        :param user_message_as_validated_type: A WebSocketUserMessage Data Model instance.
+        Args:
+            user_message_as_validated_type (WebSocketUserMessage): The validated user message to process.
         """
 
         try:
@@ -151,18 +176,15 @@ class WebSocketMessageHandler:
             self._workflow_schema_type = user_message_as_validated_type.schema_type
             self._conversation_id = user_message_as_validated_type.conversation_id
 
-            content: BaseModel | None = await self.process_user_message_content(user_message_as_validated_type)
-
-            if content is None:
-                raise ValueError(f"User message content could not be found: {user_message_as_validated_type}")
+            message_content: typing.Any = await self._process_websocket_user_message(user_message_as_validated_type)
 
-            if isinstance(content, TextContent) and (self._running_workflow_task is None):
+            if (self._running_workflow_task is None):
 
-                def _done_callback(task: asyncio.Task):
+                def _done_callback(_task: asyncio.Task):
                     self._running_workflow_task = None
 
                 self._running_workflow_task = asyncio.create_task(
-                    self._run_workflow(payload=content.text,
+                    self._run_workflow(payload=message_content,
                                        user_message_id=self._message_parent_id,
                                        conversation_id=self._conversation_id,
                                        result_type=self._schema_output_mapping[self._workflow_schema_type],
@@ -180,13 +202,14 @@ class WebSocketMessageHandler:
     async def create_websocket_message(self,
                                        data_model: BaseModel,
                                        message_type: str | None = None,
-                                       status: str = WebSocketMessageStatus.IN_PROGRESS) -> None:
+                                       status: WebSocketMessageStatus = WebSocketMessageStatus.IN_PROGRESS) -> None:
         """
         Creates a websocket message that will be ready for routing based on message type or data model.
 
-        :param data_model: Message content model.
-        :param message_type: Message content model.
-        :param status: Message content model.
+        Args:
+            data_model (BaseModel): Message content model.
+            message_type (str | None): Message content model.
+            status (WebSocketMessageStatus): Message content model.
         """
         try:
             message: BaseModel | None = None
@@ -196,8 +219,8 @@ class WebSocketMessageHandler:
 
             message_schema: type[BaseModel] = await self._message_validator.get_message_schema_by_type(message_type)
 
-            if 'id' in data_model.model_fields:
-                message_id: str = data_model.id
+            if hasattr(data_model, 'id'):
+                message_id: str = str(getattr(data_model, 'id'))
             else:
                 message_id = str(uuid.uuid4())
 
@@ -253,12 +276,15 @@ class WebSocketMessageHandler:
         Registered human interaction callback that processes human interactions and returns
         responses from websocket connection.
 
-        :param prompt: Incoming interaction content data model.
-        :return: A Text Content Base Pydantic model.
+        Args:
+            prompt: Incoming interaction content data model.
+
+        Returns:
+            A Text Content Base Pydantic model.
         """
 
         # First create a future from the loop for the human response
-        human_response_future: asyncio.Future[HumanResponse] = asyncio.get_running_loop().create_future()
+        human_response_future: asyncio.Future[TextContent] = asyncio.get_running_loop().create_future()
 
         # Then add the future to the outstanding human prompts dictionary
         self._user_interaction_response = human_response_future
@@ -274,10 +300,10 @@ class WebSocketMessageHandler:
                 return HumanResponseNotification()
 
             # Wait for the human response future to complete
-            interaction_response: HumanResponse = await human_response_future
+            text_content: TextContent = await human_response_future
 
             interaction_response: HumanResponse = await self._message_validator.convert_text_content_to_human_response(
-                interaction_response, prompt.content)
+                text_content, prompt.content)
 
             return interaction_response
 
@@ -293,13 +319,12 @@ class WebSocketMessageHandler:
                             output_type: type | None = None) -> None:
 
         try:
-            async with self._session_manager.session(
-                    user_message_id=user_message_id,
-                    conversation_id=conversation_id,
-                    http_connection=self._socket,
-                    user_input_callback=self.human_interaction_callback,
-                    user_authentication_callback=(self._flow_handler.authenticate
-                                                  if self._flow_handler else None)) as session:
+            auth_callback = self._flow_handler.authenticate if self._flow_handler else None
+            async with self._session_manager.session(user_message_id=user_message_id,
+                                                     conversation_id=conversation_id,
+                                                     http_connection=self._socket,
+                                                     user_input_callback=self.human_interaction_callback,
+                                                     user_authentication_callback=auth_callback) as session:
 
                 async for value in generate_streaming_response(payload,
                                                                session_manager=session,

nat/front_ends/fastapi/message_validator.py

@@ -240,8 +240,7 @@ class MessageValidator:
         thread_id: str = "default",
         parent_id: str = "default",
         conversation_id: str | None = None,
-        content: SystemResponseContent
-        | Error = SystemResponseContent(),
+        content: SystemResponseContent | Error = SystemResponseContent(),
         status: WebSocketMessageStatus = WebSocketMessageStatus.IN_PROGRESS,
         timestamp: str = str(datetime.datetime.now(datetime.UTC))
     ) -> WebSocketSystemResponseTokenMessage | None:

nat/front_ends/mcp/mcp_front_end_config.py

@@ -13,13 +13,17 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import logging
 from typing import Literal
 
 from pydantic import Field
+from pydantic import model_validator
 
 from nat.authentication.oauth2.oauth2_resource_server_config import OAuth2ResourceServerConfig
 from nat.data_models.front_end import FrontEndBaseConfig
 
+logger = logging.getLogger(__name__)
+
 
 class MCPFrontEndConfig(FrontEndBaseConfig, name="mcp"):
     """MCP front end configuration.
@@ -43,3 +47,31 @@ class MCPFrontEndConfig(FrontEndBaseConfig, name="mcp"):
 
     server_auth: OAuth2ResourceServerConfig | None = Field(
         default=None, description=("OAuth 2.0 Resource Server configuration for token verification."))
+
+    @model_validator(mode="after")
+    def validate_security_configuration(self):
+        """Validate security configuration to prevent accidental misconfigurations."""
+        # Check if server is bound to a non-localhost interface without authentication
+        localhost_hosts = {"localhost", "127.0.0.1", "::1"}
+        if self.host not in localhost_hosts and self.server_auth is None:
+            logger.warning(
+                "MCP server is configured to bind to '%s' without authentication. "
+                "This may expose your server to unauthorized access. "
+                "Consider either: (1) binding to localhost for local-only access, "
+                "or (2) configuring server_auth for production deployments on public interfaces.",
+                self.host)
+
+        # Check if SSE transport is used (which doesn't support authentication)
+        if self.transport == "sse":
+            if self.server_auth is not None:
+                logger.warning("SSE transport does not support authentication. "
+                               "The configured server_auth will be ignored. "
+                               "For production use with authentication, use 'streamable-http' transport instead.")
+            elif self.host not in localhost_hosts:
+                logger.warning(
+                    "SSE transport does not support authentication and is bound to '%s'. "
+                    "This configuration is not recommended for production use. "
+                    "For production deployments, use 'streamable-http' transport with server_auth configured.",
+                    self.host)
+
+        return self

nat/observability/register.py

@@ -77,6 +77,14 @@ async def console_logging_method(config: ConsoleLoggingMethodConfig, builder: Bu
     level = getattr(logging, config.level.upper(), logging.INFO)
     handler = logging.StreamHandler(stream=sys.stdout)
     handler.setLevel(level)
+
+    # Set formatter to match the default CLI format
+    formatter = logging.Formatter(
+        fmt="%(asctime)s - %(levelname)-8s - %(name)s:%(lineno)d - %(message)s",
+        datefmt="%Y-%m-%d %H:%M:%S",
+    )
+    handler.setFormatter(formatter)
+
     yield handler
 
 
@@ -95,4 +103,12 @@ async def file_logging_method(config: FileLoggingMethod, builder: Builder):
     level = getattr(logging, config.level.upper(), logging.INFO)
     handler = logging.FileHandler(filename=config.path, mode="a", encoding="utf-8")
     handler.setLevel(level)
+
+    # Set formatter to match the default CLI format
+    formatter = logging.Formatter(
+        fmt="%(asctime)s - %(levelname)-8s - %(name)s:%(lineno)d - %(message)s",
+        datefmt="%Y-%m-%d %H:%M:%S",
+    )
+    handler.setFormatter(formatter)
+
     yield handler

nat/runtime/runner.py

@@ -196,8 +196,7 @@ class Runner:
 
             return result
         except Exception as e:
-            err_msg = f": {e}" if str(e).strip() else "."
-            logger.error("Error running workflow%s", err_msg)
+            logger.error("Error running workflow: %s", e)
             event_stream = self._context_state.event_stream.get()
             if event_stream:
                 event_stream.on_complete()

nat/runtime/session.py

@@ -192,7 +192,7 @@ class SessionManager:
                                     user_message_id: str | None,
                                     conversation_id: str | None) -> None:
         """
-        Extracts and sets user metadata for Websocket connections.
+        Extracts and sets user metadata for WebSocket connections.
         """
 
         # Extract cookies from WebSocket headers (similar to HTTP request)

nat/tool/memory_tools/add_memory_tool.py

@@ -30,10 +30,10 @@ logger = logging.getLogger(__name__)
 class AddToolConfig(FunctionBaseConfig, name="add_memory"):
     """Function to add memory to a hosted memory platform."""
 
-    description: str = Field(default=("Tool to add memory about a user's interactions to a system "
+    description: str = Field(default=("Tool to add a memory about a user's interactions to a system "
                                       "for retrieval later."),
                              description="The description of this function's use for tool calling agents.")
-    memory: MemoryRef = Field(default="saas_memory",
+    memory: MemoryRef = Field(default=MemoryRef("saas_memory"),
                               description=("Instance name of the memory client instance from the workflow "
                                            "configuration object."))
 
@@ -46,7 +46,7 @@ async def add_memory_tool(config: AddToolConfig, builder: Builder):
     from langchain_core.tools import ToolException
 
     # First, retrieve the memory client
-    memory_editor = builder.get_memory_client(config.memory)
+    memory_editor = await builder.get_memory_client(config.memory)
 
     async def _arun(item: MemoryItem) -> str:
         """

nat/tool/memory_tools/delete_memory_tool.py

@@ -30,10 +30,9 @@ logger = logging.getLogger(__name__)
 class DeleteToolConfig(FunctionBaseConfig, name="delete_memory"):
     """Function to delete memory from a hosted memory platform."""
 
-    description: str = Field(default=("Tool to retrieve memory about a user's "
-                                      "interactions to help answer questions in a personalized way."),
+    description: str = Field(default="Tool to delete a memory from a hosted memory platform.",
                              description="The description of this function's use for tool calling agents.")
-    memory: MemoryRef = Field(default="saas_memory",
+    memory: MemoryRef = Field(default=MemoryRef("saas_memory"),
                               description=("Instance name of the memory client instance from the workflow "
                                            "configuration object."))
 
@@ -47,7 +46,7 @@ async def delete_memory_tool(config: DeleteToolConfig, builder: Builder):
     from langchain_core.tools import ToolException
 
     # First, retrieve the memory client
-    memory_editor = builder.get_memory_client(config.memory)
+    memory_editor = await builder.get_memory_client(config.memory)
 
     async def _arun(user_id: str) -> str:
         """

nat/tool/memory_tools/get_memory_tool.py

@@ -30,10 +30,10 @@ logger = logging.getLogger(__name__)
 class GetToolConfig(FunctionBaseConfig, name="get_memory"):
     """Function to get memory to a hosted memory platform."""
 
-    description: str = Field(default=("Tool to retrieve memory about a user's "
+    description: str = Field(default=("Tool to retrieve a memory about a user's "
                                       "interactions to help answer questions in a personalized way."),
                              description="The description of this function's use for tool calling agents.")
-    memory: MemoryRef = Field(default="saas_memory",
+    memory: MemoryRef = Field(default=MemoryRef("saas_memory"),
                               description=("Instance name of the memory client instance from the workflow "
                                            "configuration object."))
 
@@ -49,7 +49,7 @@ async def get_memory_tool(config: GetToolConfig, builder: Builder):
     from langchain_core.tools import ToolException
 
     # First, retrieve the memory client
-    memory_editor = builder.get_memory_client(config.memory)
+    memory_editor = await builder.get_memory_client(config.memory)
 
     async def _arun(search_input: SearchMemoryInput) -> str:
         """

nat/utils/type_converter.py

@@ -93,6 +93,14 @@ class TypeConverter:
         if to_type is None or decomposed.is_instance(data):
             return data
 
+        # 2) If data is a union type, try to convert to each type in the union
+        if decomposed.is_union:
+            for union_type in decomposed.args:
+                result = self._convert(data, union_type)
+                if result is not None:
+                    return result
+            return None
+
         root = decomposed.root
 
         # 2) Attempt direct in *this* converter

nvidia_nat-1.3.0rc4.dist-info/METADATA

@@ -0,0 +1,195 @@
+Metadata-Version: 2.4
+Name: nvidia-nat
+Version: 1.3.0rc4
+Summary: NVIDIA NeMo Agent toolkit
+Author: NVIDIA Corporation
+Maintainer: NVIDIA Corporation
+License: Apache-2.0
+Project-URL: documentation, https://docs.nvidia.com/nemo/agent-toolkit/latest/
+Project-URL: source, https://github.com/NVIDIA/NeMo-Agent-Toolkit
+Keywords: ai,rag,agents
+Classifier: Programming Language :: Python
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Requires-Python: <3.14,>=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE-3rd-party.txt
+License-File: LICENSE.md
+Requires-Dist: aioboto3>=11.0.0
+Requires-Dist: authlib~=1.5
+Requires-Dist: click~=8.1
+Requires-Dist: colorama~=0.4.6
+Requires-Dist: datasets~=4.0
+Requires-Dist: expandvars~=1.0
+Requires-Dist: fastapi~=0.115.5
+Requires-Dist: httpx~=0.27
+Requires-Dist: jinja2~=3.1
+Requires-Dist: jsonpath-ng~=1.7
+Requires-Dist: mcp~=1.13
+Requires-Dist: nest-asyncio~=1.6
+Requires-Dist: networkx~=3.4
+Requires-Dist: numpy~=1.26; python_version < "3.12"
+Requires-Dist: numpy~=2.3; python_version >= "3.12"
+Requires-Dist: openinference-semantic-conventions~=0.1.14
+Requires-Dist: openpyxl~=3.1
+Requires-Dist: optuna~=4.4.0
+Requires-Dist: pip>=24.3.1
+Requires-Dist: pkce==1.0.3
+Requires-Dist: pkginfo~=1.12
+Requires-Dist: platformdirs~=4.3
+Requires-Dist: pydantic~=2.11
+Requires-Dist: pymilvus~=2.4
+Requires-Dist: python-dotenv~=1.1.1
+Requires-Dist: PyYAML~=6.0
+Requires-Dist: ragas~=0.2.14
+Requires-Dist: rich~=13.9
+Requires-Dist: tabulate~=0.9
+Requires-Dist: uvicorn[standard]<0.36
+Requires-Dist: wikipedia~=1.4
+Provides-Extra: all
+Requires-Dist: nvidia-nat-all; extra == "all"
+Provides-Extra: adk
+Requires-Dist: nvidia-nat-adk; extra == "adk"
+Provides-Extra: agno
+Requires-Dist: nvidia-nat-agno; extra == "agno"
+Provides-Extra: crewai
+Requires-Dist: nvidia-nat-crewai; extra == "crewai"
+Provides-Extra: data-flywheel
+Requires-Dist: nvidia-nat-data-flywheel; extra == "data-flywheel"
+Provides-Extra: ingestion
+Requires-Dist: nvidia-nat-ingestion; extra == "ingestion"
+Provides-Extra: langchain
+Requires-Dist: nvidia-nat-langchain; extra == "langchain"
+Provides-Extra: llama-index
+Requires-Dist: nvidia-nat-llama-index; extra == "llama-index"
+Provides-Extra: mcp
+Requires-Dist: nvidia-nat-mcp; extra == "mcp"
+Provides-Extra: mem0ai
+Requires-Dist: nvidia-nat-mem0ai; extra == "mem0ai"
+Provides-Extra: opentelemetry
+Requires-Dist: nvidia-nat-opentelemetry; extra == "opentelemetry"
+Provides-Extra: phoenix
+Requires-Dist: nvidia-nat-phoenix; extra == "phoenix"
+Provides-Extra: profiling
+Requires-Dist: nvidia-nat-profiling; extra == "profiling"
+Provides-Extra: ragaai
+Requires-Dist: nvidia-nat-ragaai; extra == "ragaai"
+Provides-Extra: mysql
+Requires-Dist: nvidia-nat-mysql; extra == "mysql"
+Provides-Extra: redis
+Requires-Dist: nvidia-nat-redis; extra == "redis"
+Provides-Extra: s3
+Requires-Dist: nvidia-nat-s3; extra == "s3"
+Provides-Extra: semantic-kernel
+Requires-Dist: nvidia-nat-semantic-kernel; extra == "semantic-kernel"
+Provides-Extra: telemetry
+Requires-Dist: nvidia-nat-opentelemetry; extra == "telemetry"
+Requires-Dist: nvidia-nat-phoenix; extra == "telemetry"
+Requires-Dist: nvidia-nat-weave; extra == "telemetry"
+Requires-Dist: nvidia-nat-ragaai; extra == "telemetry"
+Provides-Extra: weave
+Requires-Dist: nvidia-nat-weave; extra == "weave"
+Provides-Extra: zep-cloud
+Requires-Dist: nvidia-nat-zep-cloud; extra == "zep-cloud"
+Provides-Extra: examples
+Requires-Dist: nat_adk_demo; extra == "examples"
+Requires-Dist: nat_agno_personal_finance; extra == "examples"
+Requires-Dist: nat_haystack_deep_research_agent; extra == "examples"
+Requires-Dist: nat_alert_triage_agent; extra == "examples"
+Requires-Dist: nat_automated_description_generation; extra == "examples"
+Requires-Dist: nat_email_phishing_analyzer; extra == "examples"
+Requires-Dist: nat_multi_frameworks; extra == "examples"
+Requires-Dist: nat_plot_charts; extra == "examples"
+Requires-Dist: nat_por_to_jiratickets; extra == "examples"
+Requires-Dist: nat_profiler_agent; extra == "examples"
+Requires-Dist: nat_redact_pii; extra == "examples"
+Requires-Dist: nat_router_agent; extra == "examples"
+Requires-Dist: nat_semantic_kernel_demo; extra == "examples"
+Requires-Dist: nat_sequential_executor; extra == "examples"
+Requires-Dist: nat_simple_auth; extra == "examples"
+Requires-Dist: nat_simple_auth_mcp; extra == "examples"
+Requires-Dist: nat_simple_web_query; extra == "examples"
+Requires-Dist: nat_simple_web_query_eval; extra == "examples"
+Requires-Dist: nat_simple_calculator; extra == "examples"
+Requires-Dist: nat_simple_calculator_custom_routes; extra == "examples"
+Requires-Dist: nat_simple_calculator_eval; extra == "examples"
+Requires-Dist: nat_simple_calculator_mcp; extra == "examples"
+Requires-Dist: nat_simple_calculator_observability; extra == "examples"
+Requires-Dist: nat_simple_calculator_hitl; extra == "examples"
+Requires-Dist: nat_simple_rag; extra == "examples"
+Requires-Dist: nat_swe_bench; extra == "examples"
+Requires-Dist: nat_user_report; extra == "examples"
+Provides-Extra: gunicorn
+Requires-Dist: gunicorn~=23.0; extra == "gunicorn"
+Provides-Extra: async-endpoints
+Requires-Dist: aiosqlite~=0.21; extra == "async-endpoints"
+Requires-Dist: dask==2023.6; extra == "async-endpoints"
+Requires-Dist: distributed==2023.6; extra == "async-endpoints"
+Requires-Dist: sqlalchemy[asyncio]~=2.0; extra == "async-endpoints"
+Provides-Extra: nvidia-haystack
+Requires-Dist: nvidia-haystack~=0.3.0; extra == "nvidia-haystack"
+Provides-Extra: litellm
+Requires-Dist: litellm==1.74.9; extra == "litellm"
+Provides-Extra: openai
+Requires-Dist: openai~=1.106; extra == "openai"
+Dynamic: license-file
+
+<!--
+SPDX-FileCopyrightText: Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+SPDX-License-Identifier: Apache-2.0
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+![NVIDIA NeMo Agent Toolkit](https://media.githubusercontent.com/media/NVIDIA/NeMo-Agent-Toolkit/refs/heads/main/docs/source/_static/banner.png "NeMo Agent toolkit banner image")
+
+# NVIDIA NeMo Agent Toolkit
+
+NeMo Agent toolkit is a flexible library designed to seamlessly integrate your enterprise agents—regardless of framework—with various data sources and tools. By treating agents, tools, and agentic workflows as simple function calls, NeMo Agent toolkit enables true composability: build once and reuse anywhere.
+
+## Key Features
+
+- [**Framework Agnostic:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/extend/plugins.html) Works with any agentic framework, so you can use your current technology stack without replatforming.
+- [**Reusability:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/extend/sharing-components.html) Every agent, tool, or workflow can be combined and repurposed, allowing developers to leverage existing work in new scenarios.
+- [**Rapid Development:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/tutorials/index.html) Start with a pre-built agent, tool, or workflow, and customize it to your needs.
+- [**Profiling:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/profiler.html) Profile entire workflows down to the tool and agent level, track input/output tokens and timings, and identify bottlenecks.
+- [**Observability:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/observe/observe-workflow-with-phoenix.html) Monitor and debug your workflows with any OpenTelemetry-compatible observability tool, with examples using [Phoenix](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/observe/observe-workflow-with-phoenix.html) and [W&B Weave](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/observe/observe-workflow-with-weave.html).
+- [**Evaluation System:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/evaluate.html) Validate and maintain accuracy of agentic workflows with built-in evaluation tools.
+- [**User Interface:**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/quick-start/launching-ui.html) Use the NeMo Agent toolkit UI chat interface to interact with your agents, visualize output, and debug workflows.
+- [**MCP Compatibility**](https://docs.nvidia.com/nemo/agent-toolkit/1.3/workflows/mcp/mcp-client.html) Compatible with Model Context Protocol (MCP), allowing tools served by MCP Servers to be used as NeMo Agent toolkit functions.
+
+With NeMo Agent toolkit, you can move quickly, experiment freely, and ensure reliability across all your agent-driven projects.
+
+## Links
+ * [Documentation](https://docs.nvidia.com/nemo/agent-toolkit/1.3/index.html): Explore the full documentation for NeMo Agent toolkit.
+
+## First time user?
+ If this is your first time using NeMo Agent toolkit, it is recommended to install the latest version from the [source repository](https://github.com/NVIDIA/NeMo-Agent-Toolkit?tab=readme-ov-file#quick-start) on GitHub. This package is intended for users who are familiar with NeMo Agent toolkit applications and need to add NeMo Agent toolkit as a dependency to their project.
+
+## Feedback
+
+We would love to hear from you! Please file an issue on [GitHub](https://github.com/NVIDIA/NeMo-Agent-Toolkit/issues) if you have any feedback or feature requests.
+
+## Acknowledgements
+
+We would like to thank the following open source projects that made NeMo Agent toolkit possible:
+
+- [CrewAI](https://github.com/crewAIInc/crewAI)
+- [FastAPI](https://github.com/tiangolo/fastapi)
+- [LangChain](https://github.com/langchain-ai/langchain)
+- [Llama-Index](https://github.com/run-llama/llama_index)
+- [Mem0ai](https://github.com/mem0ai/mem0)
+- [Ragas](https://github.com/explodinggradients/ragas)
+- [Semantic Kernel](https://github.com/microsoft/semantic-kernel)
+- [uv](https://github.com/astral-sh/uv)