nvidia-nat-langchain 1.3.0.dev2__py3-none-any.whl → 1.4.0a20251216__py3-none-any.whl

nat/meta/pypi.md

@@ -18,6 +18,6 @@ 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 Subpackage
-This is a subpackage for LangChain and LangGraph integration in NeMo Agent toolkit.
+This is a subpackage for LangChain/LangGraph integration in NeMo Agent toolkit.
 
 For more information about the NVIDIA NeMo Agent toolkit, please visit the [NeMo Agent toolkit GitHub Repo](https://github.com/NVIDIA/NeMo-Agent-Toolkit).

nat/plugins/langchain/embedder.py

@@ -13,8 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# pylint: disable=unused-argument
-
 from nat.builder.builder import Builder
 from nat.builder.framework_enum import LLMFrameworkEnum
 from nat.cli.register_workflow import register_embedder_client
@@ -30,7 +28,13 @@ async def azure_openai_langchain(embedder_config: AzureOpenAIEmbedderModelConfig
 
     from langchain_openai import AzureOpenAIEmbeddings
 
-    client = AzureOpenAIEmbeddings(**embedder_config.model_dump(exclude={"type"}, by_alias=True))
+    client = AzureOpenAIEmbeddings(
+        **embedder_config.model_dump(exclude={"type", "api_version"},
+                                     by_alias=True,
+                                     exclude_none=True,
+                                     exclude_unset=True),
+        api_version=embedder_config.api_version,
+    )
 
     if isinstance(embedder_config, RetryMixin):
         client = patch_with_retry(client,
@@ -46,7 +50,8 @@ async def nim_langchain(embedder_config: NIMEmbedderModelConfig, builder: Builde
 
     from langchain_nvidia_ai_endpoints import NVIDIAEmbeddings
 
-    client = NVIDIAEmbeddings(**embedder_config.model_dump(exclude={"type"}, by_alias=True))
+    client = NVIDIAEmbeddings(
+        **embedder_config.model_dump(exclude={"type"}, by_alias=True, exclude_none=True, exclude_unset=True))
 
     if isinstance(embedder_config, RetryMixin):
         client = patch_with_retry(client,
@@ -62,7 +67,8 @@ async def openai_langchain(embedder_config: OpenAIEmbedderModelConfig, builder:
 
     from langchain_openai import OpenAIEmbeddings
 
-    client = OpenAIEmbeddings(**embedder_config.model_dump(exclude={"type"}, by_alias=True))
+    client = OpenAIEmbeddings(
+        **embedder_config.model_dump(exclude={"type"}, by_alias=True, exclude_none=True, exclude_unset=True))
 
     if isinstance(embedder_config, RetryMixin):
         client = patch_with_retry(client,

nat/plugins/langchain/llm.py

@@ -12,24 +12,82 @@
 # 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.
+# pylint: disable=unused-argument
+
+import logging
+from collections.abc import Sequence
+from typing import TypeVar
 
 from nat.builder.builder import Builder
 from nat.builder.framework_enum import LLMFrameworkEnum
 from nat.cli.register_workflow import register_llm_client
+from nat.data_models.llm import APITypeEnum
+from nat.data_models.llm import LLMBaseConfig
 from nat.data_models.retry_mixin import RetryMixin
+from nat.data_models.thinking_mixin import ThinkingMixin
 from nat.llm.aws_bedrock_llm import AWSBedrockModelConfig
 from nat.llm.azure_openai_llm import AzureOpenAIModelConfig
+from nat.llm.litellm_llm import LiteLlmModelConfig
 from nat.llm.nim_llm import NIMModelConfig
 from nat.llm.openai_llm import OpenAIModelConfig
+from nat.llm.utils.thinking import BaseThinkingInjector
+from nat.llm.utils.thinking import FunctionArgumentWrapper
+from nat.llm.utils.thinking import patch_with_thinking
 from nat.utils.exception_handlers.automatic_retries import patch_with_retry
+from nat.utils.responses_api import validate_no_responses_api
+from nat.utils.type_utils import override
 
+logger = logging.getLogger(__name__)
 
-@register_llm_client(config_type=AWSBedrockModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
-async def aws_bedrock_langchain(llm_config: AWSBedrockModelConfig, _builder: Builder):
+ModelType = TypeVar("ModelType")
 
-    from langchain_aws import ChatBedrockConverse
 
-    client = ChatBedrockConverse(**llm_config.model_dump(exclude={"type", "context_size"}, by_alias=True))
+def _patch_llm_based_on_config(client: ModelType, llm_config: LLMBaseConfig) -> ModelType:
+
+    from langchain_core.language_models import LanguageModelInput
+    from langchain_core.messages import BaseMessage
+    from langchain_core.messages import HumanMessage
+    from langchain_core.messages import SystemMessage
+    from langchain_core.prompt_values import PromptValue
+
+    class LangchainThinkingInjector(BaseThinkingInjector):
+
+        @override
+        def inject(self, messages: LanguageModelInput, *args, **kwargs) -> FunctionArgumentWrapper:
+            """
+            Inject a system prompt into the messages.
+
+            The messages are the first (non-object) argument to the function.
+            The rest of the arguments are passed through unchanged.
+
+            Args:
+                messages: The messages to inject the system prompt into.
+                *args: The rest of the arguments to the function.
+                **kwargs: The rest of the keyword arguments to the function.
+
+            Returns:
+                FunctionArgumentWrapper: An object that contains the transformed args and kwargs.
+
+            Raises:
+                ValueError: If the messages are not a valid type for LanguageModelInput.
+            """
+            if isinstance(messages, PromptValue):
+                messages = messages.to_messages()
+            elif isinstance(messages, str):
+                messages = [HumanMessage(content=messages)]
+
+            if isinstance(messages, Sequence) and all(isinstance(m, BaseMessage) for m in messages):
+                for i, message in enumerate(messages):
+                    if isinstance(message, SystemMessage):
+                        if self.system_prompt not in str(message.content):
+                            messages = list(messages)
+                            messages[i] = SystemMessage(content=f"{message.content}\n{self.system_prompt}")
+                        break
+                else:
+                    messages = list(messages)
+                    messages.insert(0, SystemMessage(content=self.system_prompt))
+                return FunctionArgumentWrapper(messages, *args, **kwargs)
+            raise ValueError(f"Unsupported message type: {type(messages)}")
 
     if isinstance(llm_config, RetryMixin):
         client = patch_with_retry(client,
@@ -37,7 +95,37 @@ async def aws_bedrock_langchain(llm_config: AWSBedrockModelConfig, _builder: Bui
                                   retry_codes=llm_config.retry_on_status_codes,
                                   retry_on_messages=llm_config.retry_on_errors)
 
-    yield client
+    if isinstance(llm_config, ThinkingMixin) and llm_config.thinking_system_prompt is not None:
+        client = patch_with_thinking(
+            client,
+            LangchainThinkingInjector(
+                system_prompt=llm_config.thinking_system_prompt,
+                function_names=[
+                    "invoke",
+                    "ainvoke",
+                    "stream",
+                    "astream",
+                ],
+            ))
+
+    return client
+
+
+@register_llm_client(config_type=AWSBedrockModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
+async def aws_bedrock_langchain(llm_config: AWSBedrockModelConfig, _builder: Builder):
+
+    from langchain_aws import ChatBedrockConverse
+
+    validate_no_responses_api(llm_config, LLMFrameworkEnum.LANGCHAIN)
+
+    client = ChatBedrockConverse(**llm_config.model_dump(
+        exclude={"type", "context_size", "thinking", "api_type"},
+        by_alias=True,
+        exclude_none=True,
+        exclude_unset=True,
+    ))
+
+    yield _patch_llm_based_on_config(client, llm_config)
 
 
 @register_llm_client(config_type=AzureOpenAIModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
@@ -45,15 +133,17 @@ async def azure_openai_langchain(llm_config: AzureOpenAIModelConfig, _builder: B
 
     from langchain_openai import AzureChatOpenAI
 
-    client = AzureChatOpenAI(**llm_config.model_dump(exclude={"type"}, by_alias=True))
+    validate_no_responses_api(llm_config, LLMFrameworkEnum.LANGCHAIN)
 
-    if isinstance(llm_config, RetryMixin):
-        client = patch_with_retry(client,
-                                  retries=llm_config.num_retries,
-                                  retry_codes=llm_config.retry_on_status_codes,
-                                  retry_on_messages=llm_config.retry_on_errors)
+    client = AzureChatOpenAI(
+        **llm_config.model_dump(exclude={"type", "thinking", "api_type", "api_version"},
+                                by_alias=True,
+                                exclude_none=True,
+                                exclude_unset=True),
+        api_version=llm_config.api_version,
+    )
 
-    yield client
+    yield _patch_llm_based_on_config(client, llm_config)
 
 
 @register_llm_client(config_type=NIMModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
@@ -61,15 +151,20 @@ async def nim_langchain(llm_config: NIMModelConfig, _builder: Builder):
 
     from langchain_nvidia_ai_endpoints import ChatNVIDIA
 
-    client = ChatNVIDIA(**llm_config.model_dump(exclude={"type"}, by_alias=True))
+    validate_no_responses_api(llm_config, LLMFrameworkEnum.LANGCHAIN)
 
-    if isinstance(llm_config, RetryMixin):
-        client = patch_with_retry(client,
-                                  retries=llm_config.num_retries,
-                                  retry_codes=llm_config.retry_on_status_codes,
-                                  retry_on_messages=llm_config.retry_on_errors)
+    # prefer max_completion_tokens over max_tokens
+    client = ChatNVIDIA(
+        **llm_config.model_dump(
+            exclude={"type", "max_tokens", "thinking", "api_type"},
+            by_alias=True,
+            exclude_none=True,
+            exclude_unset=True,
+        ),
+        max_completion_tokens=llm_config.max_tokens,
+    )
 
-    yield client
+    yield _patch_llm_based_on_config(client, llm_config)
 
 
 @register_llm_client(config_type=OpenAIModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
@@ -77,18 +172,37 @@ async def openai_langchain(llm_config: OpenAIModelConfig, _builder: Builder):
 
     from langchain_openai import ChatOpenAI
 
-    # Default kwargs for OpenAI to include usage metadata in the response. If the user has set stream_usage to False, we
-    # will not include this.
-    default_kwargs = {"stream_usage": True}
+    if llm_config.api_type == APITypeEnum.RESPONSES:
+        client = ChatOpenAI(stream_usage=True,
+                            use_responses_api=True,
+                            use_previous_response_id=True,
+                            **llm_config.model_dump(
+                                exclude={"type", "thinking", "api_type"},
+                                by_alias=True,
+                                exclude_none=True,
+                                exclude_unset=True,
+                            ))
+    else:
+        # If stream_usage is specified, it will override the default value of True.
+        client = ChatOpenAI(stream_usage=True,
+                            **llm_config.model_dump(
+                                exclude={"type", "thinking", "api_type"},
+                                by_alias=True,
+                                exclude_none=True,
+                                exclude_unset=True,
+                            ))
 
-    kwargs = {**default_kwargs, **llm_config.model_dump(exclude={"type"}, by_alias=True)}
+    yield _patch_llm_based_on_config(client, llm_config)
 
-    client = ChatOpenAI(**kwargs)
 
-    if isinstance(llm_config, RetryMixin):
-        client = patch_with_retry(client,
-                                  retries=llm_config.num_retries,
-                                  retry_codes=llm_config.retry_on_status_codes,
-                                  retry_on_messages=llm_config.retry_on_errors)
+@register_llm_client(config_type=LiteLlmModelConfig, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
+async def litellm_langchain(llm_config: LiteLlmModelConfig, _builder: Builder):
+
+    from langchain_litellm import ChatLiteLLM
+
+    validate_no_responses_api(llm_config, LLMFrameworkEnum.LANGCHAIN)
+
+    client = ChatLiteLLM(**llm_config.model_dump(
+        exclude={"type", "thinking", "api_type"}, by_alias=True, exclude_none=True, exclude_unset=True))
 
-    yield client
+    yield _patch_llm_based_on_config(client, llm_config)

nat/plugins/langchain/register.py

@@ -13,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# pylint: disable=unused-import
 # flake8: noqa
 # isort:skip_file
 

nat/plugins/langchain/tool_wrapper.py

@@ -24,7 +24,7 @@ logger = logging.getLogger(__name__)
 
 
 @register_tool_wrapper(wrapper_type=LLMFrameworkEnum.LANGCHAIN)
-def langchain_tool_wrapper(name: str, fn: Function, builder: Builder):  # pylint: disable=unused-argument
+def langchain_tool_wrapper(name: str, fn: Function, builder: Builder):
 
     import asyncio
 

nat/plugins/langchain/tools/code_generation_tool.py

@@ -61,6 +61,6 @@ You are a helpful code assistant that can teach a junior developer how to code.
         response = await tool.ainvoke({"question": query})
         if config.verbose:
             log.debug('Tool input was: %s\nTool output is: \n%s', query, response)
-        return response.content
+        return response.text()
 
     yield FunctionInfo.from_fn(_inner, description=config.description)

nat/plugins/langchain/tools/register.py

@@ -13,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# pylint: disable=unused-import
 # flake8: noqa
 # isort:skip_file
 

nat/plugins/langchain/tools/tavily_internet_search.py

@@ -13,9 +13,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from pydantic import Field
+
 from nat.builder.builder import Builder
 from nat.builder.function_info import FunctionInfo
 from nat.cli.register_workflow import register_function
+from nat.data_models.common import SerializableSecretStr
+from nat.data_models.common import get_secret_value
 from nat.data_models.function import FunctionBaseConfig
 
 
@@ -26,35 +30,40 @@ class TavilyInternetSearchToolConfig(FunctionBaseConfig, name="tavily_internet_s
     Requires a TAVILY_API_KEY.
     """
     max_results: int = 3
-    api_key: str = ""
+    api_key: SerializableSecretStr = Field(default="", description="The API key for the Tavily service.")
 
 
 @register_function(config_type=TavilyInternetSearchToolConfig)
 async def tavily_internet_search(tool_config: TavilyInternetSearchToolConfig, builder: Builder):
     import os
 
-    from langchain_community.tools import TavilySearchResults
+    from langchain_tavily import TavilySearch
 
     if not os.environ.get("TAVILY_API_KEY"):
-        os.environ["TAVILY_API_KEY"] = tool_config.api_key
+        if tool_config.api_key:
+            os.environ["TAVILY_API_KEY"] = get_secret_value(tool_config.api_key)
     # This tavily tool requires an API Key and it must be set as an environment variable (TAVILY_API_KEY)
     # Refer to create_customize_workflow.md for instructions of getting the API key
 
     async def _tavily_internet_search(question: str) -> str:
+        """This tool retrieves relevant contexts from web search (using Tavily) for the given question.
+
+        Args:
+            question (str): The question to be answered.
+
+        Returns:
+            str: The web search results.
+        """
         # Search the web and get the requested amount of results
-        tavily_search = TavilySearchResults(max_results=tool_config.max_results)
+        tavily_search = TavilySearch(max_results=tool_config.max_results)
         search_docs = await tavily_search.ainvoke({'query': question})
         # Format
         web_search_results = "\n\n---\n\n".join(
-            [f'<Document href="{doc["url"]}"/>\n{doc["content"]}\n</Document>' for doc in search_docs])
+            [f'<Document href="{doc["url"]}"/>\n{doc["content"]}\n</Document>' for doc in search_docs["results"]])
         return web_search_results
 
     # Create a Generic NAT tool that can be used with any supported LLM framework
     yield FunctionInfo.from_fn(
         _tavily_internet_search,
-        description=("""This tool retrieves relevant contexts from web search (using Tavily) for the given question.
-
-                        Args:
-                            question (str): The question to be answered.
-                    """),
+        description=_tavily_internet_search.__doc__,
     )

{nvidia_nat_langchain-1.3.0.dev2.dist-info → nvidia_nat_langchain-1.4.0a20251216.dist-info}/METADATA

@@ -1,19 +1,32 @@
 Metadata-Version: 2.4
 Name: nvidia-nat-langchain
-Version: 1.3.0.dev2
-Summary: Subpackage for LangChain and LangGraph integration in NeMo Agent toolkit
+Version: 1.4.0a20251216
+Summary: Subpackage for LangChain/LangGraph integration in 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
-Requires-Python: <3.13,>=3.11
+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
-Requires-Dist: nvidia-nat==v1.3.0-dev2
-Requires-Dist: langchain-aws~=0.2.1
-Requires-Dist: langchain-core~=0.3.7
-Requires-Dist: langchain-nvidia-ai-endpoints~=0.3.5
-Requires-Dist: langchain-milvus~=0.1.5
-Requires-Dist: langchain-openai~=0.3.5
-Requires-Dist: langgraph~=0.2.50
-Requires-Dist: langchain-milvus~=0.1.8
+License-File: LICENSE-3rd-party.txt
+License-File: LICENSE.md
+Requires-Dist: nvidia-nat==v1.4.0a20251216
+Requires-Dist: langchain~=0.3.27
+Requires-Dist: langchain-aws~=0.2.31
+Requires-Dist: langchain-core~=0.3.75
+Requires-Dist: langchain-litellm~=0.2.3
+Requires-Dist: langchain-milvus~=0.2.1
+Requires-Dist: langchain-nvidia-ai-endpoints~=0.3.17
+Requires-Dist: langchain-openai~=0.3.32
+Requires-Dist: langchain-tavily~=0.2.11
+Requires-Dist: langgraph~=0.6.7
+Dynamic: license-file
 
 <!--
 SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
@@ -35,6 +48,6 @@ 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 Subpackage
-This is a subpackage for LangChain and LangGraph integration in NeMo Agent toolkit.
+This is a subpackage for LangChain/LangGraph integration in NeMo Agent toolkit.
 
 For more information about the NVIDIA NeMo Agent toolkit, please visit the [NeMo Agent toolkit GitHub Repo](https://github.com/NVIDIA/NeMo-Agent-Toolkit).

nvidia_nat_langchain-1.4.0a20251216.dist-info/RECORD

@@ -0,0 +1,19 @@
+nat/meta/pypi.md,sha256=T_KFtTXVxhFM8Y6K3OlNByA5sTXLQuqqUpHgNOCvZBU,1120
+nat/plugins/langchain/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nat/plugins/langchain/embedder.py,sha256=Ie-J4N4lvygW0zNKklKZVSYxYFcRW6p_QlRdcz0WxcE,3607
+nat/plugins/langchain/llm.py,sha256=FaEPleu_aBCsQ-6tt3ofr1A2Oa_ta60tMb0yGGhiWSo,8561
+nat/plugins/langchain/register.py,sha256=jgq6wSJoGQIZFJhS8RbUs25cLgNJjCkFu4M6qaWJS_4,906
+nat/plugins/langchain/retriever.py,sha256=SWbXXOezEUuPACnmSSU497NAmEVEMj2SrFJGodkRg34,2644
+nat/plugins/langchain/tool_wrapper.py,sha256=Zgb2_XB4bEhjPPeqS-ZH_OJT_pcQmteX7u03N_qCLfc,2121
+nat/plugins/langchain/tools/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+nat/plugins/langchain/tools/code_generation_tool.py,sha256=f5pna0WMOx3QOS4WnaMFKD7tBZ1-tS0PfI0IMYobtTQ,2723
+nat/plugins/langchain/tools/register.py,sha256=uemxqLxcNk1bGX4crV52oMphLTZWonStzkXwTZeG2Rw,889
+nat/plugins/langchain/tools/tavily_internet_search.py,sha256=W5sdZ9hobPc3xbnWPSbtFBClIn14EM8xT0XUVF2HpWo,2928
+nat/plugins/langchain/tools/wikipedia_search.py,sha256=431YwLsjoC_mdvMZ_gY0Q37Uqaue2ASnAHpwr4jWCaU,2197
+nvidia_nat_langchain-1.4.0a20251216.dist-info/licenses/LICENSE-3rd-party.txt,sha256=fOk5jMmCX9YoKWyYzTtfgl-SUy477audFC5hNY4oP7Q,284609
+nvidia_nat_langchain-1.4.0a20251216.dist-info/licenses/LICENSE.md,sha256=QwcOLU5TJoTeUhuIXzhdCEEDDvorGiC6-3YTOl4TecE,11356
+nvidia_nat_langchain-1.4.0a20251216.dist-info/METADATA,sha256=RbqeB8fHZdx2VKxjL9gxJkgyeeu6njb_k7_8ES9H6F4,2263
+nvidia_nat_langchain-1.4.0a20251216.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+nvidia_nat_langchain-1.4.0a20251216.dist-info/entry_points.txt,sha256=4deXsMn97I012HhDw0UjoqcZ8eEoZ7BnqaRx5QmzebY,123
+nvidia_nat_langchain-1.4.0a20251216.dist-info/top_level.txt,sha256=8-CJ2cP6-f0ZReXe5Hzqp-5pvzzHz-5Ds5H2bGqh1-U,4
+nvidia_nat_langchain-1.4.0a20251216.dist-info/RECORD,,