nvidia-nat 1.2.0rc5__py3-none-any.whl

aiq/runtime/session.py

@@ -0,0 +1,158 @@
+# SPDX-FileCopyrightText: Copyright (c) 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.
+
+import asyncio
+import contextvars
+import typing
+from collections.abc import Awaitable
+from collections.abc import Callable
+from contextlib import asynccontextmanager
+from contextlib import nullcontext
+
+from starlette.requests import HTTPConnection
+
+from aiq.builder.context import AIQContext
+from aiq.builder.context import AIQContextState
+from aiq.builder.workflow import Workflow
+from aiq.data_models.authentication import AuthenticatedContext
+from aiq.data_models.authentication import AuthFlowType
+from aiq.data_models.authentication import AuthProviderBaseConfig
+from aiq.data_models.config import AIQConfig
+from aiq.data_models.interactive import HumanResponse
+from aiq.data_models.interactive import InteractionPrompt
+
+_T = typing.TypeVar("_T")
+
+
+class UserManagerBase:
+    pass
+
+
+class AIQSessionManager:
+
+    def __init__(self, workflow: Workflow, max_concurrency: int = 8):
+        """
+        The AIQSessionManager class is used to run and manage a user workflow session. It runs and manages the context,
+        and configuration of a workflow with the specified concurrency.
+
+        Parameters
+        ----------
+        workflow : Workflow
+            The workflow to run
+        max_concurrency : int, optional
+            The maximum number of simultaneous workflow invocations, by default 8
+        """
+
+        if (workflow is None):
+            raise ValueError("Workflow cannot be None")
+
+        self._workflow: Workflow = workflow
+
+        self._max_concurrency = max_concurrency
+        self._context_state = AIQContextState.get()
+        self._context = AIQContext(self._context_state)
+
+        # We save the context because Uvicorn spawns a new process
+        # for each request, and we need to restore the context vars
+        self._saved_context = contextvars.copy_context()
+
+        if (max_concurrency > 0):
+            self._semaphore = asyncio.Semaphore(max_concurrency)
+        else:
+            # If max_concurrency is 0, then we don't need to limit the concurrency but we still need a context
+            self._semaphore = nullcontext()
+
+    @property
+    def config(self) -> AIQConfig:
+        return self._workflow.config
+
+    @property
+    def workflow(self) -> Workflow:
+        return self._workflow
+
+    @property
+    def context(self) -> AIQContext:
+        return self._context
+
+    @asynccontextmanager
+    async def session(self,
+                      user_manager=None,
+                      request: HTTPConnection | None = None,
+                      conversation_id: str | None = None,
+                      user_input_callback: Callable[[InteractionPrompt], Awaitable[HumanResponse]] = None,
+                      user_authentication_callback: Callable[[AuthProviderBaseConfig, AuthFlowType],
+                                                             Awaitable[AuthenticatedContext | None]] = None):
+
+        token_user_input = None
+        if user_input_callback is not None:
+            token_user_input = self._context_state.user_input_callback.set(user_input_callback)
+
+        token_user_manager = None
+        if user_manager is not None:
+            token_user_manager = self._context_state.user_manager.set(user_manager)
+
+        token_user_authentication = None
+        if user_authentication_callback is not None:
+            token_user_authentication = self._context_state.user_auth_callback.set(user_authentication_callback)
+
+        if conversation_id is not None and request is None:
+            self._context_state.conversation_id.set(conversation_id)
+
+        self.set_metadata_from_http_request(request)
+
+        try:
+            yield self
+        finally:
+            if token_user_manager is not None:
+                self._context_state.user_manager.reset(token_user_manager)
+            if token_user_input is not None:
+                self._context_state.user_input_callback.reset(token_user_input)
+            if token_user_authentication is not None:
+                self._context_state.user_auth_callback.reset(token_user_authentication)
+
+    @asynccontextmanager
+    async def run(self, message):
+        """
+        Start a workflow run
+        """
+        async with self._semaphore:
+            # Apply the saved context
+            for k, v in self._saved_context.items():
+                k.set(v)
+
+            async with self._workflow.run(message) as runner:
+                yield runner
+
+    def set_metadata_from_http_request(self, request: HTTPConnection | None) -> None:
+        """
+        Extracts and sets user metadata request attributes from a HTTP request.
+        If request is None, no attributes are set.
+        """
+        if request is None:
+            return
+
+        self._context.metadata._request.method = getattr(request, "method", None)
+        self._context.metadata._request.url_path = request.url.path
+        self._context.metadata._request.url_port = request.url.port
+        self._context.metadata._request.url_scheme = request.url.scheme
+        self._context.metadata._request.headers = request.headers
+        self._context.metadata._request.query_params = request.query_params
+        self._context.metadata._request.path_params = request.path_params
+        self._context.metadata._request.client_host = request.client.host
+        self._context.metadata._request.client_port = request.client.port
+        self._context.metadata._request.cookies = request.cookies
+
+        if request.headers.get("conversation-id"):
+            self._context_state.conversation_id.set(request.headers["conversation-id"])

aiq/runtime/user_metadata.py

@@ -0,0 +1,130 @@
+# 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.
+
+from starlette.datastructures import Headers
+from starlette.datastructures import QueryParams
+
+from aiq.data_models.api_server import Request
+
+
+class RequestAttributes:
+    """
+    The RequestAttributes class is responsible for managing user http and webscoket session
+    metadata. It provides a way to store and expose session attributes to workflow tools.
+    """
+
+    def __init__(self) -> None:
+        self._request: Request = Request()
+
+    @property
+    def method(self) -> str | None:
+        """
+        This property retrieves the HTTP method of the request.
+        It can be GET, POST, PUT, DELETE, etc.
+
+        Returns:
+            str | None
+        """
+        return self._request.method
+
+    @property
+    def url_path(self) -> str | None:
+        """
+        This property retrieves the path from the URL of the request.
+
+        Returns:
+            str | None
+        """
+        return self._request.url_path
+
+    @property
+    def url_port(self) -> int | None:
+        """
+        This property retrieves the port number from the URL of the request.
+
+        Returns:
+            int | None
+        """
+        return self._request.url_port
+
+    @property
+    def url_scheme(self) -> str | None:
+        """
+        This property retrieves the scheme from the URL of the request.
+
+        Returns:
+            str | None
+        """
+        return self._request.url_scheme
+
+    @property
+    def headers(self) -> Headers | None:
+        """
+        This property retrieves the headers from the request stored in a dictionary-like object.
+
+        Returns:
+            Headers | None
+        """
+        return self._request.headers
+
+    @property
+    def query_params(self) -> QueryParams | None:
+        """
+        This property retrieves the query parameters from the request stored in a dictionary-like object.
+
+        Returns:
+            QueryParams | None
+        """
+        return self._request.query_params
+
+    @property
+    def path_params(self) -> dict[str, str] | None:
+        """
+        This property retrieves the path parameters from the request stored in a dictionary-like object.
+
+        Returns:
+            dict[str, str] | None
+        """
+        return self._request.path_params
+
+    @property
+    def client_host(self) -> str | None:
+        """
+        This property retrieves the clients remote hostname or IP address.
+
+        Returns:
+            str | None
+        """
+        return self._request.client_host
+
+    @property
+    def client_port(self) -> int | None:
+        """
+        This property retrieves the clients remote port number from which the client is connecting to.
+
+        Returns:
+            int | None
+        """
+        return self._request.client_port
+
+    @property
+    def cookies(self) -> dict[str, str] | None:
+        """
+        This property retrieves the cookies from the request stored in a dictionary-like object.
+
+        Returns:
+            dict[str, str] | None
+        """
+        return self._request.cookies

aiq/settings/global_settings.py

@@ -0,0 +1,318 @@
+# 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.
+
+import json
+import logging
+import os
+import typing
+from collections.abc import Callable
+from contextlib import contextmanager
+from copy import deepcopy
+
+from platformdirs import user_config_dir
+from pydantic import ConfigDict
+from pydantic import Discriminator
+from pydantic import Tag
+from pydantic import ValidationError
+from pydantic import ValidationInfo
+from pydantic import ValidatorFunctionWrapHandler
+from pydantic import field_validator
+
+from aiq.cli.type_registry import GlobalTypeRegistry
+from aiq.cli.type_registry import RegisteredInfo
+from aiq.data_models.common import HashableBaseModel
+from aiq.data_models.common import TypedBaseModel
+from aiq.data_models.common import TypedBaseModelT
+from aiq.data_models.registry_handler import RegistryHandlerBaseConfig
+
+logger = logging.getLogger(__name__)
+
+
+class Settings(HashableBaseModel):
+
+    model_config = ConfigDict(extra="forbid")
+
+    # Registry Handeler Configuration
+    channels: dict[str, RegistryHandlerBaseConfig] = {}
+
+    _configuration_directory: typing.ClassVar[str]
+    _settings_changed_hooks: typing.ClassVar[list[Callable[[], None]]] = []
+    _settings_changed_hooks_active: bool = True
+
+    @field_validator("channels", mode="wrap")
+    @classmethod
+    def validate_components(cls, value: typing.Any, handler: ValidatorFunctionWrapHandler, info: ValidationInfo):
+
+        try:
+            return handler(value)
+        except ValidationError as err:
+
+            for e in err.errors():
+                if e['type'] == 'union_tag_invalid' and len(e['loc']) > 0:
+                    requested_type = e['loc'][0]
+
+                    if (info.field_name == "channels"):
+                        registered_keys = GlobalTypeRegistry.get().get_registered_registry_handlers()
+                    else:
+                        assert False, f"Unknown field name {info.field_name} in validator"
+
+                    # Check and see if the there are multiple full types which match this short type
+                    matching_keys = [k for k in registered_keys if k.local_name == requested_type]
+
+                    assert len(matching_keys) != 1, "Exact match should have been found. Contact developers"
+
+                    matching_key_names = [x.full_type for x in matching_keys]
+                    registered_key_names = [x.full_type for x in registered_keys]
+
+                    if (len(matching_keys) == 0):
+                        # This is a case where the requested type is not found. Show a helpful message about what is
+                        # available
+                        raise ValueError(
+                            f"Requested {info.field_name} type `{requested_type}` not found. "
+                            "Have you ensured the necessary package has been installed with `uv pip install`?"
+                            "\nAvailable {} names:\n - {}".format(info.field_name,
+                                                                  '\n - '.join(registered_key_names))) from err
+
+                    # This is a case where the requested type is ambiguous.
+                    raise ValueError(f"Requested {info.field_name} type `{requested_type}` is ambiguous. " +
+                                     f"Matched multiple {info.field_name} by their local name: {matching_key_names}. " +
+                                     f"Please use the fully qualified {info.field_name} name." +
+                                     "\nAvailable {} names:\n - {}".format(info.field_name,
+                                                                           '\n - '.join(registered_key_names))) from err
+
+            raise
+
+    @classmethod
+    def rebuild_annotations(cls):
+
+        def compute_annotation(cls: type[TypedBaseModelT], registrations: list[RegisteredInfo[TypedBaseModelT]]):
+
+            while (len(registrations) < 2):
+                registrations.append(RegisteredInfo[TypedBaseModelT](full_type=f"_ignore/{len(registrations)}",
+                                                                     config_type=cls))
+
+            short_names: dict[str, int] = {}
+            type_list: list[tuple[str, type[TypedBaseModelT]]] = []
+
+            # For all keys in the list, split the key by / and increment the count of the last element
+            for key in registrations:
+                short_names[key.local_name] = short_names.get(key.local_name, 0) + 1
+
+                type_list.append((key.full_type, key.config_type))
+
+            # Now loop again and if the short name is unique, then create two entries, for the short and full name
+            for key in registrations:
+
+                if (short_names[key.local_name] == 1):
+                    type_list.append((key.local_name, key.config_type))
+
+            # pylint: disable=consider-alternative-union-syntax
+            return typing.Union[tuple(typing.Annotated[x_type, Tag(x_id)] for x_id, x_type in type_list)]
+
+        RegistryHandlerAnnotation = dict[
+            str,
+            typing.Annotated[compute_annotation(RegistryHandlerBaseConfig,
+                                                GlobalTypeRegistry.get().get_registered_registry_handlers()),
+                             Discriminator(TypedBaseModel.discriminator)]]
+
+        should_rebuild = False
+
+        channels_field = cls.model_fields.get("channels")
+        if channels_field is not None and channels_field.annotation != RegistryHandlerAnnotation:
+            channels_field.annotation = RegistryHandlerAnnotation
+            should_rebuild = True
+
+        if (should_rebuild):
+            cls.model_rebuild(force=True)
+
+    @property
+    def channel_names(self) -> list:
+        return list(self.channels.keys())
+
+    @property
+    def configuration_directory(self) -> str:
+        return self._configuration_directory
+
+    @property
+    def configuration_file(self) -> str:
+        return os.path.join(self.configuration_directory, "config.json")
+
+    @staticmethod
+    def from_file():
+
+        configuration_directory = os.getenv("AIQ_CONFIG_DIR", user_config_dir(appname="aiq"))
+
+        if not os.path.exists(configuration_directory):
+            os.makedirs(configuration_directory, exist_ok=True)
+
+        configuration_file = os.path.join(configuration_directory, "config.json")
+
+        file_path = os.path.join(configuration_directory, "config.json")
+
+        if (not os.path.exists(configuration_file)):
+            loaded_config = {}
+        else:
+            with open(file_path, mode="r", encoding="utf-8") as f:
+                loaded_config = json.load(f)
+
+        settings = Settings(**loaded_config)
+        settings.set_configuration_directory(configuration_directory)
+        return settings
+
+    def set_configuration_directory(self, directory: str, remove: bool = False) -> None:
+        if (remove):
+            if os.path.exists(self.configuration_directory):
+                os.rmdir(self.configuration_directory)
+        self.__class__._configuration_directory = directory
+
+    def reset_configuration_directory(self, remove: bool = False) -> None:
+        if (remove):
+            if os.path.exists(self.configuration_directory):
+                os.rmdir(self.configuration_directory)
+        self._configuration_directory = os.getenv("AIQ_CONFIG_DIR", user_config_dir(appname="aiq"))
+
+    def _save_settings(self) -> None:
+
+        if not os.path.exists(self.configuration_directory):
+            os.mkdir(self.configuration_directory)
+
+        with open(self.configuration_file, mode="w", encoding="utf-8") as f:
+            f.write(self.model_dump_json(indent=4, by_alias=True, serialize_as_any=True))
+
+        self._settings_changed()
+
+    def update_settings(self, config_obj: "dict | Settings"):
+        self._update_settings(config_obj)
+
+    def _update_settings(self, config_obj: "dict | Settings"):
+
+        if isinstance(config_obj, Settings):
+            config_obj = config_obj.model_dump(serialize_as_any=True, by_alias=True)
+
+        self._revalidate(config_dict=config_obj)
+
+        self._save_settings()
+
+    def _revalidate(self, config_dict) -> bool:
+
+        try:
+            validated_data = self.__class__(**config_dict)
+
+            for field in validated_data.model_fields:
+                match field:
+                    case "channels":
+                        self.channels = validated_data.channels
+                    case _:
+                        raise ValueError(f"Encountered invalid model field: {field}")
+
+            return True
+
+        except Exception as e:
+            logger.exception("Unable to validate user settings configuration: %s", e, exc_info=True)
+            return False
+
+    def print_channel_settings(self, channel_type: str | None = None) -> None:
+
+        import yaml
+
+        remote_channels = self.model_dump(serialize_as_any=True, by_alias=True)
+
+        if (not remote_channels or not remote_channels.get("channels")):
+            logger.warning("No configured channels to list.")
+            return
+
+        if (channel_type is not None):
+            filter_channels = []
+            for channel, settings in remote_channels.items():
+                if (settings["type"] != channel_type):
+                    filter_channels.append(channel)
+            for channel in filter_channels:
+                del remote_channels[channel]
+
+        if (remote_channels):
+            logger.info(yaml.dump(remote_channels, allow_unicode=True, default_flow_style=False))
+
+    def override_settings(self, config_file: str) -> "Settings":
+
+        from aiq.utils.io.yaml_tools import yaml_load
+
+        override_settings_dict = yaml_load(config_file)
+
+        settings_dict = self.model_dump()
+        updated_settings = {**override_settings_dict, **settings_dict}
+        self._update_settings(config_obj=updated_settings)
+
+        return self
+
+    def _settings_changed(self):
+
+        if (not self._settings_changed_hooks_active):
+            return
+
+        for hook in self._settings_changed_hooks:
+            hook()
+
+    @contextmanager
+    def pause_settings_changed_hooks(self):
+
+        self._settings_changed_hooks_active = False
+
+        try:
+            yield
+        finally:
+            self._settings_changed_hooks_active = True
+
+            # Ensure that the registration changed hooks are called
+            self._settings_changed()
+
+    def add_settings_changed_hook(self, cb: Callable[[], None]) -> None:
+
+        self._settings_changed_hooks.append(cb)
+
+
+GlobalTypeRegistry.get().add_registration_changed_hook(lambda: Settings.rebuild_annotations())
+
+
+class GlobalSettings:
+
+    _global_settings: Settings | None = None
+
+    @staticmethod
+    def get() -> Settings:
+
+        if (GlobalSettings._global_settings is None):
+            from aiq.runtime.loader import PluginTypes
+            from aiq.runtime.loader import discover_and_register_plugins
+
+            discover_and_register_plugins(PluginTypes.REGISTRY_HANDLER)
+
+            GlobalSettings._global_settings = Settings.from_file()
+
+        return GlobalSettings._global_settings
+
+    @staticmethod
+    @contextmanager
+    def push():
+
+        saved = GlobalSettings.get()
+        settings = deepcopy(saved)
+
+        try:
+            GlobalSettings._global_settings = settings
+
+            yield settings
+        finally:
+            GlobalSettings._global_settings = saved
+            GlobalSettings._global_settings._settings_changed()

aiq/test/.namespace

@@ -0,0 +1 @@
+Note: This is a python namespace package and this directory should remain empty. Do NOT add a `__init__.py` file or any other files to this directory. This file is also needed to ensure the directory exists in git.

aiq/tool/chat_completion.py

@@ -0,0 +1,74 @@
+# 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.
+"""
+Simple Completion Function for AIQ Toolkit
+
+This module provides a simple completion function that can handle
+natural language queries and perform basic text completion tasks.
+"""
+
+from pydantic import Field
+
+from aiq.builder.builder import Builder
+from aiq.builder.framework_enum import LLMFrameworkEnum
+from aiq.cli.register_workflow import register_function
+from aiq.data_models.component_ref import LLMRef
+from aiq.data_models.function import FunctionBaseConfig
+
+
+class ChatCompletionConfig(FunctionBaseConfig, name="chat_completion"):
+    """Configuration for the Chat Completion Function."""
+
+    system_prompt: str = Field(("You are a helpful AI assistant. Provide clear, accurate, and helpful "
+                                "responses to user queries. You can give general advice, recommendations, "
+                                "tips, and engage in conversation. Be helpful and informative."),
+                               description="The system prompt to use for chat completion.")
+
+    llm_name: LLMRef = Field(description="The LLM to use for generating responses.")
+
+
+@register_function(config_type=ChatCompletionConfig)
+async def register_chat_completion(config: ChatCompletionConfig, builder: Builder):
+    """Registers a chat completion function that can handle natural language queries."""
+
+    # Get the LLM from the builder context using the configured LLM reference
+    # Use LangChain framework wrapper since we're using LangChain-based LLM
+    llm = await builder.get_llm(config.llm_name, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
+
+    async def _chat_completion(query: str) -> str:
+        """A simple chat completion function that responds to natural language queries.
+
+        Args:
+            query: The user's natural language query
+
+        Returns:
+            A helpful response to the query
+        """
+        try:
+            # Create a simple prompt with the system message and user query
+            prompt = f"{config.system_prompt}\n\nUser: {query}\n\nAssistant:"
+
+            # Generate response using the LLM
+            response = await llm.ainvoke(prompt)
+
+            return response
+
+        except Exception as e:
+            # Fallback response if LLM call fails
+            return (f"I apologize, but I encountered an error while processing your "
+                    f"query: '{query}'. Please try rephrasing your question or try "
+                    f"again later. Error: {str(e)}")
+
+    yield _chat_completion