unique_toolkit 0.7.7__py3-none-any.whl → 1.23.0__py3-none-any.whl

unique_toolkit/__init__.py

@@ -2,16 +2,43 @@
 from unique_toolkit.chat import ChatService
 from unique_toolkit.content import ContentService
 from unique_toolkit.embedding import EmbeddingService
-from unique_toolkit.language_model import LanguageModelMessages, LanguageModelService
+from unique_toolkit.framework_utilities.openai.client import (
+    get_async_openai_client,
+    get_openai_client,
+)
+from unique_toolkit.language_model import (
+    LanguageModelMessages,
+    LanguageModelName,
+    LanguageModelService,
+    LanguageModelToolDescription,
+)
+from unique_toolkit.services.knowledge_base import KnowledgeBaseService
 from unique_toolkit.short_term_memory import ShortTermMemoryService
 
+# Conditionally import langchain utilities if langchain is installed
+try:
+    from unique_toolkit.framework_utilities.langchain.client import get_langchain_client  # noqa: F401, I001
+
+    _LANGCHAIN_AVAILABLE = True
+except ImportError:
+    _LANGCHAIN_AVAILABLE = False
+
 # You can add other classes you frequently use here as well
 
 __all__ = [
     "LanguageModelService",
     "LanguageModelMessages",
+    "LanguageModelName",
+    "LanguageModelToolDescription",
     "ChatService",
     "ContentService",
     "EmbeddingService",
     "ShortTermMemoryService",
+    "KnowledgeBaseService",
+    "get_openai_client",
+    "get_async_openai_client",
 ]
+
+# Add langchain-specific exports if available
+if _LANGCHAIN_AVAILABLE:
+    __all__.append("get_langchain_client")

unique_toolkit/_common/api_calling/human_verification_manager.py

@@ -0,0 +1,343 @@
+import hashlib
+from datetime import datetime
+from logging import Logger
+from typing import Any, Generic
+
+import jinja2
+from pydantic import BaseModel, ValidationError
+
+from unique_toolkit._common.endpoint_builder import (
+    ApiOperationProtocol,
+    PathParamsSpec,
+    PathParamsType,
+    PayloadParamSpec,
+    PayloadType,
+    ResponseType,
+)
+from unique_toolkit._common.endpoint_requestor import (
+    RequestContext,
+    RequestorType,
+    build_requestor,
+)
+from unique_toolkit._common.pydantic_helpers import (
+    create_complement_model,
+    create_union_model,
+)
+from unique_toolkit._common.string_utilities import (
+    dict_to_markdown_table,
+    extract_dicts_from_string,
+)
+from unique_toolkit.chat.schemas import ChatMessage, ChatMessageRole
+
+
+class HumanConfirmation(BaseModel):
+    payload_hash: str
+    time_stamp: datetime
+
+
+NEXT_USER_MESSAGE_JINJA2_TEMPLATE = jinja2.Template("""I confirm the api call with the following data:
+```json
+{{ api_call_as_json }}
+```""")
+
+
+ASSISTANT_CONFIRMATION_MESSAGE_JINJA2_TEMPLATE = jinja2.Template(
+    """
+\n
+{{ api_call_as_markdown_table }}
+\n\n
+[{{ button_text }}](https://prompt={{ next_user_message | urlencode }})"""
+)
+
+
+class HumanVerificationManagerForApiCalling(
+    Generic[
+        PathParamsSpec,
+        PathParamsType,
+        PayloadParamSpec,
+        PayloadType,
+        ResponseType,
+    ]
+):
+    """
+    Manages human verification for api calling.
+
+    The idea is that the manager is able to produce the verification message to the user
+    and to detect an api call from the user message.
+
+    If it detects such a verification message in the user message, it will call the api
+    and incorporate the response into the user message.
+    """
+
+    def __init__(
+        self,
+        *,
+        logger: Logger,
+        operation: type[
+            ApiOperationProtocol[
+                PathParamsSpec,
+                PathParamsType,
+                PayloadParamSpec,
+                PayloadType,
+                ResponseType,
+            ]
+        ],
+        requestor_type: RequestorType = RequestorType.REQUESTS,
+        environment_payload_params: BaseModel | None = None,
+        modifiable_payload_params_model: type[BaseModel] | None = None,
+        **kwargs: dict[str, Any],
+    ):
+        """
+        Manages human verification for api calling.
+
+        Args:
+            logger: The logger to use for logging.
+            operation: The operation to use for the api calling.
+            requestor_type: The requestor type to use for the api calling.
+            environment_payload_params: The environment payload params to use for the api calling.
+                If None, the modifiable params model will be the operation payload model.
+                This can be useful for parameters in the payload that should not be modified by the user.
+            modifiable_payload_params_model: The modifiable payload params model to use for the api calling.
+                If None, a complement model will be created using the operation payload model
+                and the environment payload params.
+                If provided, it will be used instead of the complement model.
+                This is necessary if the modifiable params model is required
+                to use custom validators or serializers.
+            **kwargs: Additional keyword arguments to pass to the requestor.
+        """
+        self._logger = logger
+        self._operation = operation
+        self._environment_payload_params = environment_payload_params
+        # Create internal models for this manager instance
+
+        if self._environment_payload_params is None:
+            self._modifiable_payload_params_model = self._operation.payload_model()
+        else:
+            if modifiable_payload_params_model is None:
+                self._modifiable_payload_params_model = create_complement_model(
+                    model_type_a=self._operation.payload_model(),
+                    model_type_b=type(self._environment_payload_params),
+                )
+            else:
+                # This is necessary if the modifiable params model is required
+                # to use custom validators or serializers.
+                self._modifiable_payload_params_model = modifiable_payload_params_model
+
+        if self._environment_payload_params is not None:
+            combined_keys = set(
+                self._modifiable_payload_params_model.model_fields.keys()
+            ) | set(type(self._environment_payload_params).model_fields.keys())
+            payload_keys = set(self._operation.payload_model().model_fields.keys())
+            if not payload_keys.issubset(combined_keys):
+                raise ValueError(
+                    "The modifiable params model + the environment parameters do not have all the keys of the operation payload model."
+                )
+
+        class VerificationModel(BaseModel):
+            confirmation: HumanConfirmation
+            modifiable_params: self._modifiable_payload_params_model  # type: ignore
+
+        self._verification_model = VerificationModel
+
+        self._requestor_type = requestor_type
+        self._combined_params_model = create_union_model(
+            model_type_a=self._operation.path_params_model(),
+            model_type_b=self._operation.payload_model(),
+        )
+        self._requestor = build_requestor(
+            requestor_type=requestor_type,
+            operation_type=operation,
+            combined_model=self._combined_params_model,
+            **kwargs,
+        )
+
+    def detect_api_calls_from_user_message(
+        self,
+        *,
+        last_assistant_message: ChatMessage,
+        user_message: str,
+    ) -> PayloadType | None:
+        user_message_dicts = extract_dicts_from_string(user_message)
+        if len(user_message_dicts) == 0:
+            return None
+
+        user_message_dicts.reverse()
+        for user_message_dict in user_message_dicts:
+            try:
+                # Convert dict to payload model first, then create payload
+                verfication_data = self._verification_model.model_validate(
+                    user_message_dict, by_alias=True, by_name=True
+                )
+                if self._verify_human_verification(
+                    verfication_data.confirmation, last_assistant_message
+                ):
+                    payload_dict = verfication_data.modifiable_params.model_dump()
+                    if self._environment_payload_params is not None:
+                        payload_dict.update(
+                            self._environment_payload_params.model_dump()
+                        )
+
+                    return self._operation.payload_model().model_validate(
+                        payload_dict, by_alias=True, by_name=True
+                    )
+
+            except Exception as e:
+                self._logger.error(f"Error detecting api calls from user message: {e}")
+
+        return None
+
+    def _verify_human_verification(
+        self, confirmation: HumanConfirmation, last_assistant_message: ChatMessage
+    ) -> bool:
+        if (
+            last_assistant_message.role != ChatMessageRole.ASSISTANT
+            or last_assistant_message.content is None
+        ):
+            self._logger.error(
+                "Last assistant message is not an assistant message or content is empty."
+            )
+            return False
+
+        return confirmation.payload_hash in last_assistant_message.content
+
+    def _create_next_user_message(self, payload: PayloadType) -> str:
+        # Extract only the modifiable fields from the payload
+        payload_dict = payload.model_dump()
+        if self._environment_payload_params is not None:
+            # Remove environment params from payload to avoid validation errors
+            environment_fields = set(
+                type(self._environment_payload_params).model_fields.keys()
+            )
+            modifiable_dict = {
+                k: v for k, v in payload_dict.items() if k not in environment_fields
+            }
+        else:
+            modifiable_dict = payload_dict
+
+        modifiable_params = self._modifiable_payload_params_model.model_validate(
+            modifiable_dict,
+            by_alias=True,
+            by_name=True,
+        )
+        api_call = self._verification_model(
+            modifiable_params=modifiable_params,
+            confirmation=HumanConfirmation(
+                payload_hash=hashlib.sha256(
+                    modifiable_params.model_dump_json().encode()
+                ).hexdigest(),
+                time_stamp=datetime.now(),
+            ),
+        )
+        return NEXT_USER_MESSAGE_JINJA2_TEMPLATE.render(
+            api_call_as_json=api_call.model_dump_json(indent=2)
+        )
+
+    def create_assistant_confirmation_message(
+        self, *, payload: PayloadType, button_text: str = "Confirm"
+    ) -> str:
+        return ASSISTANT_CONFIRMATION_MESSAGE_JINJA2_TEMPLATE.render(
+            api_call_as_markdown_table=dict_to_markdown_table(payload.model_dump()),
+            button_text=button_text,
+            next_user_message=self._create_next_user_message(payload),
+        )
+
+    def call_api(
+        self,
+        *,
+        context: RequestContext,
+        path_params: PathParamsType,
+        payload: PayloadType,
+    ) -> ResponseType:
+        """
+        Call the api with the given path params, payload and secured payload params.
+
+        The `secured payload params` are params that are enforced by the application.
+        It should generally be not possible for the user to adapt those but here we
+        ensure that the application has the last word.
+
+        """
+        params = path_params.model_dump()
+        params.update(payload.model_dump())
+
+        response = self._requestor.request(
+            context=context,
+            **params,
+        )
+        try:
+            return self._operation.handle_response(response)
+        except ValidationError as e:
+            self._logger.error(f"Error calling api: {e}. Response: {response}")
+            raise e
+
+
+if __name__ == "__main__":
+    import logging
+    from string import Template
+
+    from unique_toolkit._common.endpoint_builder import (
+        EndpointMethods,
+        build_api_operation,
+    )
+
+    class GetUserPathParams(BaseModel):
+        user_id: int
+
+    class GetUserRequestBody(BaseModel):
+        include_profile: bool = False
+
+    class UserResponse(BaseModel):
+        id: int
+        name: str
+
+    class CombinedParams(GetUserPathParams, GetUserRequestBody):
+        pass
+
+    UserApiOperation = build_api_operation(
+        method=EndpointMethods.GET,
+        path_template=Template("/users/{user_id}"),
+        path_params_constructor=GetUserPathParams,
+        payload_constructor=GetUserRequestBody,
+        response_model_type=UserResponse,
+    )
+
+    human_verification_manager = HumanVerificationManagerForApiCalling(
+        logger=logging.getLogger(__name__),
+        operation=UserApiOperation,
+        requestor_type=RequestorType.FAKE,
+        return_value={"id": 100, "name": "John Doe"},
+    )
+
+    payload = GetUserRequestBody(include_profile=True)
+
+    api_call = human_verification_manager._verification_model(
+        modifiable_params=payload,
+        confirmation=HumanConfirmation(
+            payload_hash=hashlib.sha256(payload.model_dump_json().encode()).hexdigest(),
+            time_stamp=datetime.now(),
+        ),
+    )
+
+    last_assistant_message = ChatMessage(
+        role=ChatMessageRole.ASSISTANT,
+        text=api_call.confirmation.payload_hash,
+        chat_id="123",
+    )
+
+    user_message_with_api_call = human_verification_manager._create_next_user_message(
+        payload=payload
+    )
+
+    print(user_message_with_api_call)
+
+    payload = human_verification_manager.detect_api_calls_from_user_message(
+        user_message=user_message_with_api_call,
+        last_assistant_message=last_assistant_message,
+    )
+
+    if payload is None:
+        print("❌ Detection failed - payload is None")
+        exit(1)
+    else:
+        print("✅ Detection successful!")
+        print(f"Payload: {payload.model_dump()}")
+        print("✅ Dict extraction from string works correctly!")

unique_toolkit/_common/base_model_type_attribute.py

@@ -0,0 +1,303 @@
+"""
+The following can be used to define a pydantic BaseModel that has has
+an attribute of type Pydantic BaseModel.
+
+This is useful for:
+- Tooldefinition for large language models (LLMs) with flexible parameters.
+- General Endpoint defintions from configuration
+"""
+
+import json
+from enum import StrEnum
+from typing import Annotated, Any, TypeVar, Union, get_args, get_origin
+
+from jambo import SchemaConverter
+from jambo.types.json_schema_type import JSONSchema
+from pydantic import (
+    BaseModel,
+    BeforeValidator,
+    Field,
+    create_model,
+)
+
+
+def _get_actual_type(python_type: type) -> type | None | Any:
+    if get_origin(python_type) is not None:
+        origin = get_origin(python_type)
+        args = get_args(python_type)
+
+        if origin is Annotated:
+            # For Annotated types, the first argument is the actual type
+            if args:
+                actual_type = args[0]
+                # Recursively handle nested generic types (e.g., Annotated[Optional[str], ...])
+                if get_origin(actual_type) is not None:
+                    return _get_actual_type(actual_type)
+            else:
+                raise ValueError(f"Invalid Annotated type: {python_type}")
+        elif origin is Union:
+            # For Union types (including Optional), use the first non-None type
+            if args:
+                for arg in args:
+                    if arg is not type(None):  # Skip NoneType
+                        return _get_actual_type(arg)
+                raise ValueError(f"Union type contains only None: {python_type}")
+            else:
+                raise ValueError(f"Invalid Union type: {python_type}")
+        else:
+            # Other generic types, use the origin
+            actual_type = origin
+    else:
+        # Regular type
+        actual_type = python_type
+
+    return actual_type
+
+
+class ParameterType(StrEnum):
+    STRING = "string"
+    INTEGER = "integer"
+    NUMBER = "number"
+    BOOLEAN = "boolean"
+
+    def to_python_type(self) -> type:
+        """Convert ParameterType to Python type"""
+
+        match self:
+            case ParameterType.STRING:
+                return str
+            case ParameterType.INTEGER:
+                return int
+            case ParameterType.NUMBER:
+                return float
+            case ParameterType.BOOLEAN:
+                return bool
+            case _:
+                raise ValueError(f"Invalid ParameterType: {self}")
+
+    @classmethod
+    def from_python_type(cls, python_type: type) -> "ParameterType":
+        type_to_check = _get_actual_type(python_type)
+
+        # Ensure we have a class before calling issubclass
+        if not isinstance(type_to_check, type):
+            raise ValueError(f"Invalid Python type: {python_type}")
+
+        # Check bool first since bool is a subclass of int in Python
+        if issubclass(type_to_check, bool):
+            return cls.BOOLEAN
+        if issubclass(type_to_check, int):
+            return cls.INTEGER
+        if issubclass(type_to_check, float):
+            return cls.NUMBER
+        if issubclass(type_to_check, str):
+            return cls.STRING
+        raise ValueError(f"Invalid Python type: {python_type}")
+
+
+class Parameter(BaseModel):
+    type: ParameterType
+    name: str
+    description: str
+    required: bool
+
+
+def create_pydantic_model_from_parameter_list(
+    title: str, parameter_list: list[Parameter]
+) -> type[BaseModel]:
+    """Create a Pydantic model from MCP tool's input schema"""
+
+    # Convert JSON schema properties to Pydantic fields
+    fields = {}
+    for parameter in parameter_list:
+        if parameter.required:
+            field = Field(description=parameter.description)
+        else:
+            field = Field(default=None, description=parameter.description)
+
+        fields[parameter.name] = (
+            parameter.type.to_python_type(),
+            field,
+        )
+
+    return create_model(title, **fields)
+
+
+def convert_to_base_model_type(
+    value: type[BaseModel] | str | list[Parameter] | None,
+) -> type[BaseModel]:
+    """
+    BeforeValidator that ensures the final type is always of type[BaseModel].
+
+    If the input is already a BaseModel class, returns it as-is.
+    If the input is a list of Parameter as defined above, converts it to a BaseModel class
+    If the input is a str (JSON schema), converts it to a BaseModel class using SchemaConverter from Jambo.
+    """
+    if isinstance(value, type) and issubclass(value, BaseModel):
+        return value
+
+    if isinstance(value, list):
+        if all(isinstance(item, Parameter) for item in value):
+            return create_pydantic_model_from_parameter_list("Parameters", value)
+
+    converter = SchemaConverter()
+    if isinstance(value, str):
+        return converter.build(JSONSchema(**json.loads(value)))
+
+    raise ValueError(f"Invalid value: {value}")
+
+
+def base_model_to_parameter_list(model: type[BaseModel]) -> list[Parameter]:
+    parameter = []
+    for field_name, field_info in model.model_fields.items():
+        parameter.append(
+            Parameter(
+                type=ParameterType.from_python_type(field_info.annotation or str),
+                name=field_name,
+                description=field_info.description or "",
+                required=field_info.is_required(),
+            )
+        )
+    return parameter
+
+
+# Create the annotated type that ensures BaseModel and generates clean JSON schema
+
+TModel = TypeVar("TModel", bound=BaseModel)
+
+
+class BaseModelTypeTitle(StrEnum):
+    LIST_OF_PARAMETERS = "List of Parameters"
+    JSON_SCHEMA_AS_STRING = "JSON Schema as String"
+    USE_MODEL_FROM_CODE = "Use Model from Code"
+
+
+ListOfParameters = Annotated[
+    list[Parameter], Field(title=BaseModelTypeTitle.LIST_OF_PARAMETERS.value)
+]
+JSONSchemaString = Annotated[
+    str, Field(title=BaseModelTypeTitle.JSON_SCHEMA_AS_STRING.value)
+]
+CodefinedModelType = Annotated[
+    None, Field(title=BaseModelTypeTitle.USE_MODEL_FROM_CODE.value)
+]
+
+
+BaseModelType = Annotated[
+    type[TModel],
+    BeforeValidator(
+        convert_to_base_model_type,
+        json_schema_input_type=ListOfParameters | JSONSchemaString | CodefinedModelType,
+    ),
+]
+
+
+def get_json_schema_extra_for_base_model_type(model: type[BaseModel]):
+    """
+    Returns a json_schema_extra mutator that injects defaults
+    into both the 'string' and 'list[Parameter]' branches.
+
+    This is used to define default for the "oneOf"/"anyOf" validation
+    of the parameters attribute.
+    """
+    sample_params = base_model_to_parameter_list(model)
+
+    def _mutate(schema: dict) -> None:
+        json_default = json.dumps(model.model_json_schema())
+        params_default = [p.model_dump() for p in sample_params]
+
+        for key in ("oneOf", "anyOf"):
+            if key in schema:
+                for entry in schema[key]:
+                    if (
+                        entry.get("type") == "string"
+                        and entry.get("title")
+                        == BaseModelTypeTitle.JSON_SCHEMA_AS_STRING.value
+                    ):
+                        entry["default"] = json_default
+                    if (
+                        entry.get("type") == "array"
+                        and entry.get("title")
+                        == BaseModelTypeTitle.LIST_OF_PARAMETERS.value
+                    ):
+                        entry["default"] = params_default
+
+    return _mutate
+
+
+if __name__ == "__main__":
+    import json
+    from pathlib import Path
+    from typing import Generic
+
+    from openai.types.chat.chat_completion_tool_param import ChatCompletionToolParam
+    from openai.types.shared_params.function_definition import FunctionDefinition
+    from pydantic import BaseModel, Field, field_serializer
+
+    class ToolDescription(BaseModel, Generic[TModel]):
+        name: str = Field(
+            ...,
+            pattern=r"^[a-zA-Z1-9_-]+$",
+            description="Name must adhere to the pattern ^[a-zA-Z1-9_-]+$",
+        )
+        description: str = Field(
+            ...,
+            description="Description of what the tool is doing the tool",
+        )
+
+        strict: bool = Field(
+            default=False,
+            description="Setting strict to true will ensure function calls reliably adhere to the function schema, instead of being best effort.",
+        )
+
+        parameters: BaseModelType[TModel] = Field(
+            ...,
+            description="Json Schema for the tool parameters. Must be valid JSON Schema and able to convert to a Pydantic model",
+        )
+
+        @field_serializer("parameters")
+        def serialize_parameters(self, parameters: type[BaseModel]):
+            return parameters.model_json_schema()
+
+        def to_openai(self) -> ChatCompletionToolParam:
+            return ChatCompletionToolParam(
+                function=FunctionDefinition(
+                    name=self.name,
+                    description=self.description,
+                    parameters=self.parameters.model_json_schema(),
+                    strict=self.strict,
+                ),
+                type="function",
+            )
+
+    class WeatherToolParameterModel(BaseModel):
+        lon: float = Field(
+            ..., description="The longitude of the location to get the weather for"
+        )
+        lat: float = Field(
+            ..., description="The latitude of the location to get the weather for"
+        )
+        name: str = Field(
+            ..., description="The name of the location to get the weather for"
+        )
+
+    class GetWeatherTool(ToolDescription[WeatherToolParameterModel]):
+        parameters: BaseModelType[WeatherToolParameterModel] = Field(
+            default=WeatherToolParameterModel,
+            json_schema_extra=get_json_schema_extra_for_base_model_type(
+                WeatherToolParameterModel
+            ),
+        )
+
+    # The json schema can be used in the RSJF library to create a valid frontend component.
+    # You can test it on https://rjsf-team.github.io/react-jsonschema-form/
+    file = Path(__file__).parent / "weather_tool_schema.json"
+    with file.open("w") as f:
+        f.write(json.dumps(GetWeatherTool.model_json_schema(), indent=2))
+
+    # Notice that the t.parameters is a pydantic model with type annotations
+    t = GetWeatherTool(
+        name="GetWeather", description="Get the weather for a given location"
+    )
+    t.parameters(lon=100, lat=100, name="Test")
+    print(t.model_dump())