feat!: Add the Base SDK support to the Toolbox SDK

This is done by updating the underlying `AsyncToolboxTool` class to return a function-like tool instead of a specific LangChain orchestration's `BaseTool` type.

Author: anubhav756

src/toolbox_langchain/async_tools.py

@@ -12,31 +12,33 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import inspect
 from copy import deepcopy
-from typing import Any, Callable, TypeVar, Union
+from typing import Any, Callable, Type, Union
 from warnings import warn
 
 from aiohttp import ClientSession
-from langchain_core.tools import BaseTool
+from pydantic import BaseModel
 
 from .utils import (
+    ParameterSchema,
     ToolSchema,
     _find_auth_params,
     _find_bound_params,
     _invoke_tool,
+    _parse_type,
+    _schema_to_docstring,
     _schema_to_model,
 )
 
-T = TypeVar("T")
-
 
 # This class is an internal implementation detail and is not exposed to the
 # end-user. It should not be used directly by external code. Changes to this
 # class will not be considered breaking changes to the public API.
-class AsyncToolboxTool(BaseTool):
+class AsyncToolboxTool:
     """
-    A subclass of LangChain's BaseTool that supports features specific to
-    Toolbox, like bound parameters and authenticated tools.
+    A class that supports features specific to Toolbox, like bound parameters
+    and authenticated tools.
     """
 
     def __init__(
@@ -110,51 +112,70 @@ def __init__(
 
         # Bind values for parameters present in the schema that don't require
         # authentication.
-        bound_params = {
+        __bound_params = {
             param_name: param_value
             for param_name, param_value in bound_params.items()
             if param_name in [param.name for param in non_auth_bound_params]
         }
 
         # Update the tools schema to validate only the presence of parameters
         # that neither require authentication nor are bound.
-        schema.parameters = non_auth_non_bound_params
-
-        # Due to how pydantic works, we must initialize the underlying
-        # BaseTool class before assigning values to member variables.
-        super().__init__(
-            name=name,
-            description=schema.description,
-            args_schema=_schema_to_model(model_name=name, schema=schema.parameters),
-        )
+        __updated_schema = deepcopy(schema)
+        __updated_schema.parameters = non_auth_non_bound_params
 
         self.__name = name
-        self.__schema = schema
+        self.__schema = __updated_schema
+        self.__model = _schema_to_model(self.__name, self.__schema.parameters)
         self.__url = url
         self.__session = session
         self.__auth_tokens = auth_tokens
         self.__auth_params = auth_params
-        self.__bound_params = bound_params
+        self.__bound_params = __bound_params
 
         # Warn users about any missing authentication so they can add it before
         # tool invocation.
         self.__validate_auth(strict=False)
 
-    def _run(self, **kwargs: Any) -> dict[str, Any]:
-        raise NotImplementedError("Synchronous methods not supported by async tools.")
+        # Store parameter definitions for the function signature and annotations
+        sig_params = []
+        annotations = {}
+        for param in self.__schema.parameters:
+            param_type = _parse_type(param)
+            annotations[param.name] = param_type
+            sig_params.append(
+                inspect.Parameter(
+                    param.name,
+                    inspect.Parameter.POSITIONAL_OR_KEYWORD,
+                    annotation=param_type,
+                )
+            )
+
+        # Set function name, docstring, signature and annotations
+        self.__name__ = self.__name
+        self.__qualname__ = self.__name
+        self.__doc__ = _schema_to_docstring(self.__schema)
+        self.__signature__ = inspect.Signature(
+            parameters=sig_params, return_annotation=dict[str, Any]
+        )
+        self.__annotations__ = annotations
 
-    async def _arun(self, **kwargs: Any) -> dict[str, Any]:
+    async def __call__(self, *args: Any, **kwargs: Any) -> dict[str, Any]:
         """
         The coroutine that invokes the tool with the given arguments.
 
         Args:
-            **kwargs: The arguments to the tool.
+            **args: The positional arguments to the tool.
+            **kwargs: The keyword arguments to the tool.
 
         Returns:
             A dictionary containing the parsed JSON response from the tool
             invocation.
         """
 
+        # Validate arguments
+        validated_args = self.__signature__.bind(*args, **kwargs).arguments
+        self.__model.model_validate(validated_args)
+
         # If the tool had parameters that require authentication, then right
         # before invoking that tool, we check whether all these required
         # authentication sources have been registered or not.
@@ -169,10 +190,10 @@ async def _arun(self, **kwargs: Any) -> dict[str, Any]:
                 evaluated_params[param_name] = param_value
 
         # Merge bound parameters with the provided arguments
-        kwargs.update(evaluated_params)
+        validated_args.update(evaluated_params)
 
         return await _invoke_tool(
-            self.__url, self.__session, self.__name, kwargs, self.__auth_tokens
+            self.__url, self.__session, self.__name, validated_args, self.__auth_tokens
         )
 
     def __validate_auth(self, strict: bool = True) -> None:

src/toolbox_langchain/tools.py

@@ -17,14 +17,12 @@
 from threading import Thread
 from typing import Any, Awaitable, Callable, TypeVar, Union
 
-from langchain_core.tools import BaseTool
-
 from .async_tools import AsyncToolboxTool
 
 T = TypeVar("T")
 
 
-class ToolboxTool(BaseTool):
+class ToolboxTool:
     """
     A subclass of LangChain's BaseTool that supports features specific to
     Toolbox, like bound parameters and authenticated tools.
@@ -45,14 +43,6 @@ def __init__(
             thread: The thread to run blocking operations in.
         """
 
-        # Due to how pydantic works, we must initialize the underlying
-        # BaseTool class before assigning values to member variables.
-        super().__init__(
-            name=async_tool.name,
-            description=async_tool.description,
-            args_schema=async_tool.args_schema,
-        )
-
         self.__async_tool = async_tool
         self.__loop = loop
         self.__thread = thread
@@ -77,11 +67,8 @@ async def __run_as_async(self, coro: Awaitable[T]) -> T:
             asyncio.run_coroutine_threadsafe(coro, self.__loop)
         )
 
-    def _run(self, **kwargs: Any) -> dict[str, Any]:
-        return self.__run_as_sync(self.__async_tool._arun(**kwargs))
-
-    async def _arun(self, **kwargs: Any) -> dict[str, Any]:
-        return await self.__run_as_async(self.__async_tool._arun(**kwargs))
+    def __call__(self, *args: Any, **kwargs: Any) -> dict[str, Any]:
+        return self.__run_as_sync(self.__async_tool(*args, **kwargs))
 
     def add_auth_tokens(
         self, auth_tokens: dict[str, Callable[[], str]], strict: bool = True

src/toolbox_langchain/utils.py

@@ -266,3 +266,29 @@ def _find_bound_params(
             _non_bound_params.append(param)
 
     return (_bound_params, _non_bound_params)
+
+
+def _schema_to_docstring(tool_schema: ToolSchema) -> str:
+    """Generates a Google Style docstring from a ToolSchema object.
+
+    If the schema has parameters, the docstring includes an 'Args:' section
+    detailing each parameter's name, type, and description. If no parameters are
+    present, only the tool's description is returned.
+
+    Args:
+        tool_schema: The schema object defining the tool's interface,
+            including its description and parameters.
+
+    Returns:
+        str: A Google Style formatted docstring.
+    """
+
+    if not tool_schema.parameters:
+        return tool_schema.description
+
+    docstring = f"{tool_schema.description}\n\nArgs:"
+    for param in tool_schema.parameters:
+        docstring += (
+            f"\n    {param.name} ({_parse_type(param).__name__}): {param.description}"
+        )
+    return docstring