mirror of
https://github.com/hwchase17/langchain
synced 2024-11-11 19:11:02 +00:00
32c5be8b73
## Summary This PR implements the "Connery Action Tool" and "Connery Toolkit". Using them, you can integrate Connery actions into your LangChain agents and chains. Connery is an open-source plugin infrastructure for AI. With Connery, you can easily create a custom plugin with a set of actions and seamlessly integrate them into your LangChain agents and chains. Connery will handle the rest: runtime, authorization, secret management, access management, audit logs, and other vital features. Additionally, Connery and our community offer a wide range of ready-to-use open-source plugins for your convenience. Learn more about Connery: - GitHub: https://github.com/connery-io/connery-platform - Documentation: https://docs.connery.io - Twitter: https://twitter.com/connery_io ## TODOs - [x] API wrapper - [x] Integration tests - [x] Connery Action Tool - [x] Docs - [x] Example - [x] Integration tests - [x] Connery Toolkit - [x] Docs - [x] Example - [x] Formatting (`make format`) - [x] Linting (`make lint`) - [x] Testing (`make test`)
164 lines
5.5 KiB
Python
164 lines
5.5 KiB
Python
import asyncio
|
|
from functools import partial
|
|
from typing import Any, Dict, List, Optional, Type
|
|
|
|
from langchain_core.callbacks.manager import (
|
|
AsyncCallbackManagerForToolRun,
|
|
CallbackManagerForToolRun,
|
|
)
|
|
from langchain_core.pydantic_v1 import BaseModel, Field, create_model, root_validator
|
|
from langchain_core.tools import BaseTool
|
|
|
|
from langchain_community.tools.connery.models import Action, Parameter
|
|
|
|
|
|
class ConneryAction(BaseTool):
|
|
"""
|
|
A LangChain Tool wrapping a Connery Action.
|
|
"""
|
|
|
|
name: str
|
|
description: str
|
|
args_schema: Type[BaseModel]
|
|
|
|
action: Action
|
|
connery_service: Any
|
|
|
|
def _run(
|
|
self,
|
|
run_manager: Optional[CallbackManagerForToolRun] = None,
|
|
**kwargs: Dict[str, str],
|
|
) -> Dict[str, str]:
|
|
"""
|
|
Runs the Connery Action with the provided input.
|
|
Parameters:
|
|
kwargs (Dict[str, str]): The input dictionary expected by the action.
|
|
Returns:
|
|
Dict[str, str]: The output of the action.
|
|
"""
|
|
|
|
return self.connery_service.run_action(self.action.id, kwargs)
|
|
|
|
async def _arun(
|
|
self,
|
|
run_manager: Optional[AsyncCallbackManagerForToolRun] = None,
|
|
**kwargs: Dict[str, str],
|
|
) -> Dict[str, str]:
|
|
"""
|
|
Runs the Connery Action asynchronously with the provided input.
|
|
Parameters:
|
|
kwargs (Dict[str, str]): The input dictionary expected by the action.
|
|
Returns:
|
|
Dict[str, str]: The output of the action.
|
|
"""
|
|
|
|
func = partial(self._run, **kwargs)
|
|
return await asyncio.get_event_loop().run_in_executor(None, func)
|
|
|
|
def get_schema_json(self) -> str:
|
|
"""
|
|
Returns the JSON representation of the Connery Action Tool schema.
|
|
This is useful for debugging.
|
|
Returns:
|
|
str: The JSON representation of the Connery Action Tool schema.
|
|
"""
|
|
|
|
return self.args_schema.schema_json(indent=2)
|
|
|
|
@root_validator()
|
|
def validate_attributes(cls, values: dict) -> dict:
|
|
"""
|
|
Validate the attributes of the ConneryAction class.
|
|
Parameters:
|
|
values (dict): The arguments to validate.
|
|
Returns:
|
|
dict: The validated arguments.
|
|
"""
|
|
|
|
# Import ConneryService here and check if it is an instance
|
|
# of ConneryService to avoid circular imports
|
|
from .service import ConneryService
|
|
|
|
if not isinstance(values.get("connery_service"), ConneryService):
|
|
raise ValueError(
|
|
"The attribute 'connery_service' must be an instance of ConneryService."
|
|
)
|
|
|
|
if not values.get("name"):
|
|
raise ValueError("The attribute 'name' must be set.")
|
|
if not values.get("description"):
|
|
raise ValueError("The attribute 'description' must be set.")
|
|
if not values.get("args_schema"):
|
|
raise ValueError("The attribute 'args_schema' must be set.")
|
|
if not values.get("action"):
|
|
raise ValueError("The attribute 'action' must be set.")
|
|
if not values.get("connery_service"):
|
|
raise ValueError("The attribute 'connery_service' must be set.")
|
|
|
|
return values
|
|
|
|
@classmethod
|
|
def create_instance(cls, action: Action, connery_service: Any) -> "ConneryAction":
|
|
"""
|
|
Creates a Connery Action Tool from a Connery Action.
|
|
Parameters:
|
|
action (Action): The Connery Action to wrap in a Connery Action Tool.
|
|
connery_service (ConneryService): The Connery Service
|
|
to run the Connery Action. We use Any here to avoid circular imports.
|
|
Returns:
|
|
ConneryAction: The Connery Action Tool.
|
|
"""
|
|
|
|
# Import ConneryService here and check if it is an instance
|
|
# of ConneryService to avoid circular imports
|
|
from .service import ConneryService
|
|
|
|
if not isinstance(connery_service, ConneryService):
|
|
raise ValueError(
|
|
"The connery_service must be an instance of ConneryService."
|
|
)
|
|
|
|
input_schema = cls._create_input_schema(action.inputParameters)
|
|
description = action.title + (
|
|
": " + action.description if action.description else ""
|
|
)
|
|
|
|
instance = cls(
|
|
name=action.id,
|
|
description=description,
|
|
args_schema=input_schema,
|
|
action=action,
|
|
connery_service=connery_service,
|
|
)
|
|
|
|
return instance
|
|
|
|
@classmethod
|
|
def _create_input_schema(cls, inputParameters: List[Parameter]) -> Type[BaseModel]:
|
|
"""
|
|
Creates an input schema for a Connery Action Tool
|
|
based on the input parameters of the Connery Action.
|
|
Parameters:
|
|
inputParameters: List of input parameters of the Connery Action.
|
|
Returns:
|
|
Type[BaseModel]: The input schema for the Connery Action Tool.
|
|
"""
|
|
|
|
dynamic_input_fields: Dict[str, Any] = {}
|
|
|
|
for param in inputParameters:
|
|
default = ... if param.validation and param.validation.required else None
|
|
title = param.title
|
|
description = param.title + (
|
|
": " + param.description if param.description else ""
|
|
)
|
|
type = param.type
|
|
|
|
dynamic_input_fields[param.key] = (
|
|
str,
|
|
Field(default, title=title, description=description, type=type),
|
|
)
|
|
|
|
InputModel = create_model("InputSchema", **dynamic_input_fields)
|
|
return InputModel
|