Ai abstraction refactor

nextpy/ai/agent/assistant_agent.py

@@ -0,0 +1,214 @@
+from typing import Any, Callable, Tuple
+from nextpy.ai.agent.base_agent import BaseAgent
+import logging
+from pathlib import Path
+from nextpy.ai import engine
+import inspect
+import asyncio
+import logging
+
+
+def _call_functions(functions):
+ for function, arguments, keyword_args in functions:
+ if inspect.iscoroutinefunction(function):
+ try:
+ other_loop = asyncio.get_event_loop()
+ import nest_asyncio
+
+ nest_asyncio.apply(other_loop)
+ except RuntimeError:
+ pass
+ loop = asyncio.new_event_loop()
+ loop.run_until_complete(function(*arguments, **keyword_args))
+ else:
+ function(*arguments, **keyword_args)
+
+
+async def _a_call_functions(functions):
+ for function, arguments, keyword_args in functions:
+ if inspect.iscoroutinefunction(function):
+ await function(*arguments, **keyword_args)
+ else:
+ function(*arguments, **keyword_args)
+
+
+class AssistantAgent(BaseAgent):
+ """
+ AssistantAgent class represents an assistant agent that interacts with users in a conversational manner.
+
+ :param name: The name of the assistant agent.
+ :type name: str
+ :param llm: The language model used by the assistant agent.
+ :type llm: LanguageModel
+ :param memory: The memory used by the assistant agent.
+ :type memory: Memory
+ :param async_mode: Whether the assistant agent should run in asynchronous mode or not. Default is True.
+ :type async_mode: bool, optional
+ :param system_message: The system message included in the prompt. Default is None.
+ :type system_message: str, optional
+ :param functions_before_call: List of functions to be called before the main function call. Default is None.
+ :type functions_before_call: List[Callable], optional
+ :param functions_after_call: List of functions to be called after the main function call. Default is None.
+ :type functions_after_call: List[Callable], optional
+
+ The assistant agent is built on top of the existing BaseAgent and serves as a simple interface for creating an AI assistant agent. 
+ It provides a convenient way to define an AI assistant agent that can interact with users in a conversational manner. 
+ The assistant agent can be customized with a name, language model, memory, and other parameters. 
+ It also supports asynchronous mode, allowing it to handle multiple conversations simultaneously. 
+
+ MultiagentManager can be used to manage multiple assistant agents and coordinate their interactions with users.
+
+
+ Example:
+
+
+ tailwind_agent = AssistantAgent(name='Tailwind Class Generator', llm=llm, memory=None, async_mode=False,
+ system_message='''automates the creation of Tailwind CSS classes, streamlining the process of building stylish and responsive user interfaces. By leveraging advanced algorithms and design principles, the Tailwind Class Generator analyzes your design elements and dynamically generates the optimal set of Tailwind utility classes. 
+ This tool is designed to enhance efficiency in web development, allowing developers to focus more on high-level design decisions and less on manually crafting individual CSS rules. With the Tailwind Class Generator, achieving a visually appealing and consistent design becomes a seamless experience.
+ '''
+ )
+ """
+
+ DEFAULT_PROMPT = '''
+ {{#system~}} {{name}}, you are working in the following team :{{agents}}
+ {{~/system}}
+
+ {{#user~}}
+ Read the following CONVERSATION :
+ {{messages}}
+ Respond as {{name}}. Do not thank any team member or show appreciation."
+ {{~/user}}
+
+ {{#assistant~}}
+ {{gen 'answer' temperature=0 max_tokens=500}}
+ {{~/assistant}}
+ '''
+
+ def __init__(self,
+ name,
+ llm=None,
+ memory=None,
+ async_mode: bool = False,
+ system_message: str | None = None,
+ custom_engine=None,
+ functions_before_call: Tuple[Callable,
+ Tuple[Any], Tuple[Any]] | None = None,
+ functions_after_call: Tuple[Callable,
+ Tuple[Any], Tuple[Any]] | None = None,
+ description: str = "Helpful AI Assistant Agent",
+ **kwargs):
+ """
+ Initializes an instance of the AssistantAgent class.
+
+ :param name: The name of the assistant agent.
+ :type name: str
+ :param llm: The language model used by the assistant agent.
+ :type llm: LanguageModel
+ :param memory: The memory used by the assistant agent.
+ :type memory: Memory
+ :param async_mode: Whether the assistant agent should run in asynchronous mode or not. Default is True.
+ :type async_mode: bool, optional
+ :param system_message: The system message to be displayed to the user. Default is None.
+ :type system_message: str, optional
+ :param engine: The engine used by the assistant agent. Either llm or engine must be provided.
+ :type engine: Engine, optional
+ :param functions_before_call: List of functions, args and kwargs, to be called before the main function call. Default is None.
+ :type functions_before_call: List[Callable], optional
+ :param functions_after_call: List of functions, args and kwargs to be called after the main function call. Default is None.
+ :type functions_after_call: List[Callable], optional
+ :param kwargs: Additional keyword arguments.
+ """
+ super().__init__(llm=llm, **kwargs)
+ self.name = name
+ self.prompt = self.DEFAULT_PROMPT
+ self.system_message = system_message
+ # This is used by multiagent manager to determine whether to use receive or a_receive
+ self.async_mode = async_mode
+
+ if system_message is not None:
+ try:
+ system_message = Path(system_message).read_text()
+ except Exception:
+ pass
+ self.prompt = self.prompt[:self.prompt.find(
+ '{{~/system}}')] + system_message + self.prompt[self.prompt.find('{{~/system}}'):]
+
+ # Either llm or engine must be provided
+ if llm is not None or engine is not None:
+ logging.debug("Warning! Either llm or engine must be provided.")
+
+ self.engine = custom_engine if custom_engine is not None else engine(
+ template=self.prompt, llm=llm, memory=memory, async_mode=async_mode, **kwargs)
+ self.output_key = 'answer'
+ self.functions_before_call = functions_before_call
+ self.functions_after_call = functions_after_call
+ self.description = description
+
+ @staticmethod
+ def function_call_decorator(func):
+ """
+ Decorator function that wraps the main function call with additional functions to be called before and after.
+
+ :param func: The main function to be called.
+ :type func: Callable
+ :return: The wrapped function.
+ :rtype: Callable
+ """
+ if inspect.iscoroutinefunction(func):
+ async def a_inner(self, *args, **kwargs):
+ if self.functions_before_call is not None:
+ await _a_call_functions(self.functions_before_call)
+
+ result = await func(self, *args, **kwargs)
+
+ if self.functions_after_call is not None:
+ await _a_call_functions(self.functions_after_call)
+
+ return result
+ return a_inner
+ else:
+ def inner(self, *args, **kwargs):
+ if self.functions_before_call is not None:
+ _call_functions(self.functions_before_call)
+
+ result = func(self, *args, **kwargs)
+
+ if self.functions_after_call is not None:
+ _call_functions(self.functions_after_call)
+
+ return result
+ return inner
+
+ @function_call_decorator
+ def receive(self, agents, messages, termination_message):
+ """
+ Receives messages from other agents and generates a response.
+
+ :param agents: The list of agents involved in the conversation.
+ :type agents: List[str]
+ :param messages: The list of messages in the conversation.
+ :type messages: List[str]
+ :param termination_message: The termination message for the conversation.
+ :type termination_message: str
+ :return: The generated response.
+ :rtype: str
+ """
+ output = self.run(agents=agents, messages=messages, name=self.name)
+ return output
+
+ @function_call_decorator
+ async def a_receive(self, agents, messages, termination_message):
+ """
+ Asynchronously receives messages from other agents and generates a response.
+
+ :param agents: The list of agents involved in the conversation.
+ :type agents: List[str]
+ :param messages: The list of messages in the conversation.
+ :type messages: List[str]
+ :param termination_message: The termination message for the conversation.
+ :type termination_message: str
+ :return: The generated response.
+ :rtype: str
+ """
+ output = await self.arun(agents=agents, messages=messages, name=self.name)
+ return output

nextpy/ai/agent/base_agent.py

@@ -1,4 +1,4 @@
-# This file has been modified by the Nextpy Team in 2023 using AI tools and automation scripts. 
+# This file has been modified by the Nextpy Team in 2023 using AI skills and automation scripts.
 # We have rigorously tested these modifications to ensure reliability and performance. Based on successful test results, we are confident in the quality and stability of these changes.
 
 import argparse
@@ -7,6 +7,7 @@
 import json
 import logging
 from enum import Enum
+from nextpy.ai.skills.base import BaseSkill
 from typing import Any, Dict, List, Optional, Union
 
 import nest_asyncio
@@ -16,7 +17,6 @@
 from nextpy.ai.engine._program import extract_text
 from nextpy.ai.memory.base import BaseMemory
 from nextpy.ai.rag.doc_loader import document_loader
-from nextpy.ai.tools.basetool import BaseTool
 
 log = logging.getLogger(__name__)
 
@@ -32,16 +32,16 @@ class AgentState(Enum):
 
  IDLE = 0
  BUSY = 1
- USED_AS_TOOL = 2
+ USED_AS_skill = 2
  ERROR = 3
 
 
 class BaseAgent:
  def __init__(
  self,
  rag: Optional[Any] = None,
- tools: Optional[List[BaseTool]] = None,
  llm: Optional[Any] = None,
+ skills: Optional[List[BaseSkill]] = None,
  prompt_template: str = None,
  input_variables: Dict[str, Any] = {},
  agent_id: str = "default",
@@ -52,8 +52,8 @@ def __init__(
  ):
  self.agent_id = agent_id
  self.rag = rag
- self.tools = tools
  self.llm = llm
+ self.skills = skills
  self.prompt_template = prompt_template
  self.input_variables = input_variables
  self.memory = memory
@@ -98,14 +98,14 @@ def get_knowledge_variable(self):
  def default_llm_model(self):
  pass
 
- def add_tool(self, tool: BaseTool) -> None:
- """Add a tool to the agent's tool list."""
- self.tools.append(tool)
+ def add_skill(self, skill: BaseSkill) -> None:
+ """Add a skill to the agent's skill list."""
+ self.skills.append(skill)
 
- def remove_tool(self, tool: BaseTool) -> None:
- """Remove a tool from the agent's tool list."""
- if tool in self.tools:
- self.tools.remove(tool)
+ def remove_skill(self, skill: BaseSkill) -> None:
+ """Remove a skill from the agent's skill list."""
+ if skill in self.skills:
+ self.skills.remove(skill)
 
  def llm_instance(self) -> engine.llms.OpenAI:
  """Create an instance of the language model."""
@@ -146,7 +146,8 @@ def run(self, **kwargs) -> Union[str, Dict[str, Any]]:
  RETRIEVED_KNOWLEDGE=retrieved_knowledge, **kwargs, silent=True
  )
  else:
- raise ValueError("knowledge_variable not found in input kwargs")
+ raise ValueError(
+ "knowledge_variable not found in input kwargs")
  else:
  output = self.engine(**kwargs, silent=True, from_agent=True)
 
@@ -166,7 +167,8 @@ def run(self, **kwargs) -> Union[str, Dict[str, Any]]:
  if output.variables().get(_output_key):
  return output[_output_key]
  else:
- logging.warning("Output key not found in output, so full output returned")
+ logging.warning(
+ "Output key not found in output, so full output returned")
  return output
 
  async def arun(self, **kwargs) -> Union[str, Dict[str, Any]]:
@@ -188,7 +190,8 @@ async def arun(self, **kwargs) -> Union[str, Dict[str, Any]]:
  RETRIEVED_KNOWLEDGE=retrieved_knowledge, **kwargs, silent=True
  )
  else:
- raise ValueError("knowledge_variable not found in input kwargs")
+ raise ValueError(
+ "knowledge_variable not found in input kwargs")
  else:
  output = await self.engine(**kwargs, silent=True, from_agent=True)
  # Handle memory here
@@ -207,7 +210,8 @@ async def arun(self, **kwargs) -> Union[str, Dict[str, Any]]:
  if output.variables().get(_output_key):
  return output[_output_key]
  else:
- logging.warning("Output key not found in output, so full output returned")
+ logging.warning(
+ "Output key not found in output, so full output returned")
  return output
 
  def _handle_memory(self, new_program):
@@ -260,7 +264,8 @@ def cli(self):
  )
 
  for var in _vars:
- parser.add_argument(f"--{var}", help=f"Pass {var} as an input variable")
+ parser.add_argument(
+ f"--{var}", help=f"Pass {var} as an input variable")
 
  args = parser.parse_args()
 
@@ -329,7 +334,7 @@ def export_agent_config(self, config_path, export_json=False):
  "prompt_template": self.prompt_template,
  "input_variables": self.input_variables,
  "output_key": self.output_key,
- # 'tools': None if self.tools is None else self.tools
+ # 'skills': None if self.skills is None else self.skills
  }
  with open(config_path, "w") as f:
  yaml.dump(config, f)
@@ -352,7 +357,8 @@ def load_from_config(cls, config_file):
 
  rag = None
  if config["rag"] is not None:
- rag_module_name, rag_class_name = config["rag"]["type"].rsplit(".", 1)
+ rag_module_name, rag_class_name = config["rag"]["type"].rsplit(
+ ".", 1)
  rag_module = importlib.import_module(rag_module_name)
  rag_class = getattr(rag_module, rag_class_name)
 
@@ -391,9 +397,12 @@ def load_from_config(cls, config_file):
  vector_store_module_name, vector_store_class_name = config["rag"][
  "vector_store"
  ]["type"].rsplit(".", 1)
- vector_store_module = importlib.import_module(vector_store_module_name)
- vector_store_class = getattr(vector_store_module, vector_store_class_name)
- vector_store = vector_store_class(embedding_function=embedding_function)
+ vector_store_module = importlib.import_module(
+ vector_store_module_name)
+ vector_store_class = getattr(
+ vector_store_module, vector_store_class_name)
+ vector_store = vector_store_class(
+ embedding_function=embedding_function)
 
  rag = rag_class(
  raw_data=raw_data,