"""Module implements an agent that uses OpenAI's APIs function enabled API."""
from typing import Any, List, Optional, Sequence, Tuple, Type, Union
from langchain_core._api import deprecated
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.callbacks import BaseCallbackManager, Callbacks
from langchain_core.language_models import BaseLanguageModel
from langchain_core.messages import (
BaseMessage,
SystemMessage,
)
from langchain_core.prompts import BasePromptTemplate
from langchain_core.prompts.chat import (
BaseMessagePromptTemplate,
ChatPromptTemplate,
HumanMessagePromptTemplate,
MessagesPlaceholder,
)
from langchain_core.pydantic_v1 import root_validator
from langchain_core.runnables import Runnable, RunnablePassthrough
from langchain_core.tools import BaseTool
from langchain_core.utils.function_calling import convert_to_openai_function
from langchain.agents import BaseSingleActionAgent
from langchain.agents.format_scratchpad.openai_functions import (
format_to_openai_function_messages,
)
from langchain.agents.output_parsers.openai_functions import (
OpenAIFunctionsAgentOutputParser,
)
[docs]@deprecated("0.1.0", alternative="create_openai_functions_agent", removal="1.0")
class OpenAIFunctionsAgent(BaseSingleActionAgent):
"""An Agent driven by OpenAIs function powered API.
Args:
llm: This should be an instance of ChatOpenAI, specifically a model
that supports using `functions`.
tools: The tools this agent has access to.
prompt: The prompt for this agent, should support agent_scratchpad as one
of the variables. For an easy way to construct this prompt, use
`OpenAIFunctionsAgent.create_prompt(...)`
output_parser: The output parser for this agent. Should be an instance of
OpenAIFunctionsAgentOutputParser.
Defaults to OpenAIFunctionsAgentOutputParser.
"""
llm: BaseLanguageModel
tools: Sequence[BaseTool]
prompt: BasePromptTemplate
output_parser: Type[OpenAIFunctionsAgentOutputParser] = (
OpenAIFunctionsAgentOutputParser
)
@root_validator(pre=False, skip_on_failure=True)
def validate_prompt(cls, values: dict) -> dict:
"""Validate prompt.
Args:
values: Values to validate.
Returns:
Validated values.
Raises:
ValueError: If `agent_scratchpad` is not in the prompt.
"""
prompt: BasePromptTemplate = values["prompt"]
if "agent_scratchpad" not in prompt.input_variables:
raise ValueError(
"`agent_scratchpad` should be one of the variables in the prompt, "
f"got {prompt.input_variables}"
)
return values
@property
def input_keys(self) -> List[str]:
"""Get input keys. Input refers to user input here."""
return ["input"]
@property
def functions(self) -> List[dict]:
"""Get functions."""
return [dict(convert_to_openai_function(t)) for t in self.tools]
[docs] def plan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
with_functions: bool = True,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date,
along with observations.
callbacks: Callbacks to use. Defaults to None.
with_functions: Whether to use functions. Defaults to True.
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
If the agent is finished, returns an AgentFinish.
If the agent is not finished, returns an AgentAction.
"""
agent_scratchpad = format_to_openai_function_messages(intermediate_steps)
selected_inputs = {
k: kwargs[k] for k in self.prompt.input_variables if k != "agent_scratchpad"
}
full_inputs = dict(**selected_inputs, agent_scratchpad=agent_scratchpad)
prompt = self.prompt.format_prompt(**full_inputs)
messages = prompt.to_messages()
if with_functions:
predicted_message = self.llm.predict_messages(
messages,
functions=self.functions,
callbacks=callbacks,
)
else:
predicted_message = self.llm.predict_messages(
messages,
callbacks=callbacks,
)
agent_decision = self.output_parser._parse_ai_message(predicted_message)
return agent_decision
[docs] async def aplan(
self,
intermediate_steps: List[Tuple[AgentAction, str]],
callbacks: Callbacks = None,
**kwargs: Any,
) -> Union[AgentAction, AgentFinish]:
"""Async given input, decided what to do.
Args:
intermediate_steps: Steps the LLM has taken to date,
along with observations.
callbacks: Callbacks to use. Defaults to None.
**kwargs: User inputs.
Returns:
Action specifying what tool to use.
If the agent is finished, returns an AgentFinish.
If the agent is not finished, returns an AgentAction.
"""
agent_scratchpad = format_to_openai_function_messages(intermediate_steps)
selected_inputs = {
k: kwargs[k] for k in self.prompt.input_variables if k != "agent_scratchpad"
}
full_inputs = dict(**selected_inputs, agent_scratchpad=agent_scratchpad)
prompt = self.prompt.format_prompt(**full_inputs)
messages = prompt.to_messages()
predicted_message = await self.llm.apredict_messages(
messages, functions=self.functions, callbacks=callbacks
)
agent_decision = self.output_parser._parse_ai_message(predicted_message)
return agent_decision
[docs] def return_stopped_response(
self,
early_stopping_method: str,
intermediate_steps: List[Tuple[AgentAction, str]],
**kwargs: Any,
) -> AgentFinish:
"""Return response when agent has been stopped due to max iterations.
Args:
early_stopping_method: The early stopping method to use.
intermediate_steps: Intermediate steps.
**kwargs: User inputs.
Returns:
AgentFinish.
Raises:
ValueError: If `early_stopping_method` is not `force` or `generate`.
ValueError: If `agent_decision` is not an AgentAction.
"""
if early_stopping_method == "force":
# `force` just returns a constant string
return AgentFinish(
{"output": "Agent stopped due to iteration limit or time limit."}, ""
)
elif early_stopping_method == "generate":
# Generate does one final forward pass
agent_decision = self.plan(
intermediate_steps, with_functions=False, **kwargs
)
if isinstance(agent_decision, AgentFinish):
return agent_decision
else:
raise ValueError(
f"got AgentAction with no functions provided: {agent_decision}"
)
else:
raise ValueError(
"early_stopping_method should be one of `force` or `generate`, "
f"got {early_stopping_method}"
)
[docs] @classmethod
def create_prompt(
cls,
system_message: Optional[SystemMessage] = SystemMessage(
content="You are a helpful AI assistant."
),
extra_prompt_messages: Optional[List[BaseMessagePromptTemplate]] = None,
) -> ChatPromptTemplate:
"""Create prompt for this agent.
Args:
system_message: Message to use as the system message that will be the
first in the prompt.
extra_prompt_messages: Prompt messages that will be placed between the
system message and the new human input.
Returns:
A prompt template to pass into this agent.
"""
_prompts = extra_prompt_messages or []
messages: List[Union[BaseMessagePromptTemplate, BaseMessage]]
if system_message:
messages = [system_message]
else:
messages = []
messages.extend(
[
*_prompts,
HumanMessagePromptTemplate.from_template("{input}"),
MessagesPlaceholder(variable_name="agent_scratchpad"),
]
)
return ChatPromptTemplate(messages=messages) # type: ignore[arg-type, call-arg]
[docs]def create_openai_functions_agent(
llm: BaseLanguageModel, tools: Sequence[BaseTool], prompt: ChatPromptTemplate
) -> Runnable:
"""Create an agent that uses OpenAI function calling.
Args:
llm: LLM to use as the agent. Should work with OpenAI function calling,
so either be an OpenAI model that supports that or a wrapper of
a different model that adds in equivalent support.
tools: Tools this agent has access to.
prompt: The prompt to use. See Prompt section below for more.
Returns:
A Runnable sequence representing an agent. It takes as input all the same input
variables as the prompt passed in does. It returns as output either an
AgentAction or AgentFinish.
Raises:
ValueError: If `agent_scratchpad` is not in the prompt.
Example:
Creating an agent with no memory
.. code-block:: python
from langchain_community.chat_models import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain import hub
prompt = hub.pull("hwchase17/openai-functions-agent")
model = ChatOpenAI()
tools = ...
agent = create_openai_functions_agent(model, tools, prompt)
agent_executor = AgentExecutor(agent=agent, tools=tools)
agent_executor.invoke({"input": "hi"})
# Using with chat history
from langchain_core.messages import AIMessage, HumanMessage
agent_executor.invoke(
{
"input": "what's my name?",
"chat_history": [
HumanMessage(content="hi! my name is bob"),
AIMessage(content="Hello Bob! How can I assist you today?"),
],
}
)
Prompt:
The agent prompt must have an `agent_scratchpad` key that is a
``MessagesPlaceholder``. Intermediate agent actions and tool output
messages will be passed in here.
Here's an example:
.. code-block:: python
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
prompt = ChatPromptTemplate.from_messages(
[
("system", "You are a helpful assistant"),
MessagesPlaceholder("chat_history", optional=True),
("human", "{input}"),
MessagesPlaceholder("agent_scratchpad"),
]
)
"""
if "agent_scratchpad" not in (
prompt.input_variables + list(prompt.partial_variables)
):
raise ValueError(
"Prompt must have input variable `agent_scratchpad`, but wasn't found. "
f"Found {prompt.input_variables} instead."
)
llm_with_tools = llm.bind(functions=[convert_to_openai_function(t) for t in tools])
agent = (
RunnablePassthrough.assign(
agent_scratchpad=lambda x: format_to_openai_function_messages(
x["intermediate_steps"]
)
)
| prompt
| llm_with_tools
| OpenAIFunctionsAgentOutputParser()
)
return agent