from typing import List, Optional
from sherpa_ai.actions import GoogleSearch, SynthesizeOutput
from sherpa_ai.actions.base import BaseAction
from sherpa_ai.agents.base import BaseAgent
from sherpa_ai.config import AgentConfig
from sherpa_ai.memory import Belief
from sherpa_ai.output_parsers.citation_validation import CitationValidation
from sherpa_ai.policies import ReactPolicy
# TODO: QA Agent only contains partial implementation from the original
# task agent, more investigation is needed to add more content to it.
# Some of the feature may be added to the agent base class, such as
# the verbose logger.
[docs]
class QAAgent(BaseAgent):
"""A specialized agent for answering questions and providing information.
This agent is designed to handle question-answering tasks by searching for
information, synthesizing responses, and validating outputs. It can optionally
include citations in its responses.
Attributes:
name (str): The name of the agent, defaults to "QA Agent".
description (str): A description of the agent's purpose and capabilities.
config (AgentConfig): Configuration settings for the agent.
num_runs (int): Number of action execution cycles, defaults to 3.
global_regen_max (int): Maximum number of output regeneration attempts, defaults to 5.
citation_enabled (bool): Whether to include citations in responses, defaults to False.
Args:
custom_task_description_prompt (Optional[str]): Custom prompt for task description. If None, uses default from prompt template.
custom_action_plan_prompt (Optional[str]): Custom prompt for action plan description. If None, uses default from prompt template.
Example:
>>> from sherpa_ai.agents.qa_agent import QAAgent
>>> from sherpa_ai.config import AgentConfig
>>> agent = QAAgent(
... name="Research Assistant",
... config=AgentConfig(),
... citation_enabled=True
... )
>>> print(agent.name)
Research Assistant
>>> # Using custom prompts
>>> agent = QAAgent(
... name="Custom QA",
... custom_task_description_prompt="You are an expert QA assistant.",
... custom_action_plan_prompt="Plan to solve: {task}"
... )
>>> print(agent.description)
You are an expert QA assistant.\n\nYour name is Custom QA.
""" # noqa: E501
name: str = "QA Agent"
description: str = None
config: AgentConfig = None
num_runs: int = 3
global_regen_max: int = 5
citation_enabled: bool = False
def __init__(
self,
*args,
custom_task_description_prompt: Optional[str] = None,
custom_action_plan_prompt: Optional[str] = None,
**kwargs
):
"""Initialize a QA agent with appropriate configuration and policy.
Sets up the agent's description, policy, and belief system using provided
custom prompts or default template prompts from prompts.json.
Args:
*args: Variable length argument list.
custom_task_description_prompt (Optional[str]): Custom prompt for task description.
custom_action_plan_prompt (Optional[str]): Custom prompt for action plan description.
**kwargs: Arbitrary keyword arguments.
Example:
>>> from sherpa_ai.agents.qa_agent import QAAgent
>>> agent = QAAgent(name="Research Assistant")
>>> print(agent.name)
Research Assistant
>>> # With custom prompt
>>> agent = QAAgent(
... name="Custom QA",
... custom_task_description_prompt="Expert QA for all queries."
... )
>>> print(agent.description)
Expert QA for all queries.\n\nYour name is Custom QA.
"""
super().__init__(*args, **kwargs)
if custom_task_description_prompt is not None:
self.description = custom_task_description_prompt
else:
template = self.prompt_template
self.description = template.format_prompt(
prompt_parent_id="qa_agent_prompts",
prompt_id="TASK_AGENT_DESCRIPTION",
version="1.0",
)
if custom_action_plan_prompt is not None:
action_planner = custom_action_plan_prompt
else:
template = self.prompt_template
action_planner = template.format_prompt(
prompt_parent_id="qa_agent_prompts",
prompt_id="ACTION_PLAN_DESCRIPTION",
version="1.0",
)
self.description = self.description + "\n\n" + f"Your name is {self.name}."
if self.policy is None:
self.policy = ReactPolicy(
role_description=self.description,
output_instruction=action_planner,
llm=self.llm,
)
if self.config is None:
self.config = AgentConfig()
if self.belief is None:
self.belief = Belief()
for validation in self.validations:
if isinstance(validation, CitationValidation):
self.citation_enabled = True
break
[docs]
def create_actions(self) -> List[BaseAction]:
"""Create and return the list of actions available to this agent.
This method defines the specific actions that the QA agent can perform,
including Google search for finding information.
Returns:
List[BaseAction]: List of action objects that the agent can use.
Example:
>>> from sherpa_ai.agents.qa_agent import QAAgent
>>> agent = QAAgent()
>>> actions = agent.create_actions()
>>> print(len(actions))
1
>>> print(actions[0].__class__.__name__)
GoogleSearch
"""
return [
GoogleSearch(
role_description=self.description,
task=self.belief.current_task.content
if self.belief.current_task
else "",
llm=self.llm,
config=self.config,
belief=self.belief,
),
]
[docs]
def synthesize_output(self) -> str:
"""Generate the final answer based on the agent's actions and belief state.
This method creates a SynthesizeOutput action and executes it with the
current task, context, and internal history to produce a coherent response.
Returns:
str: The synthesized answer to the question.
Example:
>>> from sherpa_ai.agents.qa_agent import QAAgent
>>> agent = QAAgent()
>>> agent.belief.current_task.content = "What is machine learning?"
>>> # In a real scenario, this would generate a response based on
>>> # the agent's actions and belief state
>>> # result = agent.synthesize_output()
>>> # print(result)
"""
synthesize_action = SynthesizeOutput(
role_description=self.description,
llm=self.llm,
add_citation=self.citation_enabled,
)
result = synthesize_action.execute(
self.belief.current_task.content,
self.belief.get_context(self.llm.get_num_tokens),
self.belief.get_internal_history(self.llm.get_num_tokens),
)
return result
