pangukitsappdev.agent package¶

Submodules¶

pangukitsappdev.agent.agent_action module¶

class pangukitsappdev.agent.agent_action.AgentAction(*, req: Optional[str] = None, resp: Optional[str] = None, thought: Optional[str] = None, action_json: Optional[str] = None, action: Optional[str] = None, action_input: Optional[Any] = None, observation: Optional[str] = None, user_feedback: Optional[str] = None)¶

基类：BaseModel

Agent执行的action Attributes:

req: LLM的原始请求 resp: LLM的原始响应 thought: LLM的思考 action_json: LLM输出的原始Action字符串 action: LLM将要采取的行动，即Tool action_input: 行动的输入，即Tool的输入 observation: 采取行动后给LLM观察的结果，即Tool的执行结果 user_feedback: 用户对该步骤的反馈，可能需要更改工具调用参数

action: Optional[str]¶

action_input: Optional[Any]¶

action_json: Optional[str]¶

observation: Optional[str]¶

req: Optional[str]¶

resp: Optional[str]¶

thought: Optional[str]¶

user_feedback: Optional[str]¶

pangukitsappdev.agent.agent_session module¶

class pangukitsappdev.agent.agent_session.AgentSession(*, messages: List[ConversationMessage], session_id: str, current_action: Optional[AgentAction] = None, agent_session_status: str, is_by_step: Optional[bool] = None, current_message: Optional[ConversationMessage] = None)¶

基类：BaseModel

Agent运行Session，包含历史Action，当前Action，状态 Attributes:

messages: 本次session的用户的输入 session_id: UUID，在一个session内唯一 current_action: 当前Action agent_session_status: Agent状态 is_by_step: 是否是逐步执行 current_message: 当前的AssistantMessage

agent_session_status: str¶

current_action: Optional[AgentAction]¶

current_message: Optional[ConversationMessage]¶

is_by_step: Optional[bool]¶

messages: List[ConversationMessage]¶

session_id: str¶

pangukitsappdev.agent.react_agent module¶

class pangukitsappdev.agent.react_agent.ReactAgent(llm: LLMApi)¶

基类：AbstractAgent

DEFAULT_SYS_PROMPT = '您是一个智能助手，尽可能有帮助和准确地回答用户提出的问题'¶

PLACEHOLDER_ACTION = 'Action:'¶

PLACEHOLDER_OBSERVATION = 'Observation:'¶

PLACEHOLDER_THOUGHT = 'Thought:'¶

get_action(answer: str) → AgentAction¶

static get_agent_action(action_str: str) → AgentAction¶

get_tool_desc() → str¶

get_tool_names() → str¶

react(agent_session: AgentSession)¶: 迭代解决问题 :param agent_session: 历史迭代几率

pangukitsappdev.agent.react_pangu_agent module¶

class pangukitsappdev.agent.react_pangu_agent.ReactPanguAgent(llm: LLMApi)¶