nat.plugins.langchain.agent.rewoo_agent.agent#
Attributes#
Classes#
State schema for the ReWOO Agent Graph |
|
Configurable ReWOO Agent. |
Module Contents#
- logger#
- _REWOO_PLACEHOLDER_PATTERN#
- class ReWOOPlanStep(/, **data: Any)#
Bases:
pydantic.BaseModel- evidence: ReWOOEvidence#
- class ReWOOGraphState(/, **data: Any)#
Bases:
pydantic.BaseModelState schema for the ReWOO Agent Graph
Create a new model by parsing and validating input data from keyword arguments.
Raises [
ValidationError][pydantic_core.ValidationError] if the input data cannot be validated to form a valid model.selfis explicitly positional-only to allowselfas a field name.- task: langchain_core.messages.human.HumanMessage = None#
- plan: langchain_core.messages.ai.AIMessage = None#
- steps: langchain_core.messages.ai.AIMessage = None#
- evidence_map: dict[str, ReWOOPlanStep] = None#
- result: langchain_core.messages.ai.AIMessage = None#
- class ReWOOAgentGraph(
- llm: langchain_core.language_models.BaseChatModel,
- planner_prompt: langchain_core.prompts.chat.ChatPromptTemplate,
- solver_prompt: langchain_core.prompts.chat.ChatPromptTemplate,
- tools: list[langchain_core.tools.BaseTool],
- use_tool_schema: bool = True,
- callbacks: list[langchain_core.callbacks.base.AsyncCallbackHandler] | None = None,
- detailed_logs: bool = False,
- log_response_max_chars: int = 1000,
- tool_call_max_retries: int = 3,
- raise_tool_call_error: bool = True,
Bases:
nat.plugins.langchain.agent.base.BaseAgentConfigurable ReWOO Agent.
- Args:
detailed_logs: Toggles logging of inputs, outputs, and intermediate steps.
- planner_prompt#
- solver_prompt#
- tools_dict#
- tool_call_max_retries = 3#
- raise_tool_call_error = True#
- static _get_current_level_status(state: ReWOOGraphState) tuple[int, bool]#
Get the current execution level and whether it’s complete.
- Args:
state: The ReWOO graph state.
- Returns:
tuple of (current_level, is_complete). Level -1 means all execution is complete.
- static _parse_planner_output(planner_output: str) list[ReWOOPlanStep]#
- static _parse_planner_dependencies(
- steps: list[ReWOOPlanStep],
Parse planner steps to identify dependencies and create execution levels for parallel processing. This creates a dependency map and identifies which evidence placeholders can be executed in parallel.
- Args:
steps: list of plan steps from the planner.
- Returns:
A mapping from evidence placeholders to step info and execution levels for parallel processing.
- static _replace_placeholders( ) Any#
Replace ReWOO evidence placeholders in a planner-provided tool input.
Exact placeholder matches preserve the evidence output type so tools with structured schemas can receive lists, objects, or numeric values. Partial string matches replace only complete ReWOO placeholder tokens.
- static _parse_tool_input(tool_input: Any)#
- async planner_node(state: ReWOOGraphState)#
- async executor_node(state: ReWOOGraphState)#
Execute tools in parallel for the current dependency level.
This replaces the sequential execution with parallel execution of tools that have no dependencies between them.
- async _execute_single_tool(
- placeholder: str,
- step_info: ReWOOPlanStep,
- intermediate_results: dict[str, langchain_core.messages.tool.ToolMessage],
Execute a single tool with proper placeholder replacement.
- Args:
placeholder (str): The evidence placeholder (e.g., “#E1”). step_info (ReWOOPlanStep): Step information containing tool and tool_input. intermediate_results (dict[str, ToolMessage]): Current intermediate results for placeholder replacement.
- Returns:
ToolMessage with the tool execution result.
- async solver_node(state: ReWOOGraphState)#
- async conditional_edge(state: ReWOOGraphState)#
- async build_graph()#