agents.planning.rewoo_v3.agentΒΆ
ReWOO V3 Agent using Enhanced MultiAgent V3 coordination.
This module implements the ReWOO (Reasoning WithOut Observation) methodology using our proven patterns from Plan-and-Execute V3 success.
ReWOO Architecture: 1. Planner: Creates complete reasoning plan with evidence placeholders 2. Worker: Executes all tool calls to collect evidence 3. Solver: Synthesizes all evidence into final answer
Key advantages: - Token efficiency (5x improvement over iterative methods) - Parallel tool execution capability - Robust to partial failures - Fine-tuning friendly modular design
ClassesΒΆ
ReWOO V3 Agent using Enhanced MultiAgent V3 coordination. |
Module ContentsΒΆ
- class agents.planning.rewoo_v3.agent.ReWOOV3Agent(name, config, tools=None, max_steps=10, **kwargs)ΒΆ
ReWOO V3 Agent using Enhanced MultiAgent V3 coordination.
Implements ReWOO (Reasoning WithOut Observation) methodology: - Separates planning, execution, and synthesis phases - Plans complete solution upfront without tool observation - Executes all tool calls in batch/parallel - Synthesizes all evidence together for final answer
This provides significant efficiency gains over traditional iterative agent approaches while maintaining high solution quality.
Initialize ReWOO V3 Agent.
- Parameters:
name (str) β Agent identifier
config (haive.core.engine.aug_llm.AugLLMConfig) β Base LLM configuration for all sub-agents
tools (list | None) β Available tools for worker agent execution
max_steps (int) β Maximum planning steps allowed
**kwargs β Additional configuration for Enhanced MultiAgent V3
- async arun(query, context=None, max_steps=None, tools_preference=None, **kwargs)ΒΆ
Execute ReWOO V3 workflow asynchronously.
- Parameters:
- Returns:
Structured output with complete ReWOO results and metadata
- Return type:
- run(query, context=None, max_steps=None, tools_preference=None, **kwargs)ΒΆ
Synchronous wrapper for ReWOO V3 execution.
- Parameters:
- Returns:
Structured output with ReWOO results
- Return type: