haive.agents.reasoning_and_critique.tot.agent¶
Tree of Thoughts (ToT) agent implementation.
This module implements the Tree of Thoughts algorithm as a Haive agent.
Classes¶
Tree of Thoughts agent implementation. |
Functions¶
|
Create evaluator runnable. |
|
Create generator runnable. |
|
Create evaluator engine configuration. |
|
Create generator engine configuration. |
Module Contents¶
- class haive.agents.reasoning_and_critique.tot.agent.ToTAgent(config)¶
Bases:
haive.core.engine.agent.agent.Agent[haive.agents.reasoning_and_critique.tot.config.TOTAgentConfig],Generic[T]Tree of Thoughts agent implementation.
This agent implements the Tree of Thoughts search algorithm, which: 1. Generates candidate solutions 2. Scores the candidates 3. Selects the best candidates for further exploration 4. Repeats until a satisfactory solution is found or max depth reached
The implementation supports parallel processing of candidates through beam search.
Initialize the ToT agent.
- Parameters:
config (haive.agents.reasoning_and_critique.tot.config.TOTAgentConfig) – Configuration for the ToT agent
- setup_workflow()¶
Set up the Tree of Thoughts workflow.
Graph structure: - Sequential mode:
START -> generate -> evaluate_all -> select -> (loop or END)
Parallel mode: START -> generate -> [evaluate_candidate * N] -> collect -> select -> (loop or END)
- Return type:
None
- haive.agents.reasoning_and_critique.tot.agent.compose_evaluator_runnable(engine, use_structured_output=False)¶
Create evaluator runnable.
- haive.agents.reasoning_and_critique.tot.agent.compose_generator_runnable(engine, use_structured_output=False)¶
Create generator runnable.
- haive.agents.reasoning_and_critique.tot.agent.create_evaluator_engine(engine, use_structured_output=False, output_model=None)¶
Create evaluator engine configuration.
- haive.agents.reasoning_and_critique.tot.agent.create_generator_engine(engine, use_structured_output=False, output_model=None)¶
Create generator engine configuration.