agentlib-reasoning

A collection of sophisticated reasoning engines that define how an agent processes information and arrives at an answer.

Included Engines

• ReactEngine: Implements the "Reasoning and Acting" loop, allowing agents to interleave thought processes with tool usage.
• ChainOfThoughtEngine (CoT): Forces the model to generate a sequence of internal thoughts before the final response.
• ReflectEngine: A multi-pass engine that generates an initial answer, critiques it, and then revises it.
• PlannerEngine: Breaks the user request into a sequence of atomic sub-tasks (a plan) and executes them sequentially.
• AutonomousEngine: An open-ended loop where the agent continues to work until it explicitly calls a "finish" tool.

Usage

use agentlib_reasoning::ReactEngine;
use std::sync::Arc;

let agent = agent.engine(Arc::new(ReactEngine::default()));

Designing Processes

All engines implement the ReasoningEngine trait and are fully compatible with the agentlib-core runtime. Each step taken by an engine is recorded as a ReasoningStep for full observability.