@fengrru/reasoning-search

MCTS tree search reasoning engine for AI agents.

Install

npm install @fengrru/reasoning-search

Quick Start

import { ReasoningSearch } from "@fengrru/reasoning-search"

const search = new ReasoningSearch({
  provider: llmProvider,
  strategy: "mcts",
})

const result = await search.search({
  prompt: "Solve this math problem",
  maxIterations: 100,
})

console.log(result.bestPath) // reasoning chain
console.log(result.score) // confidence score

Strategies

| Strategy | Description | |----------|-------------| | mcts | Monte Carlo Tree Search | | standard_sampling | Single generation | | guided_beam_search | Beam search with heuristic | | importance_sampling | Weighted random sampling |

Features

• UCT value: exploration vs exploitation
• Softmax rewards: temperature-based sampling
• Adaptive floor: decreases with depth
• Completion detection: math/code/logic patterns

Documentation

• API Reference — TypeDoc-generated API docs
• Source Code
• Examples

License

MIT