RLM-NodeJs (@yuemi-development/rlm-js)

Recursive LLM (RLM) is a library designed to solve complex problems by breaking them down into simpler sub-questions, solving each independently, and synthesizing the results back into a coherent final answer.

This approach is highly effective for "small" LLMs that might struggle with large contexts or multifaceted reasoning tasks.

Features

• Recursive Decomposition: Automatically breaks complex prompts into manageable sub-tasks.
• Parallel Solving: Solves sub-questions in parallel to optimize speed.
• Provider Agnostic: Support for OpenAI, Anthropic, DeepSeek, OpenRouter, and custom providers.
• Load Balancing: Round-robin distribution across multiple LLM endpoints.
• Token Budgeting: Global token limit enforcement to prevent runaway costs.
• Progress Tracking: Real-time labels for decomposition, solving, and synthesis phases.

Installation

npm install @yuemi-development/rlm-js

Quick Start

const { LLMService, RLMEngine, createOpenAIProvider } = require('@yuemi-development/rlm-js')

// 1. Setup LLM Service
const llm = new LLMService()
llm.register('openai', createOpenAIProvider('your-api-key'))

// 2. Initialize Engine
const engine = new RLMEngine(llm, {
  maxDepth: 3,
  maxSubQuestions: 3,
  maxTokens: 50000,
  strategy: 'round-robin'
})

// 3. Solve!
const result = await engine.solve('Explain the nuances of quantum entanglement.')
console.log(result.answer)

Configuration

The RLMEngine accepts an options object:

• maxDepth: Max levels of recursion.
• maxSubQuestions: Max sub-questions generated per step.
• maxTokens: Global token limit for the entire process.
• strategy: 'fixed' (default) or 'round-robin'.
• providerPool: Optional array of provider IDs for load balancing.