NMLR SDK (Node.js/TypeScript)

Overview

NMLR SDK (Node.js/TypeScript) is the JavaScript/TypeScript implementation of the reasoning orchestration layer. It mirrors the functionality of the Python SDK and provides:

1. Candidate Generation: Generate multiple reasoning paths
2. Optional Robustness Scoring: Evaluate candidates with counterpoints
3. Synthesis: Merge top candidates into final answer

This SDK is provider-agnostic - users provide their own AI API keys.

Installation

npm install nmlr-sdk

Quick Start

import { NMLREngine, AnthropicAdapter } from 'nmlr-sdk';

// Initialize with your API key
const adapter = new AnthropicAdapter(
  'sk-ant-your-key-here',
  'claude-3-sonnet-20240229'
);

// Create NMLR engine
const nmlr = new NMLREngine(adapter, { nPaths: 3 });

// Ask a question
const result = await nmlr.ask('What are the key considerations for distributed systems?');

// Access results
console.log(result.answer);
console.log(result.confidence);
console.log(result.scores);

API Reference

Core Classes

| Class | Description | |-------|-------------| | LLMAdapter | Abstract base class for LLM providers | | MockAdapter | Mock adapter for testing | | AnthropicAdapter | Adapter for Anthropic Claude | | NMLREngine | Core orchestration engine |

NMLREngine Methods

class NMLREngine {
  constructor(adapter: LLMAdapter, config?: NMLREngineConfig);
  
  // Main entry point
  ask(query: string): Promise<NMLRResult>;
  
  // Generate candidate paths
  generateCandidates(query: string): Promise<Candidate[]>;
  
  // Evaluate candidate robustness
  robustnessScore(candidate: Candidate, query: string): Promise<CandidateScore>;
  
  // Merge top candidates
  synthesize(candidates: Candidate[], query: string): Promise<string>;
  
  // Extensibility
  setNPaths(nPaths: number): void;
  addScoringAxis(name: string, description: string): void;
  registerCustomEvaluator(fn: Function): void;
}

Examples

See examples/basicUsage.ts for complete examples.

With Environment Variable

export ANTHROPIC_API_KEY="sk-ant-..."
npx ts-node examples/basicUsage.ts

Using Mock Adapter

import { NMLREngine, MockAdapter } from 'nmlr-sdk';

const adapter = new MockAdapter();
const nmlr = new NMLREngine(adapter);
const result = await nmlr.ask('Test question');
console.log(result.answer);

Requirements

• Node.js 18+
• TypeScript 5.0+ (for development)

License

MIT License - See LICENSE for details.

Disclaimer

This SDK is a reasoning orchestration layer that works with user-provided API keys. It does not host or supply AI access - users must provide their own keys from their chosen LLM provider.