llm-retry-chain

Retry failed LLM calls with automatic provider fallback, circuit breaker, and exponential backoff — zero dependencies.

Install

npm install llm-retry-chain

Quick Start

const { RetryChain } = require('llm-retry-chain');

const chain = new RetryChain([
  { name: 'openai', fn: (prompt) => callOpenAI(prompt) },
  { name: 'anthropic', fn: (prompt) => callAnthropic(prompt) },
  { name: 'local', fn: (prompt) => callOllama(prompt) },
]);

// Tries openai first, falls back to anthropic, then local
const result = await chain.call('Explain quantum computing');
// → { text: "...", provider: "openai", attempts: 1, latency: 450 }

With Circuit Breaker

const chain = new RetryChain(providers, {
  maxRetries: 2,
  circuitBreaker: {
    threshold: 3,       // open after 3 consecutive failures
    recoveryMs: 30000,  // try again after 30s
  },
  backoff: {
    initialMs: 100,
    maxMs: 5000,
    multiplier: 2,
  },
  onFallback: (from, to, error) => {
    console.log(`Falling back from ${from} to ${to}: ${error.message}`);
  },
});

Features

• Provider chain: Ordered fallback across any number of providers
• Circuit breaker: Skips known-failing providers automatically
• Exponential backoff: Configurable retry delays
• Latency tracking: Know which provider responded
• Callbacks: onFallback, onSuccess, onExhausted hooks

License

MIT