⚠️ Deprecated

March 15, 2026 · View on GitHub

This package has been deprecated. The routing logic is maintained as part of the Model Router hosted service, which includes features not present here (auto-routing, analytics, billing, streaming, multi-provider failover).

If you're looking for AI model routing:

This repo is archived for reference only.

Original README

modelrouter-core

Tier-based AI model routing engine — pick economy, standard, or premium and let the router handle the rest.

A standalone, zero-dependency TypeScript library that implements the core routing logic from Model Router. Use it to build your own routing layer, or as a reference for how tier-based model selection works.

Install

npm install modelrouter-core

Quick Start

import { RoutingEngine, TIERS, MODEL_ALIASES } from 'modelrouter-core';

const engine = new RoutingEngine({
  availableProviders: new Set(['openai', 'anthropic']),
  defaultTier: 'standard',
  defaultOutputRatio: 0.33,
});

const decision = engine.selectModel({ messages: [...], model: 'gpt-4o' });
// → { provider: 'openai', model: 'gpt-4.1', tier: 'standard', prefer: 'balanced', ... }

Features

  • Tier-based routing: economy / standard / premium
  • Prefer strategies: balanced, cheap, fast, quality
  • Model aliases: send gpt-4o or claude-sonnet, get routed to the right tier
  • Circuit breaker: automatic provider failover on errors
  • Zero dependencies: pure TypeScript, ~800 lines