Minimal LLM Router
Multi-model fallback chain. Zero dependencies. Works everywhere.

Quick Start · Features · Why Styrr? · Ecosystem

---

What Is Styrr?

Styrr (Old Norse: "rudder") steers your LLM requests to the right model. If one model fails (rate limit, timeout, error), it automatically tries the next. One consistent API for OpenAI, OpenRouter, Bedrock, Ollama — or any OpenAI-compatible endpoint.

import { StyrRouter } from 'styrr';

const router = new StyrRouter({
  apiKey: process.env.OPENROUTER_API_KEY!,
  models: [
    { id: 'nvidia/nemotron-3-super-120b:free' },
    { id: 'meta-llama/llama-3.3-70b-instruct:free' },
    { id: 'qwen/qwen3-coder:free' },
  ],
});

const result = await router.prompt('Explain FinOps in 2 sentences.');
console.log(result.text);          // "FinOps is..."
console.log(result.modelUsed);     // which model responded
console.log(result.latencyMs);     // how long it took
console.log(result.fallbacksTried); // 0 if primary worked

Install

npm install styrr

Quick Start

1. Install

npm install styrr

2. Route your first prompt

import { StyrRouter } from 'styrr';

const router = new StyrRouter({
  apiKey: process.env.OPENROUTER_API_KEY!,
  models: [
    { id: 'meta-llama/llama-3.3-70b-instruct:free' },
  ],
});

const result = await router.prompt('Hello!');
console.log(result.text);

Features

Feature	Description
Multi-model fallback	If model 1 returns 429/5xx, automatically tries model 2, 3, etc.
Fail-fast on auth errors	401/400 throws immediately — don't retry with different model
Structured JSON output	Auto-parses JSON responses, strips markdown fences
Tool calling	Pass tool schemas, get parsed tool_calls back
Timeout per model	AbortSignal.timeout per call
Zero dependencies	Just fetch() — works anywhere
Observable	onFallback and onAllFailed hooks

Advanced Usage

With tools (function calling)

const result = await router.call(messages, {
  tools: [{
    type: 'function',
    function: {
      name: 'get_weather',
      description: 'Get current weather',
      parameters: { type: 'object', properties: { city: { type: 'string' } } }
    }
  }]
});

if (result.toolCalls) {
  console.log(result.toolCalls[0].name);      // 'get_weather'
  console.log(result.toolCalls[0].arguments); // { city: 'Lima' }
}

With observability hooks

const router = new StyrRouter({
  apiKey: '...',
  models: [...],
  onFallback: (failed, error, next) => {
    console.warn(`[Styrr] ${failed} failed (${error}), trying ${next}`);
  },
  onAllFailed: (errors) => {
    console.error('[Styrr] All models exhausted:', errors);
  },
});

Custom providers (Bedrock, Ollama, etc.)

const router = new StyrRouter({
  apiKey: 'not-used',
  models: [
    { id: 'llama3.2', baseUrl: 'http://localhost:11434/v1', provider: 'ollama' },
    { id: 'gpt-4o', baseUrl: 'https://api.openai.com/v1', apiKey: 'sk-...' },
  ],
});

Bedrock AgentCore cross-provider fallback

Pattern for using Styrr as the routing layer inside a Bedrock AgentCore deployment, with ordered fallback by budget + latency (Bedrock → external → free). Lazy-imports the AWS SDK, so the package stays zero-dependency:

const router = new StyrRouter({
  apiKey: process.env.OPENROUTER_API_KEY!,
  models: [
    { id: 'anthropic.claude-3-sonnet-20240229-v1:0', provider: 'bedrock' },
    { id: 'openai/o1', provider: 'openrouter' },
    { id: 'meta-llama/llama-3.3-70b-instruct:free', provider: 'openrouter' },
  ],
});

See docs/bedrock-agentcore.md for the full pattern (IAM auth, AgentCore tool integration, Sayay budget-aware degrade).

Why Styrr?

Feature	Styrr	LiteLLM	OpenRouter
Zero dependencies	✅	❌	N/A
Self-hosted	✅	✅	❌
Works in CF Workers	✅	❌	N/A
Fallback chain	✅	✅	❌
Tool calling	✅	✅	✅
Cost-aware routing	🔜	❌	❌
Size	~5KB	~500KB	—

Ecosystem

Package	Role	npm
Styrr	LLM router (this)	styrr
Sayay	Cost guardrails	GitHub
Tinkuy	Agent framework	@carloscortezcloud/tinkuy-agent
Qhaway	Agent observability	@carloscortezcloud/qhaway
TideRAG	Edge RAG pipeline	@carloscortezcloud/tiderag

License

Apache 2.0 — see LICENSE.

---

Built by engineers who got tired of vendor lock-in.
Tinkuy Labs · finoptix.dev

_{Works on free models. Zero deps. Ship it.}