AutoRouter
RoutingFeaturesDeploy
Source Launch
Open source LLM traffic control for production apps

Route AI traffic like infrastructure.

AutoRouter gives teams one hardened endpoint for providers, fallbacks, model policy, access tokens, and request telemetry. Your app stays simple while the routing layer gets smarter.

Open dashboardView source
9+
providers
1
API surface
MIT
license
app
12 req/s
worker
queued
OpenAI
28ms
Claude
fallback
Gemini
standby
99.9%
uptime
42ms
p95
$0.18
saved
OpenAIAnthropicGeminiGroqOllamaCohereMistralAzureDeepSeekOpenAIAnthropicGeminiGroqOllamaCohereMistralAzureDeepSeekOpenAIAnthropicGeminiGroqOllamaCohereMistralAzureDeepSeek

Operating model

Three moves from scattered keys to a routing layer.

01

Connect providers

Add the keys you already use and keep credentials away from product code.

02

Design routes

Choose primary models, fallbacks, and token rules for each workload.

03

Ship one endpoint

Send traffic through AutoRouter and let the control plane handle recovery.

Capabilities

Everything between your product and the model vendors.

Built for self-hosted teams that want explicit routing policy, not another opaque proxy.

Latency-aware routing

Send each request to the fastest healthy model without changing client code.

Provider control plane

Keep OpenAI, Anthropic, Gemini, Groq, Ollama, and custom endpoints behind one console.

Fallback chains

Promote recovery paths to first-class configuration instead of handling outages in app code.

Scoped access tokens

Issue keys that can only reach the providers, models, and routes you approve.

Request telemetry

Review costs, tokens, latency, status, and provider behavior from one audit trail.

OpenAI-compatible API

Point existing SDKs at AutoRouter and keep the interface your apps already understand.

Developer Experience

Drop-in compatible. Zero friction.

Change one line of configuration and keep using the same SDKs, types, and request patterns your team already knows.

Same request / response shapes
Streaming support out of the box
TypeScript types included
app.ts
1class="text-chart-2 font-medium">import OpenAI class="text-chart-2 font-medium">from class="text-primary">'openai'
2
3class="text-chart-2 font-medium">const client = new OpenAI({
4 baseURL: class="text-primary">'https://api.autorouter.io/v1',
5 apiKey: process.env.AR_TOKEN,
6})
7
8class="text-chart-2 font-medium">const response = class="text-chart-2 font-medium">await client.chat.completions.create({
9 model: class="text-primary">'router/smart',
10 messages: [{ role: class="text-primary">'user', content: class="text-primary">'Hello!' }],
11 // AutoRouter picks the best provider,
12 // handles fallbacks, and reports telemetry.
13})

Unified API

One URL. Many models. Fewer production surprises.

AutoRouter exposes an OpenAI-compatible endpoint, so existing SDKs keep working while you centralize provider policy.

Drop-in client migration

Change the base URL once and keep request payloads familiar.

Centralized secret handling

Applications receive scoped tokens instead of raw provider credentials.

Observable by default

Route decisions and provider health become visible operational data.

Self-hosted, open source, production-minded

Put routing policy where it belongs: outside your app code.

Start with one provider, then add fallback paths, access boundaries, and telemetry as your traffic grows.

Get started
© 2026 AutoRouterOpenAI-compatible routing for self-hosted teams.