The universal AI model router

One endpoint for every model.

Every model, one URL — resilient by default.

AnyRouter is a developer-first AI gateway. We eliminate vendor lock-in and hand you a single OpenAI-compatible endpoint for every major model — engineered for better prices, higher uptime, and production-grade reliability. Pool your own keys and subscriptions behind it, and let automatic failover keep your agents running when any one provider doesn't.

Start free Why AnyRouter

@anyrouterdev AnyRouter

1 URLto reach every provider

100sof models, one catalog

$4monthly credit on Go

0client rewrites needed

One key reaches every major provider

OpenAI

Anthropic

Google

Everything the gateway does

One URL, every model

A single OpenAI- and Anthropic-compatible endpoint reaches hundreds of models across providers.

Auto-failover

401, 429, and 5xx errors transparently retry on the next upstream so your run never stalls.

BYOK + subscription pooling

Pool your own provider keys and paid plans behind one gateway. BYOK never spends platform credits.

Better prices

Transparent per-token pricing, BYOK at no markup, and a curated set of free models on Go — route real traffic for a dollar or a donated key.

Higher uptime

Continuous upstream monitoring plus cross-provider fallback keeps the gateway answering when one provider doesn't.

Swap models, no code change

Presets and virtual models let you change the underlying model from your account, not your codebase.

CLI & agent launchers

anyr claude points Claude Code, Codex, or Cursor at the gateway — config injects on demand, then wipes clean.

Private by default

Prompt and completion bodies are never logged unless you opt in. Secrets are encrypted at rest with per-row salts.

MCP server built in

Manage keys, credits, and the catalog from any MCP client — the gateway speaks the protocol natively.

Architecture

How the gateway routes a request

One endpoint fans out to whichever provider serves the model, with every call logged and cached alongside it.

For coding agents

Built for coding agents

Switch the model behind Claude Code, Codex, or Cursor without editing a single config file — from the command line or straight from your code.

The CLI

anyr claude launches your coding agent on the gateway. Pass a flag to swap models mid-project — same agent, different model, no config surgery. Your keys and presets inject on demand, then wipe clean on exit.

anyr claude --model z-ai/glm-4.7-flash

The SDK

@anyr/sdk and the Vercel AI SDK provider @anyr/ai-sdk-provider let your own agents reach every model through one client — change the model id in code and routing, failover, and billing follow.

createAnyRouter().chat("anthropic/claude-opus-4")

AI-first

A gateway that improves itself

AnyRouter is an AI-first, self-improving development project: built and maintained by AI coding agents running on the very gateway they ship. New providers, fixes, and features land continuously as those agents test, deploy, and refine the platform — so the catalog stays current and the routing keeps getting better without waiting on a release cycle.

Why we exist

The model landscape changes every week — new providers, new prices, new rate limits. Wiring each one into your stack by hand is wasted effort that breaks the moment something shifts. AnyRouter absorbs that churn so you can treat models as interchangeable building blocks: pick the best one for the job, and let the gateway handle the keys, fallbacks, and plumbing underneath.

Principles

What we stand for

Private by default

We don't log prompt or completion bodies unless you opt in per account. Secrets are encrypted at rest with per-row salts.

Open and compatible

We speak the OpenAI, Anthropic, and Responses APIs you already use. No lock-in — point your client elsewhere any time.

Cheap to start

Bring your own keys at no markup, or top up and pay per token. Add a $4 monthly credit and free models with Go — $2/mo, or free when you donate a provider key.

Route your first request in 2 minutes

Run anyr claude or point your existing client at our base URL — one key, every model, automatic failover.

Start free Read the docs