routing.runDocs
Get started

Overview›Introduction

Introduction

routing.run gives you one interface for the models in its live catalog across multiple providers through one predictable REST API.

Every provider has its own details: authentication methods, rate limits, model naming, streaming protocols, response formats, and error shapes. routing.run handles that provider-specific work for you. You connect your API key once, then call any supported model with the same request pattern across all providers.

That is the point: when you authenticate, stream responses, set generation parameters, or handle errors, you do it the routing.run way. Your code calls routing.run. routing.run monitors configured routes, translates the request for the selected model, and returns a consistent OpenAI-compatible response.

Start here

You need a routing.run account and an API key.

  1. Sign up at routing.run.
  2. Activate Basic in Billing to receive monthly Routing credit.
  3. Create an API key in API Keys.
  4. Browse the Model Catalog for available models and pricing.
  5. Open the Quick Start guide to send your first request.
  6. Explore the API Reference for full endpoint documentation.

Keep your API key on your server. Do not ship it in browser code, mobile apps, public repositories, or logs.

Works with your tools

routing.run speaks the OpenAI and Anthropic APIs, so it drops straight into the coding agents and chat harnesses you already use — pick one, set the base URL and your key, and choose a currently available model from the catalog.

Point Claude Code at routing.run via the Anthropic-compatible base URL.

Environment

shell
export ANTHROPIC_BASE_URL="https://api.routing.run"export ANTHROPIC_API_KEY="rk_live_your_api_key" claude

First request shape

All API requests use this base URL pattern:

https://api.routing.run/v1/{endpoint}

For example:

https://api.routing.run/v1/chat/completions

Use POST for chat completions and embeddings, GET for listing and retrieving models.

What routing.run handles

  • Provider routing — Automatically selects the best available provider for each model.
  • Failover — Can retry a failed attempt on another configured route when it is safe to do so.
  • Format compatibility — Supports both OpenAI and Anthropic-compatible request formats.
  • Route health — Uses provider health and circuit breaking when selecting configured routes.
  • Streaming — Normalizes SSE streaming across all supported providers.

What it's not for

routing.run is not a model host — it routes requests to existing inference providers. It does not fine-tune models, store conversation data, or provide training infrastructure.