Product schematicAI control layerOpenAI · Anthropic · Google

Change the URL, keep your tools, see your routes.

RouterLab sits between your applications, agents, workflows, and model routes. It authenticates the request, applies the selected route, then attaches the response to usage and credits.

Product schematic

RouterLab integration schematic

VISIBLERouterLab API · CH / DE

Existing

Application

OpenAI SDK

Existing

AI agent

Claude Messages API

Existing

Workflow

Job / automation

Replace the base URL

from

api.openai.com/v1

to

api.routerlab.ch/v1

ROUTERLAB

API

Routes · credits · statuses

OpenAI

External provider route

Anthropic

External provider route

Google

External provider route

RouterLab

RouterLab-hosted open-source models · CH / DE

Dashboard usage

Visible response

Illustrative example

Usage

1,248 req

Credits

CHF 42.80

Response

back to client

After request

usage + credits

Concepts

Model, API format and processing provider are not the same thing.

These three levels answer three different questions: what to use, how to call it, and who executes the request.

DEFINITION

Model

The model or model family requested.

Catalog examples

GPTClaudeGeminiDeepSeekKimiGLMMiniMaxQwenGrok
DEFINITION

API format

The protocol the client uses to call RouterLab.

Catalog examples

OpenAI-compatible APIClaude Messages APIEmbeddings API
DEFINITION

Processing provider

The infrastructure that actually executes the request.

Catalog examples

OpenAIAnthropicGoogleRouterLab

Before / with RouterLab

The difference is not a new app. It is a control point.

The homepage explains why RouterLab exists. This page shows what changes in the technical flow.

Before

Without a central layer

  • Separate provider SDKs and keys
  • Costs scattered across tools
  • Routes are hard to explain
RouterLab

With RouterLab

  • One RouterLab URL on the client side
  • Selected route and destination are explicit
  • Credits, usage, and catalog are visible

Detailed flow

What happens when a request goes through RouterLab.

The sequence stays short: keep integration simple, make routing and usage observable.

01

Client / SDK

Your application, agent, or compatible SDK sends the request.

02

RouterLab API

RouterLab authenticates the key and receives the request.

03

Selected route

The selected route determines the format and destination.

04

Provider / actual infrastructure

The request is processed by the provider or infrastructure specified by the route.

05

Response + usage / credits

The response returns to the client; RouterLab attaches usage and credits.

Endpoint and compatibility

Keep your compatible SDKs and point them to RouterLab.

RouterLab exposes compatible formats. The change is the base URL and RouterLab key; the exact path then depends on the SDK you use.

https://api.routerlab.ch/v1

OpenAI SDKClaude Messages APICH / DE

Change the entry point, then read routes and usage instead of guessing what happened.

SDKs can use a slightly different base URL because each one builds its own API path. They still reach the same RouterLab layer.

  • Keep the compatible SDK
  • Targeted base URL change
  • Same request, route and response remain readable
Compatible

OpenAI SDK

https://api.routerlab.ch/v1

Compatible
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.routerlab.ch/v1",
  apiKey: process.env.ROUTERLAB_API_KEY,
});

await client.chat.completions.create({
  model: "model-id-from-/v1/models",
  messages: [{ role: "user", content: "Route this request" }],
});

Claude Messages API

https://api.routerlab.ch

Compatible
import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  baseURL: "https://api.routerlab.ch",
  apiKey: process.env.ROUTERLAB_API_KEY,
});

await client.messages.create({
  model: "model-id-with-claude-messages-route",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Route this request" }],
});

Control

What RouterLab makes visible.

The modules are concrete: keys, credits, usage, and catalog. They operate AI traffic without adding an automatic-routing promise.

Keys

One RouterLab key

Apps keep one stable entry point without exposing every provider key.

Credits

Readable balance

Cost is attached to requests and the credit ledger.

Usage

Observed routes

The dashboard shows the route, format, and cost associated with the request.

Catalog

Published models

Models, formats, and statuses remain visible when choosing a route.

Governance

Routes stay readable through to the actual infrastructure.

The catalog separates model family, API format, and processing path. RouterLab does not present a provider as if it were a model.

RouterLab

What RouterLab controls

The control layer, published routing, and visibility.

  • Authentication and API keys
  • Published model routes
  • Usage visibility
  • Credits and billing
  • RouterLab infrastructure for open-source models hosted in Switzerland / Germany
Provider

What the provider controls

Each provider operates its infrastructure and models.

  • Model infrastructure and availability
  • Provider-side request processing
  • Model behavior
  • Model lifecycle and updates

Route table

How to read a route

Model / familyAPI formatProcessing / hostingWhat it means
ClaudeClaude Messages APIExternal providerClaude is a model family called with the Messages format; infrastructure depends on the catalog route.
GPTOpenAI-compatible APIExternal providerGPT is a model family; OpenAI-compatible describes the protocol, not the final infrastructure by itself.
GeminiOpenAI-compatible APIExternal providerGemini is a model family; the call format and processing provider are separate pieces of information.
Open-source familiesOpenAI-compatible APIRouterLab when hosted; otherwise external providerOpen source describes the model family or type; hosting must be read route by route.
Embedding modelsEmbeddings APIProcessing follows the catalog routeEmbedding models use a specialized route and are not chat models.

A route can target an external provider or host an open-source model with RouterLab. Hosting and processing are read route by route; see the Trust page for relevant infrastructure information.

Trial

Try the RouterLab endpoint with your tools.

Start with a trial key, then use the catalog to choose the right routes.