How the router works

What this is

An AI router, not a token shop. You attach the Claude / ChatGPT / Kimi / Grok subscriptions and API keys you already pay for. Those become backends. Agents talk to one gateway URL. We translate protocols, map models, fail over, and log usage. Quota stays on the provider you connected.

Multiple accounts

Every user has their own backends, endpoints, keys, logs, and connectors. Two people can both name an endpoint main. The client key decides whose main is hit. Operators can disable an account (dashboard + that account’s keys). They cannot open another user’s OAuth tokens.

New accounts can use the dashboard immediately. /v1 stays locked until the email is verified.

Setup

1. Backends — Connect a plan (OAuth / local connector) or paste an API key. One account can hold several Claude, Grok, or GLM rows and fail over between them.
2. Endpoints — Match models (exact, glob, or *) and attach backends with an optional model map.
3. Keys — Mint a key. Plaintext is shown once. Optional per-key token cap.

Point an agent at it

Same host for every customer. Auth header selects the account.

ANTHROPIC_BASE_URL=https://ai-gateway.eversources.app
ANTHROPIC_API_KEY=sk-rtr-…

OpenAI-speaking clients:

OPENAI_BASE_URL=https://ai-gateway.eversources.app/v1
OPENAI_API_KEY=sk-rtr-…

Protocols

Clients send POST /v1/messages (Anthropic) or POST /v1/chat/completions (OpenAI), including streaming and tools. Per-endpoint: /e/{slug}/v1/….

Sign in · Pricing