An API gateway for AI models — routes your requests to the cheapest capable model across multiple providers. Supports OpenAI-compatible and Anthropic-native endpoints.
Instead of specifying a model, you specify a capability tier
(economy, standard,
or premium) and an optimisation preference
(cheap, fast,
balanced, or quality).
The router selects the best model from all available providers.
If a provider goes down, the circuit breaker reroutes automatically. If a cheaper model launches, you benefit without changing code. Context-window guards ensure your input always fits the selected model.
1. Sign up — go to /profile, enter your email, get a login code. New accounts get $1 in trial credits.
2. Create an API key — on the profile page, generate a key
(starts with mr_sk_).
3. Make a request — two endpoints available:
OpenAI-compatible — works with any OpenAI SDK or tool:
economy routes to the cheapest models in each tier. Your $1 trial credit covers
millions of tokens on economy, or fewer calls on
standard / premium.
Pin a specific model like claude-sonnet-4-6.
See Tiers.
Anthropic-native — use the native Anthropic Messages API for full-fidelity Claude access (including thinking blocks and interleaved reasoning):
That’s it. For the OpenAI-compatible endpoint, the response format is identical to OpenAI’s — any existing client library or tool that speaks the OpenAI API will work without modification. For the Anthropic-native endpoint, use the standard Anthropic Messages API format.
Provider costs are passed through at exact rates — no per-request markup. We charge a 4% fee on credit top-ups (minimum $0.80) via Stripe. That’s it.
All prices in USD. View your balance and usage at /profile.