Terms of service

minirouter.sh is an OpenAI-API-compatible gateway: one API key reaches many language models across multiple upstream providers, with automatic failover between them. We are a router and a meter. We do not train models, and the models’ outputs are produced by the upstream providers we route to.

Prepaid credit is denominated in credits, 1,000 credits = $1, pegged. Deposits are accepted in crypto (minimum $5) and by card (minimum $20, 3.5% processing fee). Deposits are one-way: there is no withdrawal path — not in the product, not in the API, not by support ticket. Credits buy inference and nothing else. This is a deliberate design choice: a withdrawal path would make us a money transmitter and force identity verification on every user. Credits never expire.

Refunds, where they occur, are issued as service credit, never as cash or coin. The full policy — including recovery of wrong-chain and wrong-token sends — is at /legal/refunds.

The API key and permanent dashboard link are bearer credentials. In accountless mode they are the only proof of control you have with us: if you lose both, the balance cannot be recovered, because nothing exists that could distinguish you from anyone else holding it. This is stated on screen before payment; it is repeated here. The dashboard link remains valid when API keys are rotated, with a grace window so a running agent is not cut off mid-task, and per-key rate and spend limits are yours to set.

The rules and their enforcement rationale are at /legal/acceptable-use. The one-sentence version: your traffic reaches providers through shared upstream accounts, so abuse that endangers those accounts endangers every other user’s service, and we suspend for it.

This service is operated by one person on a small number of machines. There is no SLA at launch, and we will not invent one. What you get instead: automatic failover between upstream providers when one degrades, a status page hosted separately so it survives our own incidents, and published measurements of our per-provider latency and availability once — and only once — they exist. Honest overnight incident recovery is measured in hours, not minutes.

Per-model unit rates and the markup schedule are published at /pricing. Price changes take effect as new dated versions, never retroactively; a request is always billed against the price that was live when it ran. Every response carries its own cost in the body and headers so our metering can be audited per request. Charges follow provider billing: if a provider bills us for tokens generated after you disconnected, the charge stands and is labelled as estimated; a stream that dies before 50 output tokens is written off entirely.

We list only models for which the configured upstream terms permit platform-funded routing. Models with unverified or restricted resale terms are unavailable. We do not currently accept customer provider credentials, and a catalog entry is not a representation that every upstream will remain available indefinitely.

The service is provided as-is. Our aggregate liability to you is capped at the balance you have on deposit with us — which is also the most you can lose, and one reason the minimum deposit is five dollars. We are not liable for the content models generate, for upstream provider outages we failed to route around, or for losses caused by leaked or lost bearer credentials. Nothing here excludes liability that cannot lawfully be excluded.

We may amend these terms; material changes are dated in the changelog and continued use after a change is acceptance of it. Questions: legal@minirouter.sh.