Autonomous Relay Agents is one OpenAI-compatible endpoint in front of every model provider you already pay for. You send a request naming a model; we apply the rules you have set, pick a provider that satisfies them, dispatch on your credential, stream the response back, and record exactly what it cost.
We do not resell inference. Every model with a price runs on your own provider key at the provider’s own price, and we add nothing to the cost of a token. That is enforced by a database constraint, not a pricing promise. The one exception costs nothing: a deployment may also serve the free models on its own key until you add yours, at $0 and within a daily limit.
GET /v1/free-tier says whether this one does.What that means for you
- One integration. Change a base URL and a key; your existing OpenAI client works unchanged.
- Your keys, your relationship. You contract directly with each provider. Their terms apply to you, and their rate limits are yours.
- Failover without failover code. If a provider errors before any byte has shipped, the next candidate takes the request.
- An auditable ledger. Every debit reconstructs from token counts the provider itself reported.
Base URL
https://autonomousrelay.com/v1
api.autonomousrelay.com serves the same API on the same certificate, if you prefer to keep API traffic on its own hostname.
Where to go next
- Quickstart — a working request in about two minutes.
- Authentication — there are two kinds of key and they are not interchangeable.
- Bring your own key — required for every model except the free ones.
- OpenAPI 3.1, for generating a client or handing to a coding agent: openapi.json or openapi.yaml.