How Zrelay works

From your client to inference

1. Your applicationSend a supported model ID to api.zrelay.net using a Zrelay-issued key and the compatible request format.
2. The Zrelay gatewayZrelay authenticates the key, checks its model family and spending allowance, and reserves balance for the request.
3. An independent inference supplierThe gateway supplies upstream credentials internally and passes the request to an available route for that model family. Upstream credentials are never distributed to customers.
4. Your response and usageZrelay relays the response, meters reported token usage and settles the cost against your account balance.

Zrelay forwards the requested model identifier; it does not deliberately rewrite it to a cheaper model. Actual model execution and availability depend on the upstream supplier. This is not a guarantee of independently verified model provenance.

Supported formats include Claude-compatible Messages, GPT-compatible Responses and Chat Completions, and Grok-compatible Chat Completions. See endpoints and authentication.

Your account, credentials and controls

When a BTC payment settles, Zrelay automatically credits the purchase. On the purchase receipt page, reveal and save your account login code and initial API key. The account code is a secret: it unlocks your portal. Your account ID is a separate identifier you can use when contacting support.

Paid keys start with zr_live_ and are specific to a model family. Create additional keys, set per-key spending quotas, and revoke or replace keys from the account portal. Your keys share one prepaid balance. Trials use separate zr_trial_ credentials and the currently available trial model set.

Your prepaid Zrelay balance is an internal service balance used only for API usage. It is not cash, is not transferable between customers, and cannot be withdrawn or redeemed for money. Refunds of eligible purchases follow the Refund Policy and applicable consumer law.

Why it costs less

Zrelay can offer competitive prices by combining aggregated demand, shared infrastructure, prepaid capacity and automated metering. Lower distribution and operating costs can also help compared with building a full platform for each use case.

We publish our own service rates, not copies of upstream suppliers’ price lists. Availability and execution depend on available capacity and the technical route used. Zrelay does not claim an official model-provider relationship or an identical backend for every request.

For example, the Base Starter package costs $19 and adds $75 of Base Zrelay service credit to your account. This compares the package price with Zrelay service credit. It does not measure savings against a model provider’s direct prices. Requests consume service credit at the published Zrelay per-model usage rates.

What your usage report means

Active requests temporarily reserve Zrelay service credit. When usage settles, the report shows the cost and releases the unused reservation. Input can include system context and cached content reported by the upstream, so billed token counts can differ from a count of your visible prompt.

The usage checker shows settled spend, pending reservations, spendable balance and per-model usage. Spending quotas apply per key in addition to the shared account balance.

Request data and service dependencies

Requests and responses pass through Zrelay and the upstream inference supplier. Ordinary prompt and output bodies are not kept as conversation history in Zrelay’s application database; bounded upstream error diagnostics may be retained. Infrastructure and upstream services can also process connection and request data. Read the Privacy Policy and processing provider disclosures.

Models, capacity and routes can become unavailable. Zrelay does not offer a documented uptime SLA. Check the service-status button in the navigation and contact support for a specific problem.