> ## Documentation Index
> Fetch the complete documentation index at: https://mifr.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> Gateway routes, exclusions, and alpha limits.

The gateway origin is `https://mifr-gateway-production.up.railway.app`.

Inference routes require `Authorization: Bearer sk-mifr_…`. See [Authentication](/authentication).

| Method    | Path                   | Auth   | Purpose                                       |
| --------- | ---------------------- | ------ | --------------------------------------------- |
| `POST`    | `/v1/chat/completions` | Bearer | Chat Completions, stream and non-stream       |
| `GET`     | `/v1/models`           | Bearer | Aliases currently servable at the trust floor |
| `GET`     | `/v1/providers`        | Bearer | Live Macs at the floor                        |
| `GET`     | `/health`              | none   | Liveness; body `mifr-gateway`                 |
| `OPTIONS` | any                    | none   | CORS preflight, 204, before the key check     |

The gateway does not serve `/v1/responses`, Anthropic, or `/v1/encryption-key`. Coordinator `GET /v1/catalog` and MDM attestation live on other hosts.

`/status.json` and `/metrics` are the Verified latency board. They are ops surfaces, not this product.

## Trust floor

The default floor is `hardware` (`MIFR_GATEWAY_MIN_TRUST`). If no provider is at the floor, the gateway returns `404 model_not_found` or `503 provider_unavailable`; it never serves a lower rung while keeping the request's `x-mifr-trust` label. An operator who sets `MIFR_GATEWAY_MIN_TRUST=process` is opting into low-trust mode, which is not Apple-backed. That environment setting is the only way down; there is no "best available" fallback.

## Alpha limits

There is one gateway host. A provider still serves one sealed session at a time, so concurrent load queues. Throughput and catalog size are not a caller contract.
