AI Gateway

AI Gateway provides a unified endpoint for accessing a range of large language models through a single OpenAI-compatible API, with platform-managed credentials and workspace-scoped authentication.

Overview

AI Gateway provides:

• A unified, OpenAI-compatible endpoint for multiple LLM models
• Mandatory authentication via your workspace's auth (request tokens are resolved against the configured auth namespace)
• Per-workspace isolation: each gateway is provisioned with its own platform-assigned URL
• Optional CORS allow-list for browser-based clients
• Built-in usage tracking and rate limiting (configured platform-side)

Configuration

Configure an AI Gateway using defineAIGateway():

Definition Rules:

• Multiple gateways allowed: You can define multiple AI Gateways in your config file
• Configuration location: Define in tailor.config.ts and add to the aiGateways array
• Uniqueness: Gateway names must be unique across all AI Gateways
• Name pattern: name must match ^[a-z0-9][a-z0-9-]{1,28}[a-z0-9]$ (lowercase alphanumeric and hyphens, 3-30 characters)

import { defineAIGateway, defineConfig } from "@tailor-platform/sdk";

const aiGateway = defineAIGateway("my-aigateway", {
  authNamespace: "default",
});

export default defineConfig({
  name: "my-app",
  aiGateways: [aiGateway],
});

Options

authNamespace

The auth namespace used to resolve request tokens against your workspace's auth configuration. Must match an existing auth namespace.

defineAIGateway("my-aigateway", {
  authNamespace: "default",
});

cors

Optional list of allowed origins for browser-based clients. Each entry is one of:

• * — any origin (any scheme, any host)
• http(s)://* — any host on the given scheme
• http(s)://*.example.com — any subdomain of example.com on the given scheme
• http(s)://app.example.com — an exact origin

An optional :port may be appended in all URL forms. Omitting cors (or passing []) disables cross-origin access — browsers will block any cross-origin reads.

defineAIGateway("my-aigateway", {
  authNamespace: "default",
  cors: ["https://app.example.com", "https://*.example.com"],
});

Complete Example

import {
  defineAIGateway,
  defineAuth,
  defineConfig,
  defineStaticWebSite,
} from "@tailor-platform/sdk";

const website = defineStaticWebSite("my-frontend", {
  description: "Frontend application",
});

const aiGateway = defineAIGateway("my-aigateway", {
  // Name of an auth namespace in your workspace; request tokens are resolved against it.
  authNamespace: "default",
  cors: [website.url],
});

const auth = defineAuth("my-auth", {
  // ...auth configuration...
});

export default defineConfig({
  name: "my-app",
  auth,
  staticWebsites: [website],
  aiGateways: [aiGateway],
});

Runtime Usage

Resolvers, executors, and workflow jobs can resolve a gateway's platform-assigned URL by name via aigateway.get(). The name is type-checked and autocompleted against the AI Gateways defined in aiGateways:

import { aigateway } from "@tailor-platform/sdk/runtime";

const { url } = await aigateway.get("my-aigateway");

// await aigateway.get("unknown"); // Type error — only "my-aigateway" is allowed

Type narrowing is provided by the generated tailor.d.ts (the AIGatewayNameRegistry interface). Run tailor generate (or deploy) after defining new AI Gateways to refresh it. Before the first generate run, get() accepts any string.

The same URL is also shown by tailor show, which lists the URL of each AI Gateway defined in aiGateways once it has been deployed.