The Smart Router for LLMs.

First steps

Make your first API call

The fastest path is to use SKYLINE INTELLIGENCE as the base URL for an existing OpenAI-compatible client. The public site is https://skylineintelligence.top/; customer API calls should use https://api.skylineintelligence.top/v1.

1. 1
   Create an API key

   Open API Keys, create a key, and keep the raw key from the one-time creation screen. Default customer scopes include model:invoke, chat:read, and chat:write.

2. 2
   Copy the Base URL

   Use the console connection panel or call GET /api/connection-info with your console session token. The returned baseUrl should be https://api.skylineintelligence.top/v1.

3. 3
   Send one chat request

   Use model: "auto" for router-owned selection, or choose a specific model from /v1/models or the Model Plaza.

Connect

Use existing OpenAI clients

Most apps only need two changes: set baseURL to SKYLINE INTELLIGENCE and replace the API key. The request body stays familiar.

export SKYLINE_BASE_URL="https://api.skylineintelligence.top/v1"
export SKYLINE_API_KEY="sk_aibrg_your_key"

curl "$SKYLINE_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $SKYLINE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "messages": [
      { "role": "user", "content": "Explain SKYLINE INTELLIGENCE in one sentence." }
    ]
  }'

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.SKYLINE_API_KEY,
  baseURL: process.env.SKYLINE_BASE_URL
});

const response = await client.chat.completions.create({
  model: "auto",
  messages: [
    { role: "user", content: "Write a launch checklist." }
  ],
  stream: false
});

process.stdout.write(response.choices[0].message.content);

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["SKYLINE_API_KEY"],
    base_url=os.environ["SKYLINE_BASE_URL"],
)

response = client.chat.completions.create(
    model="auto",
    messages=[
        {"role": "user", "content": "Summarize this product for a CTO."}
    ],
)

print(response.choices[0].message.content)

API reference

Runtime API surface

Customer traffic goes through the runtime service under /v1. Start with chat completions unless you specifically need Responses, Anthropic-style Messages, or file operations.

Models and routing

Choose a model, or let the router choose

SKYLINE INTELLIGENCE separates customer-facing model codes from upstream provider details. Your organization only sees models enabled by an admin. The router then picks an eligible channel based on model access, capability requirements, priority, weight, health, and fallback availability.

{
  "model": "gpt-4o-mini",
  "messages": [{ "role": "user", "content": "Hello" }]
}

{
  "model": "auto",
  "stream": true,
  "messages": [{ "role": "user", "content": "Draft release notes" }]
}

Billing and usage

Every request is authorized before it runs

SKYLINE INTELLIGENCE reserves estimated quota and billing capacity before the provider call, then settles against actual usage after completion. That keeps quota enforcement predictable during concurrent traffic and gives finance teams a request-level ledger.

Console workflow

Operate the gateway from the customer UI

The docs map directly to the current console navigation so operators and developers can speak the same language.

Errors and fallback

Handle failures explicitly

Runtime errors use a stable JSON shape with a request ID. Rate limits and temporary service pressure include Retry-After. Fallback responses expose degradation metadata so your app can decide whether to display, retry, or audit the result.

{
  "type": "error",
  "error": {
    "type": "rate_limit_error",
    "message": "API key rate limit exceeded"
  },
  "request_id": "req_..."
}

X-SKYLINE-Degraded: true
X-SKYLINE-Fallback-Reason: channel_unhealthy
X-SKYLINE-Fallback-From: claude-sonnet
X-SKYLINE-Fallback-To: gpt-4o-mini

Before live traffic

Go-live preflight

Use this as a final client-integration check before an app sends real traffic through SKYLINE INTELLIGENCE. It is not a platform production runbook.