Open-source LLM gateway

Your keys. Your gateway.

A zero-markup BYOK LLM router. OpenAI-compatible, with 20+ providers and an MIT-licensed core. Point any OpenAI SDK at KeysQuay. You pay providers directly.

This gateway is the hosted beta. You can also run the same MIT engine in your own Cloudflare account.

app.ts
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://keysquay.theserverless.dev/v1",
  apiKey: "kq_live_…",
});

await client.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Hello" }],
});

Works with the providers you already pay

  • OpenAI
  • Anthropic
  • Gemini
  • Groq
  • Mistral
  • DeepSeek
  • xAI
  • Together
  • Fireworks
  • OpenRouter
  • Cerebras
  • Moonshot
  • Qwen
  • Bedrock

One gateway for every app key

Provider credentials stay in your workspace. Apps only see a virtual key.

Bring your own keys

Store provider credentials in your workspace. Apps never see them. A virtual key is what you put in client code.

Zero markup

You pay the provider directly. KeysQuay does not sell credits, balances, or a percentage fee on inference.

OpenAI-compatible

Point the OpenAI SDK at your gateway base URL. Chat completions, streaming, embeddings, and the models list use the same shapes.

Asia and beyond

Connect DeepSeek, Moonshot, Zhipu, Qwen, Sarvam, OpenAI, OpenRouter, or any OpenAI-compatible base URL.

How a request moves

A virtual key selects one workspace. A route selects the provider keys that can answer.

01

Virtual key

The client sends Authorization: Bearer kq_live_…. KeysQuay hashes it and loads that workspace only.

02

Route

The model name matches a route in the same workspace. Fallback tries the next provider key in priority order.

03

Usage

Token counts and a cost estimate are recorded for that virtual key. The estimate is not an invoice.

Providers you can connect

Connect OpenAI, Anthropic, Gemini, Groq, Mistral, DeepSeek, xAI, Together, Fireworks, OpenRouter, Cerebras, Perplexity, Moonshot, Zhipu, Qwen, MiniMax, Sarvam, Azure OpenAI, Bedrock, or any OpenAI-compatible base URL.

Clients send OpenAI Chat Completions. Anthropic chat is translated to the Messages API, including streaming and tools. Bedrock uses its OpenAI-compatible runtime path. Anthropic embeddings are not available.

No markup

You pay the provider. That is the whole price.

KeysQuay does not sell credits and does not add a percentage fee on inference. Figures in the dashboard are estimates from public prices. Missing prices stay unknown. They are not invoices.

Read pricing

0%

KeysQuay fee on provider usage

Open source, and hostable

The gateway is MIT licensed. Self-host it on Cloudflare Workers, or use the hosted beta on this deployment. The engine is the same. Hosted accounts add sign-up, workspaces, and member invites.

View the source