Keyaos

Edge-native AI API gateway — cost-optimized routing across providers, multi-protocol support, built on Cloudflare Workers.

🌏 中文 · 🌐 Website · 🐺 Werewolf · 📖 Docs · 📡 API Reference

You subscribe to multiple AI services — OpenRouter, DeepSeek, Google AI Studio, xAI, and more. Each has its own API key, pricing, and quota. Keyaos unifies them behind multi-protocol API endpoints (OpenAI, Anthropic, and more), automatically routing every request to the cheapest healthy provider.

Built entirely on Cloudflare Workers + D1 + Cron Triggers. Self-hosted deployments require no servers and fit within Cloudflare's free tier.

🏗 Architecture

flowchart LR
    Client([Client])

    subgraph Keyaos ["Keyaos (Cloudflare Workers)"]
        direction TB
        Auth[Auth & Key Permissions]
        Router[Cost-Optimal Router]
        CB[Circuit Breaker]
        Intercept[SSE Stream Interceptor]
        Billing[Usage Tracking & Billing]
        Sync[Cron: Model & Price Sync]
    end

    subgraph Upstream [Upstream Providers]
        OR[OpenRouter]
        DI[DeepInfra]
        DS[DeepSeek]
        OAI[OpenAI]
        Anthr[Anthropic]
        More["+9 more"]
    end

    Client --> Auth
    Auth --> Router
    Router --> CB
    CB --> Intercept
    Intercept --> OR & DI & DS & OAI & Anthr & More
    Intercept -.-> Billing
    Sync -.-> Router

Request flow: Client sends a request to any supported endpoint. Auth validates the API key and checks permissions (model restrictions, quota, expiry, IP). The router ranks all available credentials by unit_price × multiplier and picks the cheapest healthy one. The circuit breaker skips providers with recent failures. The SSE interceptor tee's the response stream — forwarding it to the client in real time while extracting usage data for billing in the background. A Cron job syncs model availability and pricing every minute.

✨ Features

• Cost-optimized routing — every request goes to the cheapest available provider
• Automatic failover — quota exceeded or rate limited? The next cheapest option takes over
• Zero-latency streaming — SSE responses are tee'd and forwarded in real time
• Auto-synced catalog — model availability and pricing stay up to date via Cron
• Multi-protocol — OpenAI Chat & Embeddings, Anthropic Messages, Google Gemini, AWS Event Stream
• Multimodal — image generation, image/audio/video/PDF inputs via chat completions
• Reasoning effort — unified reasoning_effort normalization across providers
• Circuit breaker — automatic failure detection and provider bypass
• API key permissions — model restrictions, expiration, spending quota, IP allowlist
• Two modes — self-hosted (single user) or platform (multi-user with Clerk + Stripe)

🚀 Quick Start

☁️ One-Click Deploy

Click the Deploy to Cloudflare button above, then set one secret:

npx wrangler secret put ADMIN_TOKEN

Done — D1 database, Cron Triggers, and schema are all provisioned automatically.

You'll get Core mode (single admin, sign in with ADMIN_TOKEN). To enable Platform mode (multi-user Clerk auth), add these build-time variables in Cloudflare → Workers → your project → Settings → Build → Variables and secrets:

They must live in the Build tab (read by vite build), not the Worker runtime secrets.

🔧 Manual Setup

pnpm install
npx wrangler login
npx wrangler d1 create keyaos-db    # update database_id in wrangler.toml
npx wrangler secret put ADMIN_TOKEN
pnpm deploy                          # builds, applies migrations, deploys

💻 Local Development

cp .env.example .env.local           # fill in provider keys
cp .dev.vars.example .dev.vars       # fill in secrets (ADMIN_TOKEN, etc.)
pnpm db:setup:local
pnpm dev                             # http://localhost:5173

📡 Usage

OpenAI Chat Completions

curl https://keyaos.<you>.workers.dev/v1/chat/completions \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-4o-mini",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Anthropic Messages

curl https://keyaos.<you>.workers.dev/v1/messages \
  -H "x-api-key: YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Embeddings

curl https://keyaos.<you>.workers.dev/v1/embeddings \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/text-embedding-3-small",
    "input": "Hello world"
  }'

Works with Cursor, Continue, Cline, aider, LiteLLM, and any tool that supports custom OpenAI or Anthropic base URLs.

🔌 Supported Providers

Adding a new OpenAI-compatible provider takes a single entry in the provider registry.

⚙️ Core vs Platform

Core (self-hosted)           Platform (multi-user)
├── Credential pool          ├── Everything in Core, plus:
├── Cost-optimal routing     ├── Clerk authentication
├── Multi-protocol proxy     ├── Stripe billing & auto top-up
├── Circuit breaker          ├── Shared credential marketplace
├── Auto-sync catalog        ├── Gift cards / redemption codes
├── Embeddings endpoint      └── Admin console & analytics
├── API key permissions
└── ADMIN_TOKEN auth

Platform is strictly additive — Core runs independently and never depends on Platform.

🖥 Frontend

Keyaos ships with a full frontend built with React 19, Vite 7, and Tailwind CSS 4:

• Model catalog — browsable, searchable listing of all available models with live prices
• Provider directory — per-provider pages with model counts and credential health
• OHLC price charts — financial-grade candlestick charts tracking model price history
• Chat UI — built-in chat interface powered by AI SDK
• API reference — interactive OpenAPI 3.1 documentation via Scalar
• MDX docs — 16 pages of embedded documentation including multimodal guides
• Dark mode — full light / dark / system theme support
• i18n — English and Chinese

🛠 Tech Stack

🤝 Contributing

Contributions are welcome! Whether it's a bug fix, new provider integration, feature request, or documentation improvement — we'd love your help.

1. Fork the repository
2. Create a feature branch (git checkout -b feat/amazing-feature)
3. Commit your changes (git commit -m "feat: add amazing feature")
4. Push to the branch (git push origin feat/amazing-feature)
5. Open a Pull Request

If you're unsure where to start, check out the open issues or start a discussion. All contributions, big or small, are greatly appreciated.