9router

9router is a high-performance, self-hosted LLM API Gateway and Router. It allows you to manage API keys, enforce quotas, apply rate limits, and route requests to various upstream AI providers (Qwen, DeepSeek, MiniMax, GLM, etc.) through a single, unified OpenAI-compatible endpoint.

Built for reliability, scalability, and ease of use, 9router provides a modern admin dashboard to manage buyers, monitor live usage, and control model access.

---

✨ Features

• Unified API Gateway: Single OpenAI-compatible endpoint (/v1/chat/completions) for all upstream models.
• Public Dashboard: Read-only public dashboard (Leaderboard, Live Usage, Usage Charts, Models) served at / with gzip compression enabled.
• API Key Management: Create, manage, extend, and revoke API keys for individual buyers.
• Quota Management:
  • Total token quotas per key.
  • Daily Reset: Automatic token quota reset at 00:00 WIB for specific packages.
• Advanced Rate Limiting:
  • Per-key RPM (Requests Per Minute) rate limiting.
  • Bulk update capabilities to apply rate/request limits to all active keys simultaneously.
• Usage Tracking: Detailed logging of prompt/completion tokens, latency, cost estimates, and success/failure rates.
• Admin Session Tracking: Monitor active admin sessions with device, IP, and browser information, with the ability to revoke specific sessions remotely.
• Model Catalog: Dynamic model management with input/output cost configuration per million tokens.
• Modern Dashboard: Responsive, dark-mode-ready React admin panel with real-time live usage monitoring and usage charts.
• Timezone Consistency: All chart and time-display queries use Asia/Jakarta (WIB) for accurate local time representation.

---

🌐 Deployment Info

• Admin Panel + API: 43.128.67.88 (or routers9.my.id when configured)
• Public Dashboard: routers9.web.id (read-only access to Leaderboard, Live, Usage, and Models via /public/* endpoints)
• Reverse Proxy: Caddy (handles TLS/HTTPS, gzip compression, and routing to localhost:3000)

---

🛠️ Tech Stack

• Backend: Node.js, TypeScript, Fastify
• Frontend: React 18, Vite, TailwindCSS, Radix UI, Recharts
• Database: PostgreSQL 16 (with pgcrypto extension)
• Reverse Proxy: Caddy (recommended for TLS/HTTPS)
• Authentication: JWT (JSON Web Tokens) + bcrypt

---

📦 Prerequisites

• Node.js 20.x or higher
• PostgreSQL 16
• npm or pnpm

---

🚀 Installation & Setup

1. Clone the Repository

git clone https://github.com/kevinnft/9router.git
cd 9router

2. Install Dependencies

npm install

3. Environment Configuration

Create a .env file in the root directory:

# Server Configuration
PORT=3000
HOST=0.0.0.0
NODE_ENV=production

# Database Configuration
DATABASE_URL=postgresql://postgres:yourpassword@localhost:5432/9router

# Security
JWT_SECRET=your-super-secret-jwt-key-change-this
COOKIE_SECURE=false # Set to true if using HTTPS

# Rate Limiting Defaults
BUYER_CHAT_RPM=60
HIDDEN_KEY_CHAT_RPM=100

# Upstream Configuration (Example)
UPSTREAM_BASE_URL=https://api.upstream-provider.com/v1
UPSTREAM_API_KEY=your-upstream-api-key

4. Database Setup

Create the database and apply the schema:

# Create database (via psql)
createdb 9router

# Apply schema (enables pgcrypto and creates tables)
psql -d 9router -f schema.sql

5. Build and Start

# Build frontend and backend
npm run build

# Start the production server
npm start

(For development, use npm run dev)

---

📁 Project Structure

9router/
├── src/
│   ├── client/       # React frontend (Vite + Tailwind)
│   ├── server/       # Fastify backend (routes, db, security)
│   └── shared/       # Shared types and constants
├── schema.sql        # PostgreSQL database schema
├── package.json      # Dependencies and scripts
├── tsconfig.json     # TypeScript configuration
└── vite.config.ts    # Vite build configuration

---

🔌 API Endpoints Overview

Public / Proxy

• GET / - Public dashboard HTML (served at routers9.web.id).
• GET /health - Health check endpoint.
• GET /public/overview - Public overview stats (buyers, active keys, requests, tokens).
• GET /public/leaderboard - Top 100 API keys by request count (masked emails).
• GET /public/live - Live active sessions (last 5 minutes, masked emails).
• GET /public/usage - Recent 200 usage events (masked key prefixes).
• GET /public/usage/chart?range=7d - Usage chart data (ranges: 15m, 1h, 6h, 12h, 1d, 3d, 7d, 1M).
• GET /public/models - List available white-label models.
• POST /v1/chat/completions - OpenAI-compatible chat completion endpoint (requires Bearer API key).

Admin (Requires Authentication)

• POST /admin/auth/login - Admin authentication.
• GET /admin/overview - Dashboard metrics.
• GET /admin/keys - List all active API keys.
• POST /admin/create-key - Create a new buyer and API key.
• PATCH /admin/keys/:keyId/rate-limit - Update RPM limit for a specific key.
• PATCH /admin/keys/bulk-rate-limit - Apply RPM limit to all active keys.
• PATCH /admin/keys/bulk-request-limit - Apply request limit to all active keys.
• GET /admin/sessions - List active admin sessions.
• DELETE /admin/sessions/:id - Revoke a specific admin session.

---

🛡️ Security & Best Practices

1. Never commit .env: The .env file is ignored by git. Keep your JWT_SECRET and UPSTREAM_API_KEY secure.
2. Use HTTPS: Always set COOKIE_SECURE=true and run behind a reverse proxy like Caddy or Nginx with valid SSL certificates.
3. Session Management: Regularly review the "Active Sessions" in the admin settings and revoke any unrecognized devices.
4. Database Backups: Schedule regular PostgreSQL dumps (pg_dump) to prevent data loss.

---

📋 Changelog

v1.2.0 (2026-06-11)

• feat: Public dashboard at routers9.web.id with read-only Leaderboard, Live Usage, Usage Charts, and Models pages
• fix: Sync admin and public total request counts — both now exclude status_code = 400 (bad requests)
• fix: Admin chart queries now use AT TIME ZONE 'Asia/Jakarta' — time labels are consistent with WIB
• feat: Caddy gzip compression enabled on both public dashboard and admin panel
• feat: Inline data labels on public usage chart (numbers displayed directly on chart lines)

v1.1.0 (2026-06-09)

• feat: Multi-upstream support with weighted round-robin and failover (upstreams table, per-upstream sync)
• feat: Leaderboard page — top API keys ranked by request count and token usage, with sortable columns and summary cards
• fix: Model sync now auto-removes models that no longer exist in upstream (prevents stale/ghost models in admin panel)
• fix: Removed MiniMax-M2.1 from ALWAYS_AVAILABLE_MODELS (no longer available upstream)
• feat: Daily token quota reset at 00:00 WIB for configured packages
• feat: Admin session tracking with device/IP info and remote revoke
• fix: Rate limiter auto-cleanup to prevent memory leaks

---

📝 License

MIT License. See LICENSE for details.

---

Note: This project is actively maintained. For issues or feature requests, please open an issue in the repository.