openai-gemini-api-key-rotator

Node.js proxy server for automatic API key rotation across multiple LLM providers (OpenAI, Gemini, Groq, OpenRouter, etc.). Includes a built-in Telegram bot for chatting with any model. Zero external dependencies.

Features

• Automatic Key Rotation: Rotates keys on configurable status codes (default: 429)
• Universal API Compatibility: Works with any OpenAI or Gemini-compatible API
• Streaming Support: Full pass-through for SSE/streaming responses ("stream": true)
• Smart Key Shuffling: Avoids recently failed keys using intelligent rotation
• Key Management: Reorder keys, disable/enable individual keys or entire providers
• Key Usage Tracking: See how many times each key has been used (in-memory)
• Live Key Validation: API keys automatically tested before saving
• Hot Configuration: Add, edit, rename, or delete providers without restart
• Custom Status Codes: Configure which HTTP codes trigger rotation per request
• Optional Access Control: Secure providers with access keys requiring authorization
• Default Models: Pre-save models for easy curl command generation
• Modern Admin Panel: Dark/light theme support for comfortable management
• Request Monitoring: Last 100 requests logged with key usage details (which key succeeded/failed)
• File Logging: All API requests logged to logs.jsonl with debounced writes for performance
• Telegram Bot: Chat with any configured model directly from Telegram (text, images, image generation)

Quick Start

git clone https://github.com/p32929/openai-gemini-api-key-rotator.git
cd openai-gemini-api-key-rotator
cp .env.example .env
# Edit .env: Set PORT and ADMIN_PASSWORD
npm start

Access admin panel: http://localhost:8990/admin

Configuration

PORT=8990
ADMIN_PASSWORD=your-secure-password

Visit http://localhost:8990/admin to configure your providers and start using the API.

Telegram Bot

Chat with any of your configured models directly from Telegram. Set it up from the admin panel (Settings icon) or add these to your .env:

TELEGRAM_BOT_TOKEN=your-bot-token-from-botfather
TELEGRAM_ALLOWED_USERS=123456789,987654321

Leave TELEGRAM_ALLOWED_USERS empty to allow anyone.

Setup

1. Open Telegram and message @BotFather
2. Send /newbot and follow the prompts to create your bot
3. Copy the bot token
4. Go to the admin panel → click the Settings icon (next to theme toggle) → paste the token and save
5. To find your User ID, message @userinfobot on Telegram

Bot Commands

Bot Features

• Model Selection: Browse providers → fetch available models → select, or type a custom model name
• Conversation History: Maintains up to 50 messages per user
• Image Input: Send photos to vision-capable models (auto-converts to base64)
• Image Generation: Supports image gen models — renders base64 and URL responses as Telegram photos
• All requests go through the proxy, so you get key rotation, access key validation, and logging automatically

API Usage Examples

OpenAI-Compatible APIs

curl -X POST "http://localhost:8990/groq/chat/completions" \
  -H "Authorization: Bearer [STATUS_CODES:429][ACCESS_KEY:your-access-key]" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-oss-120b",
    "messages": [
      {
        "role": "user",
        "content": "Hello! Please say hello back."
      }
    ]
  }'

Gemini-Compatible APIs

curl -X POST "http://localhost:8990/gemini/models/gemini-2.5-flash:generateContent" \
  -H "x-goog-api-key: [STATUS_CODES:429][ACCESS_KEY:your-access-key]" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "parts": [
          {
            "text": "Hello! Please say hello back."
          }
        ]
      }
    ]
  }'

Note: Replace your-access-key with your provider's ACCESS_KEY if configured. If no ACCESS_KEY is set for the provider, you can omit the [ACCESS_KEY:...] parameter entirely.

File Logging

All API requests are automatically logged to logs.jsonl in the project root. Writes are debounced (5 seconds) for performance — multiple requests within the window are batched into a single disk write. Each line is a JSON object:

{"timestamp":"2026-04-01T15:33:15.221Z","requestId":"h8j4pqqot","method":"POST","endpoint":"/chat/completions","provider":"cerebras","status":200,"responseTime":497,"error":null,"clientIp":"::ffff:127.0.0.1","keyUsed":"csk-...9ft2","failedKeys":[]}

Changelog

Version 6.x.x

• Telegram Bot — chat with any model from Telegram with interactive model selection, conversation history, image input/output support
• File Logging — all API requests logged to logs.jsonl with debounced writes
• Settings Panel — gear icon in admin panel for configuring Telegram bot
• Admin panel improvements: password input autofocus, logs sorted newest-first, disabled providers shown at bottom

Version 5.x.x

• Streaming support — SSE responses piped through without buffering
• Disable/enable individual API keys (persisted via ~ prefix in .env)
• Disable/enable entire providers (_DISABLED=true in .env)
• Reorder API keys to control rotation priority
• Per-key usage tracking displayed in admin panel
• Logs now show which key was used and which keys failed per request

Version 4.x.x

• Dynamic status code configuration via headers
• Optional ACCESS_KEY for provider-level security
• Enhanced admin panel with improved UX
• Auto-generated curl commands reflect the new API format

Breaking Changes:

• API endpoints changed from /provider/v1/* to /provider/*
• Version suffix (/v1) now derived from provider's base URL configuration
• Migration: Simply copy the curl command from admin panel to see the new format in action

Version 3.x.x

• Enhanced admin panel with better UI/UX
• No breaking changes

Version 2.x.x

• Added admin panel for dynamic provider management
• No breaking changes

Version 1.x.x

• Basic API key rotation
• OpenAI and Gemini-compatible API support

Screenshot

Contributing

Contributions are warmly welcomed and greatly appreciated! Whether it's a bug fix, new feature, or improvement, your input helps make this project better for everyone.

Before submitting a pull request, please:

1. Create an issue describing the feature or bug fix you'd like to work on
2. Wait for discussion and approval to ensure alignment with project goals
3. Fork the repository and create your feature branch
4. Submit your pull request with a clear description of changes

This approach helps avoid duplicate efforts and ensures smooth collaboration. Thank you for considering contributing!

License

This project is licensed under the MIT License - see the LICENSE file for details.