LLM Relay API

This is a OpenAI API relay server, which can be used to relay requests to multiple providers.

The project supports serverless deployment on Vercel.

Installation

bun install

Usage of API

bun run dev
bun run deploy # Deploy to Vercel

CLI Tool (helper.ts)

This helper tool helps manage the api.json configuration file, allowing you to add user API keys, add providers, add keys to existing providers, add models to existing providers, and modify provider settings.

Usage

To run the helper tool, use the following command:

bun run helper

This will start an interactive menu in your terminal, guiding you through the available options:

1. Add User API Key: Generates and adds a new user API key to the configuration.
2. Add Provider: Adds a new provider to the configuration, prompting for the provider's name, base URL, models, and keys.
3. Add Key to Existing Provider: Adds a new API key to an existing provider.
4. Add Model to Existing Provider: Adds a new model to an existing provider.
5. Modify Provider Settings: Modifies the settings of an existing provider, such as the name, base URL, and Azure-specific settings.

File api.json Format

Make sure all provider supports OpenAI API format.

The file is located in ./data/api.json, make sure to create it if it doesn't exist, and keep it secure.

{
    "userKeys": [
        {
            "name": "Test",
            "key": "sk-123456" // User key, generated locally
        }
    ],
    "providers": {
        // Azure
        "azure": {
            "name": "Azure OpenAI",
            "azure": true, // Required for Azure providers
            "azureAPIVersion": "2024-10-21", // Azure API version
            "baseURL": "https://azure-openai.azure-api.net/v1",
            "models": ["gpt-4o", "gpt-4o-mini"],
            "keys": ["123456"]
        },
        // Google provider
        "google-genai": {
            "name": "Google GenAI",
            "baseURL": "https://generativelanguage.googleapis.com/v1beta/openai",
            "models": [
                // 2 Types: Object (Need casting), String (Direct)
                {
                    "request": "gpt-4o-mini", // Model name from request to be casted
                    "destination": "openai/gpt-4o-mini" // Model name sent to provider
                },
                "gemini-1.5-flash-latest", // Direct model name, without casting
            ],
            "keys": [
                // User provided
                "AIzaSyB1234567890" // Provider key, which sends requests to the provider
            ]
        }
    }
}