Llms

LLMs

A universal LLM API transformation server, initially developed for the [claude-code-router](https://github.com/musistudio/claude-code-router).

How it works

The LLM API transformation server acts as a middleware to standardize requests and responses between different LLM providers (Anthropic, Gemini, Deepseek, etc.). It uses a modular transformer system to handle provider-specific API formats.

Key Components

1. **Transformers**: Each provider (e.g., Anthropic, Gemini) has a dedicated transformer class that implements:

   • transformRequestIn: Converts the provider's request format to a unified format.
   • transformResponseIn: Converts the provider's response format to a unified format.
   • transformRequestOut: Converts the unified request format to the provider's format.
   • transformResponseOut: Converts the unified response format back to the provider's format.
   • endPoint: Specifies the API endpoint for the provider (e.g., "/v1/messages" for Anthropic).

2. **Unified Formats**:

   • Requests and responses are standardized using UnifiedChatRequest and UnifiedChatResponse types.

3. **Streaming Support**:

   • Handles real-time streaming responses for providers like Anthropic, converting chunked data into a standardized format.

Data Flow

1. **Request**:

   • Incoming provider-specific requests are transformed into the unified format.
   • The unified request is processed by the server.

2. **Response**:

   • The server's unified response is transformed back into the provider's format.
   • Streaming responses are handled with chunked data conversion.

Example Transformers

• Anthropic: Converts between OpenAI-style and Anthropic-style message formats.
• Gemini: Adjusts tool definitions and parameter formats for Gemini compatibility.
• Deepseek: Enforces token limits and handles reasoning content in streams.

Run this repo

• Install dependencies:

  npm install
  # or pnpm install