README_EN.md

🔍 LLM API Reverse Engineering Detection Tool

Current Version: v0.02

中文版 | English Version

Project Links: GitHub | Website

This LLM API reverse engineering detection tool based on official parameter support. APIs that fail the detection are highly likely to be reverse-engineered.

The project is currently in its initial phase, requiring manual comparison of response results with examples.

Features

• 🔍 Multi-Provider API Detection

  • OpenAI
  • Claude
  • Gemini

• 🛠 Comprehensive Parameter Testing

  • max_tokens parameter validation
  • logprobs parameter testing
  • n parameter verification
  • stop sequence testing
  • function_call/tools parameter validation
  • response_format parameter testing
  • image input support testing

• 📊 Intuitive Results Display

  • Real-time API request information
  • Response comparison analysis
  • Official examples reference
  • Toggle between concise/complete responses

Tech Stack

• Frontend Framework: Vue 3
• UI Components: Tailwind CSS
• Build Tool: Vite

Quick Start

Install Dependencies

npm install

Run Development Server

npm run dev

Build for Production

npm run build

Usage Guide

1. Select a model provider (OpenAI/Claude/Gemini)
2. Choose parameter types to test
3. Configure API settings
   • API Base URL
   • API Key
   • Model name
4. Click "Start Detection" button
5. Review test results and comparison data

Parameter Details

OpenAI Parameters

Parameter	Explanation
max_tokens	Official API strictly follows token limits (e.g., max_tokens=10), reverse-engineered ones ignore limits.
logprobs	Official API returns logprobs information for each token, reverse-engineered ones don't support it.
n	Official API returns multiple answers (e.g., n=2), reverse-engineered ones only return one.
stop	Official API stops generation at stop words, reverse-engineered ones ignore them.
image_url	Official API properly processes image URLs and generates descriptions, reverse-engineered ones can't handle images.
function_call	Official API returns JSON-formatted function calls, reverse-engineered ones don't call functions.
response_format	Official API returns specified formats (like JSON), reverse-engineered ones only return strings.

Claude Parameters

Parameter	Explanation
max_tokens	Official API strictly follows token limits (e.g., max_tokens=10), reverse-engineered ones ignore limits.
stop	Official API stops generation at stop words, reverse-engineered ones ignore them.
function_call	Official API returns JSON-formatted function calls, reverse-engineered ones don't call functions.

Gemini Parameters

Parameter	Explanation
max_tokens	Official API strictly follows token limits (e.g., max_tokens=10), reverse-engineered ones ignore limits.
codeExecution	Official API has built-in code execution tools, reverse-engineered implementations can't execute code.
googleSearch	Official API correctly calls Google Search and returns results, reverse-engineered implementations can't use Google Search tool.
response_format	Official API returns specified formats (like JSON), reverse-engineered implementations only return strings.

TODO

1. Implement automated reverse engineering detection
2. Implement integration with Uptime Kuma for continuous monitoring

Changelog

v0.02

1. Added reasoning model specific detection parameters
   • OpenAI: reasoning_max_tokens parameter testing
   • Claude: reasoning_stop parameter testing
   • Gemini: reasoning_max_tokens parameter testing
2. Fixed error handling to display complete response body on API errors (e.g., 520 errors) instead of showing popup alerts.

API Documentation References

• OpenAI API Documentation
• Claude API Documentation
• Gemini API Documentation

License

MIT License