This project provides a lightweight, proof-of-concept PHP web application for demonstrating basic chain-of-thought (CoT) reasoning with any Large Language Model (LLM) supporting an OpenAI-compatible API. It's designed for developers and researchers looking to quickly experiment with CoT capabilities across various local or remote LLM providers without complex setup.

How It Works

ReflectionAnyLLM uses a PHP backend (chat.php) to interact with LLMs via their OpenAI-compatible API endpoints. It dynamically generates reasoning steps, allowing the LLM to break down problems. A summary of the thought process is displayed by default, with an option to view the full details. The application maintains a history of the last 30 messages for context.

Quick Start & Requirements

• Install: Clone the repository and place index.html and chat.php into your web server's public directory (e.g., XAMPP's htdocs/).
• Prerequisites: A web server (XAMPP, MAMP, LAMP) with PHP and the curl extension enabled.
• Setup: Configure API endpoint and key in chat.php. Access via http://localhost/ or your domain.
• Demo: freemindgpt.com/reflectionanyllm (uses Gemini-pro-1.5 on OpenRouter free tier).

Highlighted Details

• Supports any OpenAI-compatible LLM API (LM Studio, Groq, OpenRouter, OpenAI).
• Implements dynamic, LLM-generated reasoning steps (up to 10, configurable).
• Solves simple reasoning tasks like the "strawberry count" problem on models >8B parameters.
• Allows downloading chat history as JSON.

Maintenance & Community

The project is a prototype, open for community contributions via pull requests. Feedback and suggestions are welcomed.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use and closed-source linking, requiring attribution.

Limitations & Caveats

This project is explicitly labeled a prototype and is not secured for production environments. The backend code lacks robust security practices. While functional for basic demonstrations, performance and advanced reasoning depend heavily on the chosen LLM.