Ensemble

Ensemble is a powerful tool that leverages multiple Large Language Models (LLMs) to provide a multi-perspective answer for any given prompt. By combining diverse responses and refining them, Ensemble delivers a clear, concise, and balanced final answer.

Ensemble's architecture involves fetching responses from multiple LLMs using the OpenRouter API, combining these responses, and then refining the combined response using a dedicated refinement model. The core logic is implemented in src/ensemble.py, while the web interface is located in the src/web directory.

Features

• Multi-LLM Integration: Interact with various language models concurrently to gain comprehensive insights.
• Response Refinement: Merge and refine multiple responses into a single, optimal answer.
• OpenRouter API Integration: Utilize the OpenRouter service to access multiple LLMs through one interface.
• Easy Configuration: Customize models, prompts, and API keys effortlessly via environment variables, with fallback to default values.

Setup

1. Install Dependencies

   • Run pip install -r requirements.txt to install all required packages.

2. Configure Environment Variables

   • Ensemble includes a default.env file with sensible defaults, but you should:
   • Create your own .env file in the project root to override any defaults:
     • OPENROUTER_API_KEY: Your OpenRouter API key.
     • MODELS: Comma-separated list of LLM model identifiers.
     • PROMPT: The prompt to send (optional; if not provided, the script will ask for input).
     • REFINEMENT_MODEL_NAME: The model to use for response refinement.

Configuration

Ensemble uses a two-tier configuration system:

1. Default Configuration: The default.env file in the project root contains sensible defaults for all required settings.

2. User Configuration: Your custom .env file overrides any defaults. You only need to specify the values you want to change.

This ensures Ensemble can run with minimal setup while still allowing for complete customization.

Usage

Execute the main script to generate and refine responses:

python src/ensemble.py

Web Interface

Ensemble provides a web interface for interacting with the system. The web interface is built using Flask and is located in the src/web directory.

To run the web application:

python src/web/app.py

License

Ensemble is licensed under the MIT License. See the LICENSE file for more information.