Andriy Karpatі created an interesting video blog dedicated to the development of the LLM Council. It is a fairly simple web application that can be run locally and features an interface similar to ChatGPT. However, its main difference is that requests are processed through Openrouter and directed to multiple language models simultaneously. Responses from all models are initially collected and mutually evaluated, then ranked based on accuracy and depth of analysis. Finally, the “main” model consolidates all the received data and generates the final answer. An OpenRouter API key is required to operate.
The concept is that instead of asking a question to a single model, you can combine several models to create what is called a “Model Council.” The process works as follows:
First, opinions are gathered. The query is sent separately to all models, and their responses are saved as tabs — so they can be reviewed and manually analyzed.
Next, the review stage begins: each model receives responses from the other participants. The identifiers remain anonymous to prevent favoritism during the evaluation of others’ results. Responses are ranked based on accuracy and depth of analysis.
Finally, the main model considers all received responses and forms a single overall summary — the final answer to the question.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby