Consilium adds multi-LLM debate with MCP server and Gradio UI

AI Impact Summary

Consilium introduces a structured, multi-LLM collaboration workflow that pits several models in roles to discuss complex questions and reach consensus through rounds and configurable topologies (Ring/Star). It runs as both a Gradio UI component and an MCP server, enabling integration with downstream apps via the Model Context Protocol, and includes a dedicated research agent that uses function calls to fetch data from Web Search, Wikipedia, arXiv, GitHub, and SEC EDGAR for evidence. The system supports multiple base models (Mistral Large, DeepSeek-R1, Meta-Llama-3.3-70B, QwQ-32B) and a lead-analyst synthesis step, which can improve decision quality but increases compute cost and latency, requiring governance around role definitions and model availability. The example references related work like MAI-DxO and Claude Desktop to illustrate the potential impact of orchestrated, multi-model panels on diagnostic or advisory tasks.