LatentMAS

About

LatentMAS is a multi-agent reasoning framework that moves agent collaboration from token space into the model’s latent space. Instead of producing long textual reasoning traces, agents communicate by passing latent thoughts through their own working memory. LatentMAS has the following key features:

Overall, LatentMAS achieves superior performance, lower token usage, and major wall-clock speedups of the multi-agent system.

Explore community-driven extensions that expand LatentMAS into new domains, architectures, and collaboration patterns:

By Prof. Markus J. Buehler & MIT LAMM Group

New Features: Extends LatentMAS for scientific modeling and material-system collaboration, enabling…