AutoReason

An open-source language-model agent framework for iterative self-refinement: independent critic/judge agents compare candidate responses, including retaining the original when revisions are not better.