Square Harness
An open-source tool for mathematical research with language models.
What is a harness?
A language model takes a prompt and produces an answer. A harness is the software that organises a longer task around the model: selecting context, calling tools, and keeping track of successive results. For example, it can request a proof attempt, send it for review in a separate model context, and return the objections for revision.
Why an open harness?
Researchers and institutions should be able to run and control their own AI systems. This needs models, computing infrastructure, and a harness.
As open-weight models improve, we want to make them useful in a mathematician’s working environment. Square Harness is our contribution to this effort. We develop it openly so that its workflows can be inspected, adapted, and run on infrastructure you control.