Monty – a sensorimotor learning system following the principles of the neocortex (github.com)

The Thousand Brains Project has open‑sourced “Monty,” an early‑beta sensorimotor learning system that implements the neocortical principles of cortical columns proposed by Vernon Mountcastle. Originally started at Numenta and now developed by the independent Thousand Brains non‑profit (partly Gates Foundation‑funded), Monty is released under an MIT license with full documentation, reproducible benchmark configs, and recorded meetings. The codebase is functional but actively evolving; each functional change triggers a suite of sensorimotor benchmarks (configs in conf/experiment/) whose results are published alongside performance summaries and a public roadmap. Monty matters because it operationalizes a biologically inspired alternative to mainstream deep learning: a modular, column‑based sensorimotor architecture aimed at rapid, robust learning and inference for embodied tasks. Key technical points include explicit sensorimotor integration, reproducible evaluation on benchmark tasks, and open contributor workflows (CLA required). The project links to theory and systems papers that argue advantages over deep nets for generalization and active perception, positioning Monty as a platform for researchers and roboticists exploring cortical principles in practice. Expect fast iteration, experimental APIs, and an emphasis on community collaboration rather than production readiness.