AlphaGo Zero (two-headed net + PUCT MCTS + self-play RL) in PyTorch, validated on tic-tac-toe. Game-agnostic pipeline.

Terminal multiplexer with a first-class agent API for shared human+agent workflows

Small, high-leverage reproductions for JEPA-style representation learning and recent Yann LeCun world-model work.

Agent orchestration for amux

Clean-room reproduction of Language Agent Tree Search (LATS, arXiv:2310.04406). First domain: HumanEval.

Clean-room reimplementation of the Huxley-Gödel Machine (arXiv:2510.21614): a self-improving coding agent with CMP + Thompson-Sampling tree search, validated in $0 simulation and live on SWE-bench (mini-swe-agent + gpt-5.4).

PTY driver for AI coding agents. Spawns agent TUIs with programmatic control via Unix socket, while keeping the full visual TUI visible.

Homebrew tap for weill-labs tools