Shipped: TheOrc

A local-first AI orchestration shell: give it a goal and a warband of agents — researcher, coder, UI, tester — plans and executes in parallel, with every file write and shell command behind an approval gate. Built for developers who would rather own the machine than rent it: in-process llama.cpp inference on your own GPU, Ollama optional, no cloud. Its source-grounded memory answered 104 of 120 held-out questions at 97.1% citation precision, against a 52/120 best baseline.