A local-first command-line and terminal-interface coding agent that runs open-weight models entirely on the user's machine through a llama.cpp fork, supporting private offline software development without cloud inference.