chad: a coding agent for your macbook pro
A coding agent that runs entirely on an Apple Silicon Mac through MLX, using one 27-billion-parameter local model with no API key, an agent loop that owns the inference backend process directly for a stable KV cache, and a terminal UI.
chad is a coding agent that runs entirely on an Apple Silicon Mac using MLX, built around a single Qwen 3.8 27B model rather than letting a user choose among many. It downloads the roughly 14 GB model once on first run and needs no API key. The PyPI package is chad-code, and it installs and runs through uvx.
Rather than talking to a standard completions endpoint, its agent loop owns the backend inference process directly, which the project says keeps the KV cache more stable and avoids long prefills, giving around 50 tokens per second in a real session versus roughly 10 tokens per second for a stock llama.cpp setup of the same model on the same hardware; the speed comes from MLX, custom kernels written for the model, and a bundled speculative-decoding drafter.
It requires an Apple Silicon Mac with at least 24 GB of RAM, ships a terminal UI, and includes an offline smoke test of four small fix-it tasks to verify a working install.