Jul 20, 2026traininglow
model training
Latest stance: no training claim
“This is the part most users care about, so we’re specific: Processing is transient. Each generation runs in an isolated, single-use sandbox. Your repository is cloned into that sandbox, used to generate your CLI, and destroyed with the sandbox when the run ends. We do not keep copies of your source code on our servers. AI providers do not train on your code. Generation uses large language models from leading AI providers, accessed through their APIs under terms that do not permit using your code to train their models. We do not train models on your code either. What we do keep is the product output: the generated CLI, its documentation, run logs, and metadata about your runs (timing, status, which repository). Run logs may quote small excerpts of your code where relevant to an error. Access credentials are scoped and guarded. GitHub tokens used to clone your repositories are short-scoped, passed to sandboxes through private channels, and never written to logs.”Open citation