On github.com: profile picture → Copilot settings → set "Allow GitHub to use my data for AI model training" to Disabled. That one toggle is what actually controls it — it is not on by default being "fixed" by a repository or org setting elsewhere.
GitHub documents this as governing "your interactions with GitHub features and services — including inputs, outputs, code snippets, and associated context" for AI training and model improvement purposes.
GitHub's own documentation scopes this to individual Copilot plans — Free, Pro, Pro+ — and notes that it does not appear the same way for Business and Enterprise subscribers, who are instead covered by GitHub's Data Protection Agreement terms rather than a personal toggle. If you're on an org-managed seat, check with whoever administers the org's Copilot policy, since an individual toggle you see (or don't see) may not be the whole picture.
A per-account setting is still a setting on someone else's platform: it can apply differently across plan tiers, it only covers Copilot specifically (not every current or future AI feature a hosted platform might add against your repository contents), and it requires trusting that the toggle stays where you left it through policy changes. Running git on your own server removes that dependency — there is no Copilot-style feature in Rithub to opt out of, and no repository content is sent to any outside AI provider at all.
Sign up freeImport from GitHubRithub's robots.txt at rithub.dev/robots.txt publishes an explicit, machine-readable ai-train=no content signal — not just product copy, a file any crawler or AI company can read directly. Run it yourself and there's no rithub.dev-level signal to trust at all; your code never leaves infrastructure you control.