Riftri

How Riftri works

Riftri changes how worktree files are stored. Git still owns branches, commits, history, and worktree registration.

View .md / Copy .md

Create once, share unchanged data

  1. Check support for the requested checkout and destination.
  2. Prepare a base: an immutable copy of the exact Git tree, built or reused.
  3. Create a worktree: Git registers a real linked worktree, the filesystem shares unchanged data from the base, and Riftri verifies Git reports it clean.

Only compatible worktrees reuse a base; different trees, checkout settings, repositories, or volumes never share one.

Edits stay private

APFS, Linux reflink filesystems, and ReFS keep changed blocks private through native copy-on-write; OverlayFS stores changes in a separate writable layer. Riftri is not in the normal read/write path and needs no daemon.

Cleanup is recorded

Lifecycle operations are recorded so interruptions can be recovered. Shared bases are removed only when unreferenced; uncertain ownership or contents stop cleanup and preserve the data.

See safety for precautions. Full component design, storage layout, and transaction state machines: View .md.


Edit on GitHub