Your other machine already pulled. reposync clones every registered repo to each peer host and fast-forwards trunk on filesystem events, skipping anything dirty, busy, or diverged.
brew install yasyf/tap/reposync
reposync repo add ~/Code/my-repo
reposync installThat's the whole setup. reposync sync runs the same converge pass the daemon now runs for you:
Driving with an agent? Paste this:
Install reposync with `brew install yasyf/tap/reposync`.
Register my most-used repo with `reposync repo add <path>`, then run `reposync install` so the daemon keeps it converged.
Run `reposync sync` and report each repo's outcome (fast-forwarded, busy, up-to-date).
Switching machines starts with a round of git pull across every repo you touched last week — or you skip it and grep three-day-old code. Register the daemon once per host:
reposync installsynckitd waits out a 15-second quiet window after a repo's git metadata changes, then converges it and notifies your peers to do the same. By the time you've sat down at the other desk, main already moved.
A new project exists on the machine where you started it and nowhere else. Register it:
reposync repo add ~/Code/new-thingThe repo joins a convergent registry that pull-merges across the mesh, and every peer clones it at the same relpath under its own ~/Code, preferring jj git clone --colocate and falling back to plain git. --local-only keeps a repo off the mesh, and reposync repo rm untracks everywhere without deleting a checkout.
A cron'd git pull doesn't know you're mid-rebase. An in-progress git or jj operation, activity within the idle threshold, a dirty tree, or a diverged trunk all gate the advance, and the skipped repo says why:
$ reposync sync
✓ notes: busy (dirty working tree)A stale lock left behind by a killed git or jj process no longer wedges a repo: sync clears a packed-refs.lock after 30 minutes, clears a jj lock of the same age once a flock probe confirms its holder is dead, and logs what it removed.
The only write it ever sends is a fast-forward push of your own trunk, and only after the repo has been quiet past push_after (a day by default).
| Command | What it does |
|---|---|
reposync |
Open the TUI: toggle discovered repos on the Repos tab, add peers on the Hosts tab |
reposync repo add <path> |
Register a repo and converge it everywhere (--local-only keeps it on this host) |
reposync repo discover |
List git/jj repos under default_location and whether each is tracked |
reposync repo ls / reposync repo rm <path> |
List registered repos / untrack one (the checkout stays) |
reposync sync |
Run the idle-safe fetch and fast-forward pass by hand |
reposync host ls |
List the peer mesh synckitd manages |
reposync install / reposync uninstall |
Register or remove the synckitd manifest that drives the daemon |
Run reposync --help for the full command tree and flags.
reposync itself is not a daemon. reposync install registers a manifest with synckitd, the shared sync daemon, which spawns reposync rpc-serve over stdio and drives it through a typed contract — no shell templates, no per-tool launchd agent. Repos are tracked by path relative to default_location, so each lands at the same place on every host, and the registry is a last-writer-wins CRDT: adds and removals pull-merge across peers, removals as tombstones. Adding a peer on the Hosts tab ssh-bootstraps it, brew-installing synckitd and reposync and cloning every tracked repo. A sync is pull-first: fetch plus a safe fast-forward, anchored to the git HEAD the fetch observed, so a git commit racing the advance aborts it instead of losing work.
State lives at ~/.config/reposync/state.json, sharing one file and one flock with synckit's host identity.
Status: pre-1.0, in daily use across my own machines. Licensed under PolyForm Noncommercial 1.0.0.