cc-sync

Edit locally, run remotely. A small Python CLI that mirrors your working tree to a remote host via one-way rsync.

Install

conda create -n cc-sync python=3.12 -y
conda activate cc-sync
cd cc-sync
pip install -e .

Quick start

cc-sync is best for quick testing and debugging, and below are the core commands for this purpose.

cd my-project
ccsync init                       # writes .ccsync.toml
ccsync push                       # one-shot rsync to remote
ccsync pull logs/                 # pull a path from remote → local (reverse of push)
ccsync watch                      # auto-sync (push and pull) on every save (Ctrl-C to stop)

My typical workflow:

• top left: editor.
• bottom left: claude code on local.
• top right: ccsync watcher, synchronize local changes to the remote machine on every local save.
• bottom right: a terminal on the remote, running my training/inference workload.

Config (.ccsync.toml)

[remote]
host = "mybox"          # ~/.ssh/config alias
path = "/home/me/proj"
# Optional — set these if `~/.ssh/config` doesn't already resolve them.
# user = "ubuntu"
# port = 22
# identity_file = "~/.ssh/id_ed25519"

# Optional — uncomment and set after `salloc` to hop from login to a compute node.
# tmux still lives on the login host; the inner ssh runs your command on `host`.
# `path` defaults to [remote].path (shared FS); only set it if it differs.
# [compute]
# host = "gh002"
# path = "/scratch/me/proj"

[sync]
exclude = [".git/", "__pycache__/", "target/", ".venv/", "logs/", "outputs/", "checkpoints/"]
debounce_ms = 500
delete = false
log_pull_interval_s = 15  # 0 disables; >0 mirrors remote .ccsync/*.log → logs/cc-sync/ during `ccsync watch`

[pull]
paths = ["logs/"]

[run]
tmux_prefix = "ccs"
log_dir = ".ccsync/logs"
shell = "bash -lc"

Working with a coding agent

This repo ships a CLAUDE.md — a short guide that teaches a local coding agent (e.g. Claude Code) how to operate with ccsync: edit locally, push to the remote, hop to a compute node, and launch long-running work in tmux. If you use a coding agent on a ccsync-mirrored project, point it at that file (or copy it to your project root) so it knows to run remote work on the remote rather than the laptop.

Other commands (still testing)

Below are some experimental commands we are still testing. They are ready to use but we are still optimizing them for better convenience and robustness.

ccsync run pytest -x              # sync + run foreground
ccsync launch train bash train.sh
ccsync logs train -f              # tail remote log; auto-pulls artifacts on exit
ccsync attach train               # jump into the tmux session
ccsync ps                         # list ccs-* sessions
ccsync install-hooks              # wire up Claude Code PostToolUse auto-sync