remote-sync

Bidirectional project sync over SSH. A simple, vendor-neutral tool to push local project code to a remote machine and pull outputs back — without Git, SDKs, or editor lock-in.

Why remote-sync exists

When training models or running long jobs on remote machines (Vast.ai, EC2, bare metal, lab servers):

• Git is too slow for iteration
• Editors require SSH sessions
• Outputs stay remote
• Re-downloading data is painful
• SDKs add lock-in

remote-sync solves exactly one problem:

Keep a local project directory and a remote project directory in sync — safely and predictably.

• Nothing more.
• What remote-sync does
• Pushes local → remote paths
• Pulls remote → local paths
• Uses rsync over SSH
• Works with any system that has SSH
• Supports raw SSH commands (Vast.ai compatible)

What remote-sync does NOT do

• ❌ Launch machines
• ❌ Manage infrastructure
• ❌ Install dependencies
• ❌ Run training jobs
• ❌ Replace your editor

It is intentionally minimal.

Requirements

• Python 3.9+
• SSH client (ssh)
• rsync

Access to a remote machine via SSH

On macOS / Linux, these are usually already installed.

Installation

Development / local install (recommended)

From the repository root:

pip install -e .

This installs the remote-sync CLI in editable mode.

Verify:

remote-sync --help

Configuration

remote-sync uses a YAML configuration file.

Default config file name remote-sync.yaml

You can override it with --config.

Basic configuration example

remote_sync:
  ssh:
    user: root
    host: 1.2.3.4
    port: 23456

  push_paths:
    - local: .
      remote: /workspace/projects/biomass-experiment

  pull_paths:
    - remote: /workspace/projects/biomass-experiment/output
      local: ./outputs

  ignore_paths:
    - .git
    - __pycache__
    - outputs
    - data
    - kaggle.json

TODO:

• Use git path command to update the remote
• Support for storing creds in session variable
• Support for rysnc + custom command (basically run rysnc + a following command)