How to copy Amazon S3 to a server you own

Sometimes the copy should be on hardware you own: an offline archive, a local working set, or a way off a provider. This job pulls an Amazon S3 bucket onto a server over SSH and keeps it current.

This job runs on the Business plan (Cloudflare R2 and Google Cloud Storage buckets, Azure sources and server destinations are Business features). Every account starts with a 14-day trial of Pro; see plans.

1. 1.Connect the source. Connect the AWS account with a one-click CloudFormation role you can revoke any time, or an access key with the minimal policy we publish. Then import the bucket under Buckets & Servers → Discover buckets, or create one there.
2. 2.Connect the destination. Under Credentials → Servers, add the machine: host, port, a user, and its private key or password. BucketPilot connects over SSH/SFTP from one fixed address, 3.111.107.194, so a firewall can allow exactly that and nothing else. The host key is remembered on first use and checked on every connection after.
3. 3.Open Backups & Migrations and press Create. Choose A bucket as the source and A server as the destination, then Continue.

1. 1.Pick the source bucket, and a folder inside it if you want only part of it.
2. 2.Pick the destination server and type the path where the files land.
3. 3.Leave the schedule on manual for a one-time move, or set one if the source keeps receiving files and you want them to follow. Choose how many files copy in parallel (threads); eight is a good default for a busy server.

1. 1.Press Create Job, then Run now. The first run copies everything; every run after it copies only what changed.

Every run is incremental: a file is copied only if its size or timestamp differs from the copy, so a nightly run of a large source moves only the day's changes and finishes in minutes.

Files are streamed from source to destination — multi-gigabyte files never sit in memory, and a run that is paused or loses its worker resumes where it stopped.

You can watch it: live progress on the job, a per-file log, and an email when a run finishes or needs attention. Transient errors retry with backoff; a run that fails outright is reported, never silently skipped.

A server keeps one copy of each file. Files that changed at the source are copied again; files removed at the source stay on the server.

The job's Restore button copies files from the server back to the bucket they came from — only what is missing, never overwriting a newer file, never deleting anything.