Your data and backups

Your server keeps its data on three Railway volumes, one per service. Your jobs’ definitions are also in git.

What is where

Where	What it keeps
Postgres volume	Paperclip’s database: jobs, tasks, runs, agents, secrets (encrypted), apps and connections.
Paperclip volume (/paperclip)	Paperclip’s hourly database dumps, uploaded files, its own repo checkouts, plugins, and the template’s state.
CloudCLI volume (/home/node)	CloudCLI’s home folder: your Claude and GitHub sign-ins, chats, repo clones, the setup page’s state, and the installed tools.
Your GitHub repos	Each job’s skill, the jobs record (.claude/jobs.json), and Claude’s shared memory.

Restarts, redeploys and updates keep all of it.

Backups

Set up all three:

1. Paperclip’s own dumps. Paperclip dumps its database every hour to its volume, in /paperclip/instances/default/data/backups. This is on by default. How long dumps are kept is set in Paperclip’s instance settings (by default 7 daily, 4 weekly and 1 monthly).
2. Railway volume backups. Turn them on for each of the three services, on its Backups tab in Railway. A template can’t turn them on for you.
3. Keep PAPERCLIP_SECRETS_MASTER_KEY. It is a variable on the Paperclip service and encrypts the secrets you store in Paperclip. Copy it somewhere safe. Without it, a restored database’s secrets can’t be read.

Even without a backup, your jobs can be rebuilt from your repos on a new server. See Restore on a new server. Run history and secret values can’t: they live only in Paperclip.

What git keeps, and what it doesn’t

In git	Only on the server
What each job does (its skill)	Run history: tasks, transcripts, tokens
When it runs, its prompt and status (.claude/jobs.json)	Secret values
The names of the secrets it needs	Apps, connections and plugins set up in Paperclip
Claude’s shared memory, once committed	Chats in CloudCLI