I wanted to post a suggestion in the roadmap relating to requesting CLI/MCP controls be brought in at the account / project level. it won't let me because this post is too similar - it wont even let me override it.. lost the post a few times. If anyone wants to suggest that, let me know. I will support that thread.

Well pretty much the only way to back up production data is through Railway backups, which can be managed in the database configuration panel.

If an AI agent (eg, Cursor) has access to Railway CLI or any access to production environments, you should always be mindful and supervise the commands it's executing.

It is being indicated that even that would not be enough to contain the blast radius of this potential issue.

https://x.com/lifeof%5Fjer/status/2048103471019434248

They did indicate the situation is resolved. I am just wondering how to avoid it.

You can create backups to S3-compatible providers as Domehane suggested above.

>for your postgres db deploy the official railway template railwayapp-templates/postgres-s3-backups on github , it runs pg_dump on a schedule and pushes backups to an external s3-compatible bucket like cloudflare r2 or backblaze b2 and since these backups live outside railway, a volume deletion can't touch them

So I have to setup and monitor an additional backend provider to backup my backend provider? This is a hard idea to break, but that seems excessive, and I have a number of different files to backup ( I made a 2nd db just for logs )...

Maybe I just start with a backup bucket inside of the same project to keep it together...

It feels like there is a better way to approach this.

you can actually use a railway bucket as your backup destination , railway's own docs list backups as a valid use case for buckets and since they're fully s3-compatible and the backup template supports it via the aws_s3_endpoint variable. so no external provider needed everything stays inside railway

and as i mentionned before if the whole project gets deleted, the bucket goes with it since buckets are services inside the project. railway does give you a 48 hour window to recover a deleted project though

Hope i help you find the answer :)

Quick context on the agent incident: we're rolling out additional safeguards around agentic workflows. One already live: delayed deletion when the API is used to drop a volume. We're also decoupling backups from volumes so a deletion can't take them with it. More coming on both safety and usability for agentic workflows 🙂

On backups generally, for anything important I'd keep an off-site copy. A Railway bucket inside the same project is convenient, but it's not a real backup if something takes the whole project down. The 3-2-1 Backup Strategy is a solid write-up if you're curious.