Backups and recovery

Daily application backup runs around 03:15 UTC and briefly pauses the gateway/dashboard.Open full size ↗Mermaid source ↓

Restore rehearsal

On EC2 as the operator:

sudo martian-hermes backup --kind daily
sudo martian-hermes verify-restore daily/EXACT_KEY.tar.gz

Use the exact uploaded object key. Verification downloads the archive, checks hashes/CRCs, imports into a temporary home and checks restored SQLite databases. It leaves the live home untouched. Repeat monthly and before upgrades.

What is covered

Data	Recovery path
Hermes home, settings, sessions and skills	Application archive
Workspace reports and evidence	Workspace archive
Synthetic lab state	Data-volume snapshot
Tool runtimes	Rebuild from approved locks
Browser logins	Reauthenticate after application restore
Target registration and Mac relay	Separate protected backup required

Backups contain credentials. Keep them out of Git and public evidence. S3 is versioned; the instance role cannot delete backup objects. Three local bundles are retained. Snapshots are crash-consistent; no cross-region copy or external backup-failure notification is configured.

Recovery

• Application: stop Hermes, install the archive’s exact code pin against a fresh home, import home.zip, then restore workspace.tar.gz.
• Whole disk: restore an encrypted EBS volume in the required AZ, attach to a clean/stopped host and verify before admitting work.

Check version, SQLite integrity, skills, services and a model/tool call. Preserve failed state and reconcile work after the recovery point. Never run older code against a newer schema.

AWS Backup reference