Legal hold
A legal hold freezes the backups of a person, a mailbox, a site, a folder, a job or a whole customer: whatever retention says, Caryvane does not prune or delete them until the hold is lifted. New backups carry on as normal.
What a hold does, and what it does not #
- It stops Caryvane removing backup copies in scope. Version history is not pruned; Xero, Shopify and WordPress retention keep every version and snapshot; Proxmox keep rules and the storage area's ceiling keep every snapshot; a cloud-to-cloud copy does not mirror a source delete onto the backup; and an Archiving job copies held files to the archive but does not remove them from the held folder. Held items are counted as held in the run, not as errors, and each run logs one line saying how many were kept.
- It does not stop new backups. Jobs keep running and keep capturing changes.
- It does not stop anyone deleting a live file. A file deleted on a laptop is still deleted there; what the hold keeps is the backup of it. On a cloud drive, a deleted file goes to the recycle bin as usual, and the bin is not emptied while the hold lasts.
- It blocks deleting held backups by hand. Deleting a held folder or file in the storage browser is refused, and so is removing a held snapshot from Proxmox's own interface.
- Moving a held backup's destination is refused. For Xero, Shopify and WordPress, moving the destination starts a fresh catalogue, so it waits until the hold is lifted.
Place a hold #
- Open Legal Holds in the console's side menu (administrators: Managers, MSP and distributor admins).
- Place hold → pick the customer, then what to hold. The list offers the customer as a whole, each job, each storage connection (then type the container and folder), the mailboxes, OneDrives, SharePoint sites and Teams its Microsoft 365 jobs back up, its Google users, Xero organisations, Shopify stores and WordPress sites, Proxmox guests (type the VMID) and Hyper-V VMs.
- Give the hold a name (the matter or case) and a reason, then Place hold.
Jobs a hold affects show a Held badge on the Jobs screen.
Lift a hold #
Legal Holds → the hold → Lift, with a reason. Normal retention resumes at each job's next run, so what the hold kept may then be removed. Lifted holds stay in the list under Show lifted.
Who did what #
Placing and lifting a hold each write an entry to the audit trail: who, when, the scope and the reason. Holds are never deleted, only lifted.
Agents must be up to date #
Holds on files, drives, Microsoft 365, Google, VSS and Proxmox backups are enforced by the agent that prunes them. An agent older than 1.0.306 does not know about holds and keeps pruning as before. Each hold lists any agent in its scope that is too old, with its version, so update those agents before relying on the hold. Xero and Shopify retention run on Caryvane's servers and the Cloud Agent fleet is always current, so those are held straight away.
A new hold reaches an agent at its next check-in (about a minute) and a Proxmox node within five minutes. A run already in progress picks up a hold on its whole job at the next check-in too.
Enforced by Caryvane, on any storage #
Caryvane's own hold works on every storage provider, because it is Caryvane's pruning that respects it. On its own it does not change anything in the bucket, so someone with direct access to the storage, outside Caryvane, could still delete the data.
The storage-level hold (Object Lock buckets) #
On storage with Object Lock (Amazon S3, Wasabi, Backblaze B2, MinIO), Caryvane also sets the bucket's own legal-hold flag on every backup version the hold covers. A flagged version cannot be deleted by anyone until the flag is removed, whatever its retention, so the data survives even outside Caryvane: if someone goes to the bucket directly, or if Caryvane's records were lost.
- When it happens. Placing a hold starts a background job that flags every version in the hold's scope; a mailbox or a whole customer can be many thousands of objects, so it takes a while. The hold shows how far it has got, then the result, for example Wasabi legal hold set on 25 of 25 version(s). Caryvane reads a flag back from the bucket to check the storage really kept it, and says so on the hold if it did not. New backups of held items are flagged as they are written while the hold lasts, and a nightly pass flags anything written since that could not be flagged at the time.
- What is flagged. For a job, the job's own backup folder; for a mailbox, OneDrive, SharePoint site, Team or Google user, that person's or site's folders inside each job that backs it up; for a storage folder, that folder. Where another customer's backups share the very same folder, those versions are never flagged.
- Lifting. Lifting the hold starts a job that removes the flags again, except on versions another hold still covers. Normal retention then resumes.
- Proxmox. The hold covers every chunk a held guest's snapshots use, not just their indexes. The cluster's nodes apply it within minutes of the hold, and again at each garbage collection.
- Which storage. Storage with immutable backups on, or a bucket whose Test showed Object Lock enabled. Elsewhere only Caryvane's own hold applies, and the hold says Storage has no Object Lock: Caryvane's hold only.
The honest limit: anyone with the s3:PutObjectLegalHold permission on the bucket can remove a legal-hold flag, including whoever holds Caryvane's storage key. The flag protects against mistakes, routine pruning and anyone without that permission; against a stolen key, Compliance-mode immutable backups are the stronger protection. Where you can, give Caryvane a key without s3:PutObjectLegalHold for everyday backups (holds are then enforced by Caryvane but not flagged in the bucket), or keep that permission on a separate key.
From an AI assistant #
The MCP tools list_legal_holds (read), and place_legal_hold and lift_legal_hold (keys with full access) do the same; get_storage_operation follows the storage-level flagging. See AI assistants & API.