Vault · Compare
The incumbent isn't a company.
It's a workaround.
Nobody owns "undo any infrastructure change". People improvise it from provider snapshots, Git, local filesystem tricks, and backup tools that were built for a different job. Here's how the improvisations hold up.
The five ways people undo today
Each covers a slice. None covers the workflow: checkpoint in seconds, on any machine, roll back in one call, survive the hardware.
| Approach | What it is | Scope | Tempo | Cost shape |
|---|---|---|---|---|
| Edge Vault | Checkpoint & rollback for any machine | Files + app-consistency hooks, any host | Seconds (warm), one-call rollback | $0.02/GB-mo stored · $0 to roll back |
| Provider snapshots (EBS, DO…) | Whole-volume images, one cloud | Entire volumes only, locked to the provider | Minutes to take, longer to restore | ~$0.05/GB-mo + restore fees & cross-region charges |
| Git | Version control for code | Tracked source files only — not configs-in-place, data, or system state | Instant, for what it covers | Free — but it was never the answer for servers |
| ZFS / btrfs / LVM snapshots | Local filesystem snapshots | One disk, one machine — dies with the hardware it protects | Instant locally | Free — until the disk that holds them fails |
| Backup products | Nightly copies, restore someday | Files, on their schedule — not a pre-change workflow | Nightly cadence; restores metered or queued | Cheap to store, expensive the day you need it |
Feature by feature
Vault against the three workarounds people actually reach for.
| Capability | Edge Vault | Provider snapshots | Git | Backup tools |
|---|---|---|---|---|
| Works on any cloud / host / bare metal | | | | |
| Pre-change checkpoint workflow | | Manual, slow | Code only | |
| One-call rollback | | Volume-level only | Code only | |
| Survives the machine dying | | | If pushed | |
| Covers configs, data & system state | | | | |
| App-consistency hooks (databases) | | Crash-consistent | | Sometimes |
| Free restores at any size | | | | Rarely |
| Agent API & MCP tools | | | | |
| Client-side encryption by default | | | | Sometimes |
| Open repository format | restic | | | Varies |
And since every checkpoint is a backup — price the disaster
2 TB stored for a month, then one full restore. Published list rates, rounded. Storage pricing is engineered to look cheap until the day you actually need it.
| Provider | Storing 2 TB | Restoring 2 TB | The wait |
|---|---|---|---|
| Edge Vault Checkpoints, apps, fleet view & verification included | $40/mo | $0 | Starts immediately |
| Backblaze B2 (DIY restic) Bring your own client, keys, hooks & monitoring | $12/mo | $0–$20* | Starts immediately |
| AWS S3 Standard Egress ~$0.09/GB, plus request fees | ~$47/mo | ~$184 | Starts immediately |
| S3 Glacier Deep Archive Retrieval + egress; your disaster waits in a queue | ~$2/mo | ~$225 | 12–48 hours |
| Tarsnap $0.25/GB storage and bandwidth, on deduplicated data | ~$500/mo† | ~$500† | Starts immediately |
* Backblaze B2 includes free egress up to 3× your average monthly storage; beyond that, $0.01/GB. † Tarsnap charges $0.25/GB-month for storage and $0.25/GB for bandwidth, on deduplicated data. All figures from published list pricing; illustrative, not a quote.
The honest corner
When the workaround is fine
If everything you care about is code in a pushed Git repo, you already have undo. If you need bootable whole-machine images inside one cloud, provider snapshots do that and Vault doesn't pretend to. If you're archiving petabytes you'll never touch, Glacier's cold rates win. And if you enjoy running restic against a bucket yourself, Edge Storage at $0.015/GB-month is a great target — Vault uses the same open format, so you can graduate in either direction without re-uploading a byte.
Vault is for the gap in the middle — the configs, data, and state on real machines, changed by real hands and real agents, that today have no undo at all.
Stop improvising your undo
Vault is coming. Join the waitlist and be first through the door.
Join the Waitlist