Recovery & Delete Control is now in early access: use it on your own database nowuse it now

Early access

Recovery & Delete Control

A copy that survives, and destruction that waits for it. Everything runs in your cloud accounts, installed with your own rights; each step below says what it depends on.

1

An installation you review first.

Before anything is installed, we check that your database engine, plan, permissions and connectivity are supported. We show you the exact rights involved, the data the copy will not cover and the impact on production.

You deploy the copy account and its storage from a template we hand over, with your credentials. The copy lives in a dedicated account of your cloud organization, in storage locked in compliance mode, with keys in your KMS and policies that deny deleting or disabling them. A verifier checks the result before the first copy.

It is an infrastructure installation, even when the templates make it short. During early access, we do it with you.

2

A copy outside production's control plane.

The copy worker connects with a replication role you create from SQL we generate. Once you confirm the estimated data transfer cost, it takes a first full copy.

Every point is encrypted with a key from your KMS and written to the copy account, append-only and locked for a set lock period. The agents, scripts and CI that work on production cannot delete a locked point or shorten its lock period. Your organization's management account keeps powers of its own: see the limits.

Continuous capture is designed to protect production first: an explicit list of tables, a bounded replication slot and a heartbeat. If the copy falls behind, it records a gap; your production disk does not fill up.

The copy worker reads production through a replication role and encrypts every point with a KMS key that DataPrism does not hold. Each point is written to a separate copy account, append-only, and stays locked for its lock period.

3

A restore you can show, on a regular schedule.

On a regular schedule, the copy is restored into a throwaway database, and what comes back is checked against what was captured. Today, it runs every week.

The result is an attestation signed with a key from your KMS. Failed runs are recorded too. You and your auditor check it without a DataPrism account, and a third party can replay the restore.

Each attestation is designed to record how long the restore took and on what instance size, so you know your recovery time before you need it.

On a regular schedule, the locked copy is restored into a throwaway database, and what comes back is checked against what was captured. The result is signed into an attestation, which an auditor verifies without a DataPrism account.

4

Destructive changes wait for a verified copy.

Agents, scripts and CI keep the access they need to work. In the design, the right to destroy (deleting a database, its backups or its keys, running terraform destroy) moves to an executor.

The executor grants a destruction only when a locked copy, restored successfully, covers the state being destroyed, and it writes every grant and every refusal to a log you check without us. It starts in observation, recording what it would have granted or refused, before it enforces anything.

On AWS, an organization policy denies those actions to every identity in your member accounts except the executor. On Supabase, agents work under a Postgres role that owns no object and cannot run DDL or TRUNCATE. We do not offer it where a platform cannot deny destruction action by action.

$deletedatabaseprodrefusedgranted
A design, not part of the demonstration. Agents, scripts and CI keep the access they need to work. An organization policy denies the destructive actions to every identity except an executor installed in the cloud account, so the only path to destroying production runs through it. The executor is designed to grant a destruction only when a locked copy, restored successfully, covers the state being destroyed, and to write every grant and every refusal to a log that is checked without DataPrism.

5

After an incident, you restore without us.

You pick a point in time, just before the bad command, and restore it with your credentials and the restore tool you hold: into a new Supabase project, a standard Postgres or Amazon RDS. No DataPrism account is involved.

The copy covers the database: tables, schemas, Supabase Auth accounts and Storage metadata. What it does not cover, and what can be lost: see the limits.

You pick a point in time just before the bad command and restore it from the locked copy with the restore tool: back into production, to the side to recover only the lost rows, or at another provider. No DataPrism account is involved.

Supabase, Neon, AWS and more.

The design is provider-agnostic: everything goes through adapters, and each promise covers only the combination you install.

Supabase
In early access. Postgres 17 is required for a replication role that cannot write.
Neon
In early access. Neon's replication role can also write, and we say so wherever it matters.
Amazon RDS and Aurora
In early access, with a copy taken on a schedule.
The copy
Stored in a separate cloud account that DataPrism cannot reach, locked for a set lock period.

See it on your own database.

Tell us where your production database runs today. Pricing on request.