Data Sandbox Management lets you rapidly create fully isolated data environments from a single source, with support for multi-level branching and instant reset. This ensures complete context separation for multi-agent workflows, A/B testing, and collaborative development.
Use cases
-
Context isolation for multi-agent sub-task flows
Multi-agent systems generate a massive number of sub-task flows. A data sandbox isolates the context for each task flow, preventing them from interfering with one another.
-
Safeguards for agent read/write operations
Agents with elevated permissions can potentially corrupt or contaminate data. Data Sandbox Management lets you create multiple independent branches from the same data source. If data corruption or contamination occurs, you can quickly reset the branch or create a new one.
-
Vibe Coding collaborative development
In Vibe Coding scenarios with collaborative development across multiple sessions, each session can have its own independent data sandbox branch for isolation. You can also create separate Dev and Prod environments to keep development and testing distinct.
-
A/B testing
Create two isolated data sandbox environments from the same data source to run parallel A/B tests.
Key features
|
Feature |
Description |
|
Branch management |
Uses copy-on-write technology for millisecond-level data cloning and supports multi-level child branches. Each branch has an independent runtime environment and isolated state, so parallel tasks do not interfere with each other. You can create a new branch from any point in time on the MAIN branch, making it ideal for parallel exploration and hypothesis testing by agents. |
|
Branch reset and restore |
If a branch's data becomes contaminated or encounters logical errors, you can reset its state from its parent branch with a single click. You can also restore a branch to a specific point in time. |
Billing
This feature is free during the invitational preview. For details on the preview schedule, see [Notice] AgenticDB for AnalyticDB for PostgreSQL Now in Invitational Preview.
Notes
-
Data Sandbox Management currently supports creating data branches for Supabase. Before you begin, you must first create a pay-as-you-go Supabase project.
-
Data Sandbox Management is currently in invitational preview. During this period, you can enable it in the China (Chengdu) region. To use it, submit a ticket.
Manage branches
Log on to the AnalyticDB for PostgreSQL console. In the left-side navigation pane, click to open the Data Sandbox Management page.
Create a branch
On the Data Sandbox Management page, your Supabase project is the MAIN branch by default. In the row for the MAIN branch, find the Operation column and click Create Branch. The new branch inherits the latest metadata and data from its parent branch at the time of creation.
-
Branch Protection: When you create a branch, you can enable this option. A protected branch cannot be manually deleted through an API and does not support branch reset.
If enabled, you can disable it later in the Branch Protection column for the target branch.
-
Create multi-level branches: You can create descendant branches, such as child branches and grandchild branches.
Reset a branch
-
You can only reset a branch to the latest data of its parent branch.
-
A branch reset is a complete overwrite, not a refresh or merge. Any local changes made to the child branch are lost during the reset process.
-
The reset temporarily interrupts existing connections, which are restored immediately afterward. Your connection details do not change.
-
The MAIN branch cannot be reset.
After delete protection is disabled, in the Operation column for the target branch, click the 
icon and select Reset from Parent Branch.
Delete a branch
If a branch has no descendant branches and delete protection is disabled, you can manually delete it by clicking Delete under the icon in the Operation column for the target branch. The Main branch cannot be deleted.
View branch details
Click the 
icon in the row of the target branch to view its connection information.
You can also click the target branch name to go to the branch details page, where you can manage the Overview, Connections and Console, and Child Branch.