Fixed an issue with the Structural Agent when LLM provider is configured from the environment settings view.
Structural Agent chat for workspace analysis and configuration - Structural’s AI features now include the Structural Agent, a chat-based tool that can help you to understand the sensitive values in your source data and then assign and configure generators to protect those values. The Agent starts when you create a workspace - each workspace has its own chat. Use natural text prompts to ask questions about the data and apply generators. The Structural Agent is automatically enabled on Structural Cloud. For self-hosted customers, the Structural Agent becomes available when they configure a connection to an LLM.
Added TLS protocol and cipher configuration for the Notifications and Worker services.
Schema caching - fixed an issue where the application incorrectly suggested that schema changes were detected.
Improved the workspace loading experience with better visibility into loading progression.
File connector - Added support for the following data types in Parquet: Binary, BinaryView, Decimal32, Decimal64, Dictionary, Duration, FixedSizeBinary, FixedSizedList, Interval, List, Map, StringView, Struct, and Union.
You can now export and import your custom generator presets. During the import:
- If the identifier of an imported preset matches an existing preset, the existing preset is updated.
- If the identifier of an imported preset doesn’t match an existing preset, a new preset with that identifier is created.
Fixed upload of large JSON files based on a prefix filter so that they validate and upload properly. Also fixed some scenarios that could fail with JSON elements larger than 2GB.
Added the Text Composition generator, which allows you to configure the output value for a column as a template that incorporates values from other columns.
For the Alphanumeric String Key generator, the new Preserve prefix option keeps a specified prefix unchanged during masking. The new Mask digits only option is used to only mask numerical digits, but preserve all of the other characters in place.
The JSON Mask generator no longer promotes JSON objects or arrays that are converted into strings into actual JSON objects or arrays.
Refined the Hostname generator output. This is a consistency breaking change.
In the workspace Schema Changes settings, Structural now preserves the state of the Automatically apply generators setting under Do not block. If you change to one of the blocking options, and then change back to Do not block, the Automatically apply generators setting has the same enabled or disabled state as before.
For post-job scripts, the Enable Warnings toggle is now enabled by default.
SQL Server - SQL Server data generation now always uses the Data Pipeline V2 process. It does not support the previous process. The Confirm Generation panel no longer allows you to choose which process to use.
A new environment setting, TONIC_LLM_ENABLE_ENHANCED_RECOMMENDATIONS_SAMPLE_DATA, determines whether LLM-based sensitivity scans include sample data. Using sample data improves the quality of the detections. When set to false, only schema information is sent. The default value is true.
Structural Cloud users who have the Account Admin permission set can change this setting from the new Organization Settings tab under Structural Settings.
SQL Server - Added an environment setting TONIC_INDEX_RANGE_PARALLELISM to use to configure parallelism for passthrough copy commands for tables that have single-column numeric indexes.
Amazon Redshirt - For subsetting, added support for date/time primary keys and fixed table identification issues.
For self-hosted instances, Structural telemetry now includes database schema information.
SQL Server - SQL Server data generation now always uses the Data Pipeline V2 process. It does not support the previous process. The Confirm Generation panel no longer allows you to choose which process to use.
LLM-based sensitivity detection on Structural Cloud
On Structural Cloud, sensitivity scans now include more comprehensive sensitivity type identifications that are driven by LLM inference. The LLM-based sensitivity detection analyzes schema metadata and a small set of 3 or fewer data samples, which allows Structural to provide better type identification and expanded generator suggestions.
For more information about AI use in Structural, go to How Structural uses AI in the Structural user guide. To learn more about how these enhanced recommendations work or to discuss how to disable them for your account, contact support@tonic.ai.
Added global permission validation for the following API endpoints:
GET api/environment/settingsGET api/environment/settings/environment_overridesGET api/external-secrets-managers
Added the following new API endpoints that include workspace permission validation:
GET api/workspace/{workspaceId}/environment/settingsGET api/workspace/{workspaceId}/environment/settings/environment_overridesGET api/workspace/{workspaceId}/external-secrets-managers
On the column and field configuration panel, all of the available reset options are combined into a single Reset menu next to the generator selection. Depending on the context, you might be able to reset to Passthrough, to the preset baseline configuration, to the configuration for the matching path generator, or to the parent workspace configuration.
Structural no longer outputs log messages and associated UI warnings about bad data or missing fields during file processing for every erroneous row. Instead, Structural emits the first error within a file, then emits a summary of the errors at the end of the file processing.
Fixed an issue where blank lines in a file were not counted towards the ‘Skip’ count, which sometimes caused incorrect processing.
Tonic Structural now supports CyberArk Central Credential Provider as a secrets manager provider.
MySQL, Oracle, PostgreSQL, and SQL Server: Added a database networking performance check to the Test Connection process, and as a data generation pre-job check .
SQL Server - Fixed overflow error when calculating large database sizes.
Snowflake - Fixed a subsetting issue where jobs failed to process if a table contained no primary keys but had rows that contained many NULL foreign keys.
Fixed an issue where very large job log files sometimes failed to download.
File connector: Updated the file group list and details pages to display additional file group settings.
Amazon Redshift - The Lambda-based data generation process for Amazon Redshift is no longer supported. All data generation for Amazon Redshift now uses the Data Pipeline V2 process.
