New Features
AI description management: generate descriptions and metadata with AI. It fills blank fields and leaves existing ones alone.
AI business relationship discovery: a new AI Definition quick-menu option that suggests likely relationships between tables using descriptions, indexes and column analysis.
Incremental ER Diagram: start from one table, then expand or collapse related tables step by step.
Custom-field test plans: filter models by custom fields (such as Epic or Feature) and build test plans from the results.
Selective paths in test plans: choose which paths from a model to include or skip when you add it to a test plan.
Windows Authentication for databases: connect to SQL Server using Windows Authentication (LDAP).
CSV support in transforms: mask CSV files with the new Mask FILE transform, use CSVs as a source or target in Transform data activities, and read from or write to CSV in Transform for Tables.
Time series generation: set generation parameters (start date, end date, frequency) for TimeVAE and TimeGAN models, in the UI and the API.
ML reports and charts: generate reports and charts during analysis, training and generation, so you can compare real, training and generated data.
REST API reference: a Swagger specification and reference for configuring and running test data jobs programmatically.
Test generation results agent: ask the AI assistant to explain why the generated paths for a profile behave as they do.
Enhancements
Bulk cosmetic changes can now select blocks by block type or condition.
Bulk-update the BPMN type of several nodes at once.
When linking subflows, only end nodes tied to the chosen start node are shown.
Labels update automatically when a subflow's end node is renamed.
Inclusive gateways now support branching inside them.
Generated test cases are listed in a logical order based on their position in the model.
The {PathNo} naming reference now starts at 1.
The Lock Model popup no longer appears when you create a new model.
Model variables have additional attributes.
Show Steps includes a new column for assigned data values.
AI path naming uses less data and respects a token budget.
Override the commit branch or connector when exporting.
Azure DevOps test suites appear in a hierarchy dropdown.
An Azure DevOps suite's Execute URL can be stored in a model variable.
Project and explorer metrics have been updated.
The job details and job queue screens have been redesigned, with live updates.
Minor changes to the activity menu.
The attach flow option has been removed from data generation.
The "Clone Dependencies" option is now limited to administrators.
The AI assistant reads notes and descriptions, and you can attach standard file types in any AI text interaction.
The MCP server now knows about projects, releases and folders, supports defaults, and has improved tool descriptions.
Column correlations support paging, filtering, sorting, select all, bulk enable and disable, and a uniqueness check.
Metadata export and import has several improvements.
A new bulk-load API for ML generation tables and schemas.
Pipeline data activities have a button to open the pipeline model, and data generation checks for an attached DLL before it starts.
New DLL actions to add, update and delete transform column rules.
The transform configuration has a CSV name template, plus a new accelerator for loading ML CSV files.
Machine learning:
Analysis can report on hyperparameter settings and apply them.
The Analyze job has new options.
A seed value makes training and generation reproducible.
Datetime columns can be split into year, month, day and hour.
Post-processing rules can be applied automatically, toggled in bulk, and copied from another rule set.
Training rule sets get smarter defaults (for example, PII columns start inactive).
Trained models keep their log file and show when they were created.
High-memory post-generation functions now show a warning.
Administrators can limit which generative engines are available in production.
Bug Fixes
Coverage profile "Ignore" settings are no longer overridden by data table row coverage.
Removing an automation function now also removes its variables.
Rules now work over subflows and within inclusive gateways.
Pools and lanes can be copied and pasted between models.
Data tables copied between models now map their variables.
End nodes from a submodel can be mapped again.
Relinking existing labels to subflows works again.
Cloned subflows are now saved in the new project.
Path-level custom fields now export correctly to Excel test cases.
Custom field values export as plain text instead of a JSON array.
The Jira connector saves correctly when you pick a project through search.
The Bitbucket connector no longer fails with too many redirects or authentication replays.
Confluence integration can retrieve documents again.
ModelGPT imports work with Azure OpenAI reasoning models.
The MCP server no longer corrupts subflow exit edges when it updates a model.
The connection dropdown can select a specific database and schema, even with many connections.
Apply Defaults on a masking rule set no longer throws an error.
Running data generation in parallel no longer errors when the VIP flow hasn't been uploaded.
The column filter on a generation rule set no longer shows an error when nothing matches.
Scanning JSON files that contain string values works again.
You can create a file definition from a new schema.
listlookupQ now uses more than 100 values from a list.
SeaTunnel masking Price() now returns a decimal.
ML data activities use the generative engine selected in configuration.
ML generation can now produce more than 100 rows.
Security fixes, including removing an exposed ML API key from training.json and October 2026 vulnerability updates.