L2025.4 SP2 is now Cloud Availability.

Release Summary:

What’s New
Primary Filters:

• Designers can now define primary filters that dynamically trim values in regular filters.
• Great for hierarchical filtering (e.g., Store → Product/Brand).
• Viewers don’t see a UI distinction; they just get context-aware filter values.
• Supports locking, background filters, AND/OR logic, and multiple primary filters (though >1 is not recommended).
• Performance note: Each primary filter triggers extra queries → potential slower dashboards + higher CDW query cost.
• Limitations: Not compatible with Custom RLS; partial issues in Jump-to-Dashboard flows.

 Improvements

• Build: Change join types (left/right/full) without rebuilding an ElastiCube.
• Dashboards: New navigation tabs for Paldi Workspace + QbeeQ Framework.
• Data Modeling: Prevents parallel imports; now shows 409 if import is already running.
• Intelligence:
  • Self-Hosted Vector DB now GA
  • Improved stability for cloud vector DBs
  • Better LLM integration (OpenAI org config, custom base URLs)

 Fixes
Build & Modeling:

• Publish Semantics button now appears when using custom columns.
• Refresh Schema for all tables works normally.
• Removing columns with relationships no longer breaks access.

Dashboards:

• ClickHouse DESC sort now treats NULLS correctly (NULLS LAST).
• PDF export now respects disabled filters via WAT.

Data Security: New import merge modes: skip, mergeOld, overwrite.
ElastiCubes: “Stop When Idle” no longer stops ECs prematurely.
Git: Pulling complex models no longer fails on false duplicate OID errors.
Multitenancy: Main-tenant datasources can no longer be shared with subtenants.
Perspectives: Assistant icon now shows consistently.
Users: Data Designers can work with shared models even when Custom Code is enabled.
Web Access Token: Page now renders correctly with the new nav bar.

Link to L2025.4 SP2 Release Notes.