Workflow definitions now support richer built-in tool steps for report generation and structured post-processing. API consumers can save workflows that call an LLM with a JSON Schema contract, write downloadable artifacts, and see whether a workflow contains non-deterministic model steps.
Workflow enablement updates can now preserve the originating DB user when called through trusted agent M2M authentication. This keeps audit columns accurate for agent-driven workflow operations while continuing to ignore caller-supplied audit fields for normal user-authenticated requests.
Skills can now mark linked workflows for deterministic pre-execution. When a skill is invoked, every auto-trigger workflow runs before the LLM reasons about the skill, guaranteeing its output is available as context.
New deploy endpoint and isDeployed parameter support across data product entity APIs, enabling a fully API-driven data product lifecycle without requiring the frontend UI.
Manage individual tiles within data app dashboards without replacing the entire dashboard configuration. These endpoints enable granular tile-level operations — add, update, or remove tiles independently while maintaining optimistic concurrency control against the parent data app.
Refresh data app tile data on demand by executing their SQL queries through the existing run-sql pipeline. These endpoints enable dashboards to fetch fresh results for all tiles at once or for a specific tile, with bounded parallel execution and per-tile error isolation.
Manage semantic model tables and relationships through dedicated CRUD endpoints. These sub-resources allow API consumers to build and modify semantic models programmatically — defining table structures, column metadata, and join relationships without editing raw YAML.
Submit feature requests programmatically via a new API endpoint. Any authenticated user can submit feature requests through the API, CLI, or MCP.
Fixed a bug in chat thread feedback filtering where admin feedback detection only considered account:{accountId} role scope. In workspaces where reviewers had global admin role but no account-scoped admin role, adminFeedback=true could incorrectly return zero results.
The chat threads list endpoint now returns separate feedback aggregates for end-user feedback and internal admin review feedback. This lets clients render clearer moderation and quality signals without inferring reviewer roles from raw feedback records.
