Release Notes

Version history for AIQu VEIL on the Snowflake Marketplace.

0.21.0

• Administrators can size the CPU training compute pool to match their workload, the same control already available for the encoding pool. See Compute sizing.

0.20.0

• A feature view always masks its feature vectors for non-bypass roles. A view that cannot be fully protected is removed rather than served.
• Protection commands report failure instead of success when a column cannot be masked.
• Removing protection from a column recognizes the app’s own policies again, so bypass roles can be changed.
• Service restarts and upgrades rebuild and re-protect every active view before it serves queries.
• Deactivating an encoder withdraws read access to its view.

0.19.0

• Encoder training standardizes each feature before measuring reconstruction error, so high-magnitude columns no longer crowd out smaller predictive ones.
• Encoded vectors retain downstream model quality close to the raw-data baseline on mixed-scale tables.