Weaviate v1.36 is now available, introducing HFresh as a technical preview — a new disk-based vector index inspired by the SPFresh algorithm that dramatically reduces memory usage for billion-scale datasets by storing vectors on disk and using a compact in-memory HNSW centroid index. Five features graduate to general availability: Server-side Batching (dynamic flow control for data ingestion), Object TTL (automatic expiration with creation/update/date-property strategies), Async Replication Improvements (collection-level configuration via asyncConfig), Drop Inverted Indices (reclaim disk space by removing unused indices), and Backup Restoration Cancellation (abort in-progress restores). Additional improvements include HNSW snapshots enabled by default, VoyageAI V4 model support, non-blocking segment deletions, and various bug fixes.
Table of contents
HFresh (Preview) Server-side Batching - General Availability Object TTL - General Availability Async Replication Improvements - General Availability Alter Schema: Drop Inverted Indices - General Availability Backup Restoration Cancellation - General Availability Multiple Performance Improvements and Fixes Community Contributions Summary Ready to start building? Don't want to miss another blog post?Sort: