Blog

Northline 1.4

2 Apr 2026

Northline 1.4 fuses keyword weight into the same query path as dense vectors. You no longer build and hydrate a second sparse index for ordinary term matching.

What changed

Term scores are computed from payload text you already store. A query can ask for dense similarity, term weight, or both, and a single fusion step orders the hits.

Filters still run during the graph walk. A tenant predicate does not move to a post-filter just because the query is hybrid.

What did not change

Collections created on 1.3 open on 1.4. Quantization settings stay put. Clients that only send a dense vector behave as before.

Upgrade

Cloud clusters move during a maintenance window with no full reindex. Self-managed nodes roll one replica at a time. Read the migration note if you pinned a sparse collection by hand.