Skip to content

Vectors, AI & ML

OverviewVectorAI

AetheriusDB serves vector search and AI workloads from the same engine that runs your SQL — no separate vector store, no model-serving sidecar to keep in sync. This section covers the stable foundation (native vector columns and kNN search) and the AI features layered on top, each with an honest availability badge.

Most teams run their vectors in a second system — a dedicated vector store beside the transactional database — and pay for it in sync jobs, double storage, and queries that can’t combine a filter with a similarity search. AetheriusDB stores embeddings as native columns and searches them in the same SQL, so a metadata filter and a nearest-neighbour ranking are one query, over one copy of the data.

  • One store, no sync — embeddings live next to the rows they describe; there’s no separate vector database to keep in step.
  • Filter and search together — pre-filter candidates with ordinary SQL predicates, then rank by vector distance, in a single statement.
  • No model-serving sidecar — the AI features run in-process against your data.
  • Indexed kNN, not a full scan. HNSW / IVFFLAT indexes answer k-nearest-neighbour queries quickly, and the SQL pre-filter shrinks the candidate set before the distance math runs.
  • Inline distances in plain SQL. Cosine / L2 ranking over a native TENSOR(N) column runs in the query engine — no export to an external service and back.

The stable foundation is two features you can use today:

  • Tensor & Vector Types — declare native fixed-dimension TENSOR(N) columns, store embeddings inline, and rank rows by cosine or L2 distance in plain SQL.
  • Vector Search (kNN) — build HNSW or IVFFLAT indexes and run fast k-nearest-neighbour queries that pre-filter with ordinary SQL predicates.
PageAvailabilityWhat it covers
Tensor & Vector TypesStableNative TENSOR(N) columns and distance functions.
Vector Search (kNN)StableHNSW / IVFFLAT indexes and kNN queries.
Semantic NLP & Text EncodingBetaHow text columns are encoded for faster joins and scans.
AI Quality ConstraintsBetaAI_CHECK anomaly detection at ingest with a dead-letter table.
AI Legacy TransmuterRoadmapMigrate a legacy schema into a queryable staging branch.
Multimodal FabricRoadmapAuto-extract embeddings, transcripts, and entities from media.
Cognitive KernelRoadmapAsk questions in natural language, in-process.