Overview

The building blocks of NYXDB — table kinds, transforms, streaming reads, storage, indexes, and the PSI routing model.

NYXDB is organized around a small set of primitives that share one internal contract. Understanding how they compose is the fastest way to model a workload.

Tables

Append, keyed, and attribute engines.

Transforms

Continuous, in-database pipelines.

Streaming reads

STREAM SELECT and STREAM HISTORICAL.

Streams

Ordered, replayable change data.

Projections

Wide, read-optimized reconstructions.

Storage model

Delta, parts, seal, compaction, policies, shards.

Indexes

Bitmap, minmax, set, bloom, HNSW.

PSI

The Predicate Subscription Index.

The query path

NYXDB keeps its query path interface-first. The parser accepts function-shaped syntax; the binder and runtime registry decide which active function version is valid for the current query.

SQL text
  -> ISqlParser
  -> IQueryBinder
  -> ILogicalPlanner
  -> IQueryOptimizer
  -> IQueryExecutor

This keeps hot-loaded functions and enterprise extensions isolated from the parser, and lets services compose shared engine modules rather than owning alternate implementations.