One engine. Streaming, querying, indexing, and durability under a single contract.
Most low-latency stacks split one workflow across an OLTP database, an analytical engine, a stream processor, a cache, and a queue. NYXDB makes them share the same internal contracts, so critical data never has to move between separate systems.
PSI — Predicate Subscription Index
The routing layer that connects committed changes to the components that react to them: current-state indexes, materialized views, schema listeners, and fan-out — matched on stream:key, not stream name.
Unified storage engine
WAL-backed durable writes, in-memory current-state indexes for low-latency reads, and columnar historical storage with compaction — under one contract, no external cache or queue.
Keyed & attribute tables
Keyed tables collapse to latest-per-key for O(1) current state. Attribute tables reconstruct entity state AS OF any point in time. Reads always reflect the writes that just landed.
Continuous transforms
CREATE TRANSFORM defines standing pipelines that run inside the database and stay fresh through PSI routing — materialized views without a separate stream processor.
SQL + Protobuf DSL
An interface-first query path — parse, bind, plan, optimize, execute — with 859 built-in native functions and a hot-loadable function runtime.
Deploy anywhere
One 154MB image runs a full single node for development. The enterprise runtime scales the same interfaces with distributed WAL ownership, RDMA transport, and cluster placement.
An interface-first query path.
The parser accepts function-shaped syntax; the binder and runtime registry decide which active function version applies. That keeps hot-loaded functions and enterprise extensions isolated from the parser, and services compose shared engine modules instead of owning alternate implementations.
SQL text -> ISqlParser -> IQueryBinder -> ILogicalPlanner -> IQueryOptimizer -> IQueryExecutor