Capabilities

What NYXDB provides — tables and versions, attribute groups, transformations, streaming queries, transactions, replication, federation, security and operations.

NYXDB is a SQL database whose tables keep exact current and versioned state, and whose transformations, materialized views and streaming queries are maintained continuously from every committed write.

The examples throughout these docs use two tables: balances, a VERSIONED table of token balances per address, and transfers, an APPEND table of token transfers.

Tables and state

Capability	What it gives you
`MODE APPEND`, `MODE UPSERT`, `MODE VERSIONED`	Event logs, keyed current state, and keyed state with its full version history. An APPEND table carries no primary key.
Attribute groups	Several producers write one row, each its own group; a group patch leaves the other groups untouched.
`VALID BY` and `REVISION BY`	The version that wins is decided by declared coordinates, per table or per group, never by arrival order.
`AS OF COMMIT` reads	Read a table exactly as it was at an earlier commit.
Indexes	`MINMAX`, `BLOOM` and `POSTINGS`, scoped to current state, history or both. An index never changes an answer.
Storage profiles and codecs	`ROW`, `COLUMNAR`, `HYBRID` per table; per-column encodings and codec chains.
Partitions and shards	`SHARD BY` with a shard count or `SHARDS AUTO`; time partitions with `PARTITION BY`.

Derived state and streaming

Capability	What it gives you
`CREATE TRANSFORMATION`	A target table, or one of its attribute groups, kept current from every committed write: filters, `GROUP BY` folds, `LOOKUP`, `INNER`, `LEFT` and `ASOF LEFT` joins, chained from one transformation into the next.
Materialized views	Engine-owned targets kept current and read by name.
Logical views	Named queries for batch reads and streams.
`STREAM SELECT`	A snapshot followed by exact changes, over a table or a `TUMBLE`, `HOP` or `SESSION` window, with `WHERE`, `GROUP BY`, `ORDER BY` and `LIMIT`.
Emit triggers	`EMIT AFTER WINDOW CLOSE`, `EMIT PERIODIC`, `EMIT ON UPDATE WITH BATCH` and `EMIT PER EVENT`.
PSI	Every committed change is published once to every index, view, transformation and stream that reads it.

Transactions and consistency

Capability	What it gives you
Atomic statements	Every statement commits atomically and durably.
Multi-statement transactions	`BEGIN`, `COMMIT` and `ROLLBACK` group several statements into one atomic unit.
Exact derived state	A derived definition is exact: its target always equals the definition over its sources.

Replication, availability and recovery

Capability	What it gives you
Write-ahead log	A write is acknowledged once it is durable.
Replication and failover	Tablets are replicated across nodes and a replica takes over when a node is lost.
Tablet placement	Tablets move, split and merge as data grows.
Backup and restore	Consistent backups of a node's tables and restore to a new node.
Graceful stop	A stopping node drains in-flight work before it exits.

Security and federation

Capability	What it gives you
Users, passwords and API keys	`CREATE`/`ALTER`/`DROP USER` and `API KEY`; a password is passed as a parameter and never appears in query history.
Roles and grants	Read, write and subscribe rights per relation.
Federation	One instance reads another's tables through a source, under an application access whose grants carry a row policy.
Settings	The `SETTINGS` clause on a query, `SET` and `RESET` on a session, and node settings with their origin.

Operations and clients

Capability	What it gives you
HTTP and WebSocket	`POST /sql` and the write routes over HTTP; multiplexed requests and streams over `GET /client`.
TypeScript client `@nyxdb/client`	Both result forms, bearer-to-session exchange, bounded batch writes and resumable bulk loads.
System relations	Read-only `system.*` relations for tables, transformations, sessions, settings and more; see System relations.
Functions	A large function library; see the function reference.

See Deployment topologies for how a deployment is laid out.