Streaming

Metrics are published to standard formats (Prometheus, CloudWatch, DataDog) with zero configuration.

Authentication tokens are rotated silently in the background; your code never handles credentials directly.

const batch = stream.pipe(filter((e) => e.type === 'payment'), batch(100), debounce(5000));

The TypeScript types are fully inferred; no manual type assertions needed even with complex nested schemas.

Streaming

Batching is automatic—control it via `batch(100)` or `batch({ size: 50, timeoutMs: 1000 })` to trade latency for throughput.

const throttled = stream.throttle( rps: 100, burst: 200 ); const deduplicated = stream.dedup( key: 'event.id', window: 60000 );

Error handling is explicit: catch failures at any stage, log them, and route bad events to a dead-letter queue without stopping the main pipeline.