Three ways to run it.

The same engine in every case. What changes is who operates it and where your bytes live.

Shape	Who runs it	Licence	Price
Hosted	We do	Managed	$10 per GB
Self-host, connected	You do, on your hardware	Signed licence with a revocation heartbeat	$10,000 per node
Air-gapped	You do, with no network at all	Offline signed licence, no phone-home, ever	$10,000 per node

why per node, not per core

The box decides the numbers, not the invoice.

The instance reads your hardware at boot and tunes every knob to it: shards map to physical cores, residency to RAM, stripes to the disks it finds, and the SIMD kernel to your CPU. We do not want a pricing model that punishes you for putting it on a bigger machine.

What you are actually buying

One 64-byte container that is simultaneously the record, the index entry, the log entry and the network frame. One copy of the data, queryable the moment it lands, through graph, search, vector, key-value, analytics and time-series lenses. The collapse from four to six systems into one is the saving, and it is an operations saving before it is a licence saving.

Where we do not win

BM25 relevance ranking over a large text corpus is Elasticsearch's home turf, and sublinear approximate nearest-neighbour at billion-vector scale is what the dedicated vector databases are built for. We do exact, not approximate, and uniform-random access over payload is bounded by physics like everyone else's. If your workload is one of those, say so and we will tell you honestly whether to buy.