Billions of nodes, ingested in seconds, from one binary

GreyCat is a single self-contained engine with SIMD/C-accelerated hot paths — no JVM, no cluster of containers — that ingests CSV at roughly 1.7 million rows/second and scales to billions of persisted nodes. Below are the public, documented numbers, the real-world deployments behind them, and a harness you can run yourself.

The headline numbers

1.7M

CSV rows/second ingested — documented at doc.greycat.io

4.8s

to ingest 8 million rows (1.3 GB) in the documented example

~3.5MB

single self-contained binary — no JVM, no containers required

Billions

of persisted nodes on the same single-node engine

These are documented figures from the official documentation; your hardware, schema and data will vary. Run the harness below to measure on your own machine.

Ingest throughput

CSV import is a C-accelerated hot path. The documentation records ingesting 8,000,000 rows (1.3 GB) in 4.8 seconds — about 1.7 million rows/second — on a single node.

8M rows in 4.8 seconds

A concrete, documented run: 8 million CSV rows (1.3 GB) imported in 4.8 seconds — roughly 1.7 million rows/second.

SIMD + C hot paths

Parsing and storage run on SIMD/C-accelerated code paths inside the engine — not a layer bolted on top of a JVM runtime.

Schema auto-inference

CSV schemas can be inferred with Csv::analyze / Csv::generate, so the same fast path covers both modelling and bulk load.

Footprint & deployment

The whole engine — database, language runtime and HTTP server — is a single self-contained binary of about 3.5 MB. No JVM, no sidecar services, no orchestration of containers required to run it.

One ~3.5 MB binary

A single self-contained executable — database, runtime and server in one file you can copy and run.

No JVM, no containers

Nothing to provision around it — no JVM, no required container fleet, no separate queue or web server to deploy.

Scales to billions of nodes

From ARM/Raspberry Pi up to terabytes and billions of persisted nodes on the same engine — vertical scaling on a single node.

Real-world public scale

Documented figures, not synthetic ones. The Kopr national electricity-grid digital twin (kopr-twin.com) runs on GreyCat at the following public scale.

1M

grid assets modelled in the Kopr digital twin

330k

delivery points tracked live

45B

meter readings ingested per year, with continuous machine learning

Reproduce it yourself

Numbers you cannot run are just claims. A reproducible benchmark harness is published as an open repository — clone it, point it at your data and measure on your own hardware.

Open benchmark harness

The harness lives at github.com/datathings/greycat-bench — run it to reproduce ingestion and query timings on your own machine.

Single-node, honestly framed

GreyCat is single-node-focused today: strong vertical scaling, but no built-in horizontal sharding. We report what is measured and documented, and label anything that is not independently reproducible.

Accuracy note: the ingestion and footprint figures above are documented at doc.greycat.io; the deployment figures are the public numbers published by the Kopr project. Results depend on hardware, schema and dataset — use the harness to verify on your own setup. We do not publish head-to-head wins against named competitors without a citable source.

Top