How does Kumo Engine scale?

This page is for Kumo Engine.

DiskGraph stores the graph index on local SSD and reads sampled neighborhoods for the model. This allows the graph to exceed system RAM without loading the full graph into GPU memory.

The October 21, 2026 synthetic commerce benchmark processed 50 billion nodes and 175.5 billion edges on one machine. Its 2.6 TiB index exceeded the machine’s 1 TiB RAM.

Warm sampling throughput normalized to 5.3 seconds per million seed evaluations at two hops, 19.9 seconds at three hops, and 38.9 seconds at four hops. These are derived rates, not measured end-to-end durations for one million customers. They exclude index preparation, seed mapping, worker initialization and model training, and depend on graph structure, sampling, cache state and hardware.

See scale and model-input results for the measured two-to-four-hop configuration and hardware details. For build time and resource use, see graph materialization. The tested graph size is a demonstrated workload, not a capacity ceiling.