Benchmark
Measured on our workloads.
Run it on yours.
Flat P99 at the engine, at production rule-set size.
The numbers
8ms
P99 at the engine
8000
patterns, throughput flat
5000
CPU cores replaced
0
GPU consumed
Third-party validated benchmark on NOL8 test workloads and rule sets. Your figures will differ. 5,000 CPU cores measured across 20 servers at production rule-set size. Get a demo →
Why the tail does not move
Inspect every payload and your latency at P99 does not move.
Deterministic matching gives the same result every time and holds throughput flat as the rule set grows, from 1 pattern to 8,000, which is where a software-only path gives out.
Three certainties
Control over every payload your AI sends and receives, on every request, at no cost to your latency or hardware.
| Certainty | The claim | What proves it |
|---|---|---|
| In what your AI does | Every payload is inspected, routed and redacted inline, on the way in and on the way out. Not a sample. | Full coverage. Inline placement means data cannot route around it. |
| In how it performs | Same input, same result, at 8ms P99 at the engine. The tail stays flat from 1 pattern to 8,000. | Deterministic pattern matching rather than probabilistic inference. |
| In what it costs | The cost of coverage stops scaling with hardware. Your CPU fleet does not grow when your rule set does. | Cost decoupled from rule set size. One engine in place of a large CPU fleet. |
Common questions
Uncertainty sucks.
Stop estimating. Your rules, your payloads, your P99.