Three layers on one engine: Filtration cleans your channels, Analytics explains and models them, Control closes the loop — supervisory, from the cloud. Prepaid credits, caps you set, no seats.
Why one platform
nlsys is a self-learning signal processing engine. Filtration cleans your channels in three modes — real-time, small-lag and offline-grade — with one-step prediction and gap recovery. Analytics explains the clean signal: anomaly events with causes, a hidden-connection graph, regime changes, per-channel health — and compiles trajectories into verified models with a Nonlinearity Passport. Control turns signal and model into device control: no-lag per-channel outputs, NMPC on the compiled model, MIMO groups; we compute setpoints, your PLC executes.
Denoise, one-step prediction and gap recovery in three latency modes; streaming API of your channels, cleaned. Nothing to train, no filters to tune.
Explained anomaly events, hidden-connection graph and root-cause, regime changes, health — plus model compiles, free-run verification, Passports, artifacts and twin supervision.
Everything in Filtration, plus device control: no-lag per-channel, latency-agnostic NMPC, MIMO-correlated groups — mixed freely; your PLC executes.
The products
Start at any layer. They share accounts, credits and one Nonlinearity Passport — from cleaning a noisy channel to a supervisory control loop.
Clean signal in three modes — real-time, small-lag, offline-grade: denoise, one-step prediction, gap recovery. Nothing to train. $29/mo.
Explore Filtration →Anomaly events with explanations, hidden-connection graph & root-cause, health — plus model compiles and the Nonlinearity Passport. $99/mo.
Explore Analytics →Everything in Filtration, plus device control: no-lag per-channel outputs, NMPC on your compiled model, MIMO groups. We compute setpoints; your PLC executes. $99/mo.
Explore Control →How the layers fit together
Enter through any product on its own. Used together, they close a loop that none of them closes alone.
Channels get cleaned in three modes — real-time, small-lag, offline-grade — with one-step prediction and gap recovery. Nothing to train.
On clean signal: explained anomaly events, a hidden-connection graph with root-cause, regime changes and health. Clean, segmented sessions compile into verified models with a Nonlinearity Passport (compiler, twin supervision).
The clean signal and the compiled model drive device control: no-lag per-channel outputs, latency-agnostic NMPC, MIMO groups. We compute setpoints; your PLC executes.
The artifact
Every compiled model ships as a verifiable artifact: the equations it recovered, the region of data it is valid in, free-run test results against held-out trajectories, and hashes — so a model can be audited, handed off and kept honest in operation.
Start
Five channels are free — real signals, honest testing, no card. Tell us what you run and we'll set you up.