This page describes how different sources reach the system. The concrete connection to machine controllers is under Process AI; what happens next is under processing.
One Interface Layer, Many Sources
The Neuronal Core needs to learn from very different kinds of sources: physical sensors on machinery, digital data streams from existing systems, and organizational information systems. The sensory integration layer abstracts away these differences, so core processing — input normalization, context retrieval, inference — consumes normalized inputs regardless of where they originally came from.
Standardized Driver Model
New sensor types are integrated via a standardized driver model without modifying core processing components. That decouples "what hardware exists at this site" from "how the Neuronal Core processes inputs" — a real advantage for deployments with a heterogeneous, organically grown sensor landscape.
Parallel, Mutually Isolated Detection Skills
In practice, the value of this architecture shows up in anomaly detection: Henri's detection system is structured as multiple independent skills, each targeting a distinct signal class. Skills operate in parallel and share no state — a false positive in one signal class cannot contaminate classification in another. Detection runs continuously, not on a polling schedule.