Context

This page describes the architecture. What it means for your plant is under Process AI; how you operate it is under the platform; and a pilot is the bounded first step to test it on your own data.

Why This Architecture Exists

Conventional software, conventional AI, and today’s industrial AI systems all fail at the same point: none of them holds a continuous state of the real operating environment. Software is rule-based and knows only what someone wrote down as a rule beforehand. A language model works one request at a time and starts from nothing after each. And most industrial AI systems are point solutions per use case, cloud-bound, with no shared basis of state.

For an operation that means: every question needs its own tool, every tool its own data path, and none of them knows anything about the others. That is the gap this architecture addresses.

What Neuronal Autonomous Architecture Is

A system architecture in which perception, context, states, relationships and assessment form one continuous layer — instead of separate tools per use case. Henri is the autonomous neuronal core inside that architecture: the instance where the layer comes together.

The Central Idea

Artificial intelligence is not the architecture. It can be a tool inside it. AI, ML and language models can be used within the architecture wherever they help. The architecture itself does not depend on any particular model existing or continuing to be maintained.

Why That Is a Structural Difference

The difference is not better models. It is that system state persists over time and is shared — and that the architecture survives a change of model. Anyone building on a model vendor today is building on someone else’s ground: an industrial line runs fifteen to twenty years, a model generation twelve to eighteen months.

Three Layers Compared

Conventional SoftwareBasis: rules · State: none · Context: hard-coded · Depends on: the developer
Artificial IntelligenceBasis: model · State: per request · Context: in the prompt · Depends on: the model vendor
Neuronal Autonomous ArchitectureBasis: architecture · State: continuous · Context: part of the system · Depends on: no external party

What Is Demonstrable Today

The points below are implemented and verified. They are the reason the system can run inside an operation today.

Runs with no outbound connectionExisting — no external calls, no dependency on SynapSync servers, no telemetry. Air-gap is the delivery state, not a special configuration
Model weights on your hardwareExisting — transferred at deployment, under your control exclusively from then on
No remote accessExisting — SynapSync has no access to deployed systems unless you explicitly provision it for support
No expiry dateExisting — perpetual license, and by construction no way to revoke it remotely
Offline updatesExisting — delivered as a package, reviewed by you, applied on your schedule
Two delivery formsExisting — pre-configured appliance or software image on your hardware; identical operating system

What the Architecture Is Designed For

The points below describe what the architecture structurally enables. They are deliberately not written as existing features — the exact status per building block is being established right now, and until then we say “designed for”, not “present”.

Continuous contextDesigned for — an ongoing context of the environment rather than isolated data points or single requests
States over timeDesigned for — past states and events can be carried forward as context and used again
Relationships across sourcesDesigned for — connections across several data sources, not isolated per machine
Assessment and responseDesigned for — taking in situations, assessing them, and responding within defined limits
Model independenceDesigned for — models are tools inside the architecture, not its foundation

Autonomy and Boundaries

Autonomous does not mean uncontrollable. The operator defines what the system may do on its own — observe, assess, propose, or act. Those limits are part of the configuration, not a setting anyone can forget.

Exactly what the system does at which level today is being established with our engineering team. Until that is settled, no figure and no degree appears here — not even as an implication.

Sovereignty

No operational data, no results, and no organizational context is transmitted outside your environment. That is not a setting but the construction: there is no mechanism through which we could see your operation, and no license check. More under the sovereignty framework and air-gapped configuration.

What Gets Built on the Architecture

The architecture is not tied to one field of use. Process AI is the first application on it — it works at the machines and is today the only one running in production. Further fields of use follow the same pattern: the same architecture, a different set of applications. Which fields those are and how far along they are is in the roadmap — as a roadmap, not as a feature.

Henri Neuronal Core works above all of it: context, relationships and states across every data source, not just across one machine. Core and application can each be deployed on their own. In both cases you work in the platform — what you buy is Process AI and/or Henri Neuronal Core, what you operate is always the same surface.

The Architecture in Detail