An AI platform that doesn’t just monitor industrial production environments — it remembers them. Four specialized reasoners on a biologically inspired memory substrate.
A public demo instance is running — no registration, no request. The credentials below are deliberately shared so anyone can look around.
E-mail
chef@foreman.de
Password
ForemanDemo2026!
The login is the plant-manager profile: it sees the whole fleet and may trigger the reasoners, so every capability is reachable from one account.
The interface is German. FOREMAN speaks the language of the shop floor it was designed for.
The data is simulated. No real plant is connected. Every prediction carries a visible simulation caveat — a deliberate design rule, not a placeholder.
It is a shared instance. Anything you enter is visible to everyone else and stays in the database. Please do not enter real personal data. Worker notes are run through name redaction before they are stored — a safety net, not a guarantee.
AI analyses are rate-limited. The reasoners call a language model on a capped budget. If an analysis is temporarily unavailable, alarms, trends and the archive keep working.
Nothing here switches anything. FOREMAN explains, it does not actuate. There is no path from this interface to a machine.
Written in May 2026, before the first commit: the problem, the three-layer architecture, its reasoning layer, and the plan. It shows the target picture, not today's build state — for that, see the table under Reasoners and the claims register. Interactive; use the arrows to navigate. Open in full screen ↗
Industry delivers the data, FOREMAN reasons, operators act. Each layer is autonomous — the platform is an open node in the industrial software stack, not a closed app.
Sensor streams, PLC logic (OPC UA, MQTT, Modbus), log files, and maintenance history — integrated through normalizing, pseudonymizing adapters. Today only the simulation adapter is built; the protocol adapters are the target picture.
Four specialized reasoners work against a shared memory substrate, routed through a local + cloud model gateway.
An operator dashboard with traceable explanations — plus an MCP interface that any third-party system can consume.
Each reasoner answers a question that a snapshot dashboard cannot.
Links incidents, sensor drift, and operator logs into coherent narratives over time.
Flags changes in a machine’s behavior against its own historical profile.
Matches the current state against consolidated precursor patterns from past incidents.
Causal deviation analysis between maintenance events and machine behavior to optimize intervals.
Load data, not load simulation. FOREMAN does not simulate load itself — that needs parameters outside its observation boundary. It exposes the observed load profiles and limits read-only over MCP, for an external simulation tool to build on.
This page describes the target picture of a project under construction. The build state lives in the repository — every row below cites its entry in the claims register.
| Target | Today | Evidence |
|---|---|---|
| Four reasoners | Three built — event-chain reconstruction, drift detection, failure prediction. Maintenance-cycle analysis waits on a real maintenance history. | C-001 |
| Ten dashboard sections | Eight built, including the platform/audit view and a 3D line view. Maintenance and load are marked placeholders. | C-003 |
| Plant protocols (OPC UA, MQTT, Modbus) | None connected. A simulation adapter drives a twin park of twelve machines. | C-005 |
| Open MCP node | Built — eleven read-only tools. The consumers named in the deck (simulation, ERP, energy management) are the target, not connected systems. | C-002 |
| Failure prediction on real data | A method demonstrator trained on simulation data; every prediction carries a simulation caveat. | C-012 |
| Memory substrate | Connected as an external service; a memory-only search mode is live in the demo. | C-033, C-073 |
Every change passes defined gates before it reaches main. The binding definition lives in the repository’s GROUND_TRUTH §10–§11.