Sovereign AI for Manufacturing Plants
Shop-floor and predictive-maintenance intelligence that runs offline on the operator's own hardware, with every action signed before it executes.
On a modern shop floor, the most valuable intelligence is not in the boardroom. It lives in the vibration signature of a spindle bearing, the thermal drift of an injection-moulding tool, the subtle harmonic that precedes a gearbox failure three shifts from now. Capturing that signal and acting on it is the whole game of predictive maintenance. The problem is that most manufacturers have been told they must ship this data to a distant cloud to make sense of it, and for a plant running proprietary tooling, safety-critical lines, or defence-adjacent work, that is a boundary they cannot cross.
We built Mickai for exactly this boundary. Mickai is a Sovereign Intelligence Operating System, a SIOS that runs on the operator's own hardware, air-gapped or on-premise, with zero data egress. The intelligence comes to the machine rather than the machine's secrets going to the intelligence. For manufacturing, that inversion changes what is possible on the line.
The forge that never leaves the floor
Predictive maintenance depends on continuity. A model that only fires when the plant has connectivity goes blind during exactly the outages, shift changes, and network partitions when equipment stress peaks. Mickai runs its sensing and inference subsystems locally, so the intelligence is always resident on the operator hardware, whether that is a ruggedised industrial PC beside a press line or a rack in a plant server room.
Because nothing is streamed off site, the plant keeps its process data, its tolerances, its recipe parameters, and its failure history entirely within its own walls. These are the crown jewels of any manufacturer, and they are the very things a public cloud pipeline would carry away. We keep the forge on the floor. The heat, the sparks, and the knowledge all stay where the work happens.
Reading the machine before it breaks
Our maintenance brains ingest the streams a plant already generates: accelerometer and acoustic data, motor current signatures, temperature and pressure telemetry, cycle counts, and controller logs. Revocable, domain-specific brains fuse these into a live picture of each asset's health, flagging the drift patterns that precede bearing wear, tool degradation, misalignment, cavitation, or lubrication loss.
The point is not a generic anomaly alarm. It is a specific, actionable forecast: this spindle has an estimated fourteen days of remaining useful life at current load, so order the part now and schedule the swap into the planned Sunday downtime. Because the brains run at the edge, the loop from sensor to insight to instruction closes in the plant, at line speed, without a round trip to anywhere.
Every action signed before it happens
In a regulated or safety-critical plant, an autonomous suggestion is only useful if it is accountable. This is where Mickai's core primitive earns its place. Before any action executes, whether that is raising a maintenance work order, adjusting a setpoint recommendation, or quarantining a suspect batch, the system writes an Operation Attestation Record, an OAR that captures what is about to happen, which brain proposed it, on what evidence, and under whose authority.
That record is signed with post-quantum cryptography, using FIPS 204 ML-DSA-65 signatures, and hash-linked into a tamper-evident ledger with SHA-3-512 chaining. The result is a cryptographically-signed audit trail that an inspector, an insurer, or a customer auditor can verify offline, with no need to trust the vendor and no way to alter history after the fact. For a plant answerable to ISO 42001, the NIST AI Risk Management Framework, or sector safety regulators, that verifiable provenance is the difference between an experiment and a system you can run in production.
High-stakes moves need more than one voice
Not every action carries the same weight. Logging a vibration trend is routine. Stopping a line, overriding a safety interlock recommendation, or authorising an emergency tool change is not. Mickai scales its guardrails to the stakes. Low-risk actions proceed on a single brain's attestation. High-consequence actions require multi-brain consensus plus voice-biometric approval from a named, authorised operator or engineer.
This means a critical decision on the floor is corroborated by more than one independent reasoning subsystem and confirmed by a human whose identity is cryptographically bound to the record. If a brain is ever found to be misbehaving or out of date, it is revocable on the spot, cut out of the decision path without tearing down the rest of the system. Autonomy and control stop being a trade-off.
Working with the cloud, not against the wall
None of this is a rejection of the public cloud. The hyperscalers, our allies at a different layer, are extraordinary at elastic, connected, non-sensitive compute. But a manufacturer running ITAR-controlled parts, or bound by NIS2 and DORA obligations on operational resilience, or protecting decades of proprietary process know-how, faces a boundary the public cloud is not designed to cross on the customer's own terms. Mickai serves that boundary. It sits alongside the existing estate as the sovereign layer where the sensitive work lives.
The capability behind all of this is protected by 104 filed UK patent applications, about 2,340 claims owned by Mickai LTD, covering the attestation, signing, and multi-brain approval mechanics that make edge autonomy trustworthy. We describe these by what they do on the floor, not as trophies. They exist so that a plant can run intelligent, self-directed maintenance and stand behind every decision it makes.
The bottom line
Manufacturing intelligence has been held hostage to a false choice: send your most sensitive data away to get smart maintenance, or stay dark and keep flying blind. Mickai dissolves that choice. The intelligence runs offline, on the operator's own hardware, reading the machines in real time, forecasting failures before they cost a shift, and signing every action into a ledger you can verify without trusting anyone. The forge stays on the floor, and so does the knowledge it produces.