Request a briefing

Industrial vision AI · On-prem

Every part inspected. On site.

Vision AI for defects, safety and line health, running on your hardware.

Explore vision analytics

Defense · Air-gapped AI

Nothing crosses unseen.

Private AI and threat detection that never leaves your boundary.

Explore defense AI
Passive OT monitoringModbus · DNP3 · S7 parsingPer-asset behavioural baselinesAir-gapped model servingSigned offline updatesRole-based access & auditNo telemetry, no call-homeSIEM / syslog exportPassive OT monitoringModbus · DNP3 · S7 parsingPer-asset behavioural baselinesAir-gapped model servingSigned offline updatesRole-based access & auditNo telemetry, no call-homeSIEM / syslog export
■ Who we serve
Industrial manufacturers

Start with vision analytics on the line.

  • Defect detection
  • PPE and safety compliance
  • Line monitoring
  • Edge hardware, no cloud egress
Vision analytics →Defense & critical infrastructure

AI that stays inside the boundary.

  • Air-gapped model serving
  • OT threat detection
  • Signed offline updates
  • Role-based access and audit
Defense AI →
01 The problem

Most security AI assumes a cloud connection. The systems that matter most don’t have one.

Plants, substations, and classified networks are isolated on purpose. That isolation also locks out the cloud-hosted tools most security vendors ship. Teams are left choosing between capability and control.

Ironmear is built for the isolated side of that line.

A

Legacy protocols

Modbus, DNP3 and S7 were designed without authentication. Malicious commands look like normal ones.

B

No outbound path

Air-gapped and segmented networks cannot send telemetry to a vendor’s cloud for analysis.

C

Sensitive data

Operational and mission data can’t leave the building to be processed by a third-party model.

02 First product · Vision analytics

Vision AI for the factory floor, on your hardware.

Cameras you already own feed an edge node on site. Machine vision models handle quality inspection, flagging surface defects, missing PPE and line stoppages in real time. Video and results stay in the plant.

  • Use casesDefects · PPE · line monitoring
  • DeploymentOn-prem edge node
  • Data egressNone by default
  • StatusNow accepting pilot partners
Vision Analytics details →
Illustrative: an edge camera inspects parts on a conveyor; one part is flagged as a defect.CAM-01DEFECT · SCRATCH 0.97ILLUSTRATIVE · NOT REAL DATAEDGE NODE · ON-PREM
Fig. 1 — Edge inspection, illustrative
03 Platform · OT detection

Anomaly and threat detection for industrial control systems.

A passive sensor learns how your process normally behaves, at the protocol and process-value level, and flags commands and readings that fall outside it. It never writes to the control network.

  • InputSPAN / TAP mirror
  • ProtocolsModbus, DNP3, S7, OPC UA (planned)
  • OutputAlerts to your SIEM / syslog
OT detection details →
Illustrative chart: observed process signal against learned baseline band, with one flagged deviation.DEVIATION · MODBUS FC16write outside baselineILLUSTRATIVE DATA · NOT A REAL INCIDENTt →
Fig. 2 — Baseline band vs. observed value
04 Platform · Enclave AI

Private AI for defense teams, sealed inside your boundary.

Language models, search, and analysis tools that run on your own hardware in disconnected environments. Models, documents, and logs stay on-site. Updates arrive on signed offline media that you verify before installing.

  • NetworkFully disconnected
  • HardwareCustomer-owned GPU servers
  • AccessRole-based, full audit trail
Enclave AI details →
Enclave: model weights, retrieval index and documents in one sealed environment, updates delivered by signed offline media.CLASSIFIED ENCLAVE · AIR-GAPPEDMODEL WEIGHTSpinned, hashedRETRIEVAL INDEXyour documentsINFERENCE · ACCESS CONTROL · AUDITANALYST UIAPI (local)SIGNEDUPDATE MEDIAoffline
Fig. 3 — Enclave components
05 How it works

Everything runs on your side of the line.

Ironmear is deployed as a node inside your network. Sensors observe; models infer; people decide. Nothing calls home.

Deployment architecture: sensors and models run inside the customer boundary; nothing crosses to the internet.CUSTOMER BOUNDARYPLC / RTULevel 0–1HMI / SCADALevel 2HISTORIANLevel 3PASSIVE TAPSPAN / mirrorread-onlyIRONMEAR NODEmodelsbaselinedetectionsaudit logon-premINTERNETnot required
Fig. 3 — Reference deployment
01 Observe

Passive collection from network mirrors and local data sources. No agents on controllers.

02 Model

Behavioural baselines and models trained and stored on the Ironmear node.

03 Act

Findings go to your operators and existing tools. Humans make every control decision.

06 Security & compliance

Built to be audited.

We publish what we hold and what we don’t. No certifications are held yet. We will update this table only when an audit is complete.

Full security posture →
FrameworkStatus
SOC 2 Type IIPlanned
ISO/IEC 27001Planned
IEC 62443Planned
Air-gapped deploymentSupported by design

Talk to an engineer, not a sales sequence.

Tell us about your environment. We’ll tell you plainly whether Ironmear fits.

Request a briefing