Questions plant teams actually ask

Treadum is an autonomous operations layer for tire and rubber plants. Nine specialist agents perceive what is happening at mixing, extrusion, calendering, building, curing and inspection, decide what should change, and act inside signed safety envelopes — with every decision recorded as audit-grade evidence.

Neither. Analytics tells you what happened. Treadum changes what the line does: setpoints, holds, cure dwell, mold allocation and material routing, inside limits your engineers signed.

No. Treadum runs on top of existing presses, builders, mixers and inspection machines, integrating through PLCs, historians, MES and machine adapters. Additional sensing is sometimes recommended, never mandatory to start.

Inference runs on edge nodes on your plant network, close to the stations. Training, simulation and fleet management run in your tenant, cloud or on-premise. The plant keeps running if the WAN link drops.

Jetson-class edge nodes per cell for inference, and GPU capacity for training and simulation. We size it during the plant assessment and validate it before deployment.

Through OPC UA, MQTT Sparkplug, Modbus and vendor-specific adapters for Siemens and Rockwell, plus certified adapters for common inspection machines. Writes always route through the envelope service.

Perception-to-decision budgets are set per station during deployment and enforced in the runtime. If a budget cannot be met, that action is not enabled — it degrades to advisory rather than acting late.

Every writable action is bounded by a control envelope: absolute limits, rate limits, interlocks and risk classes, signed by your process and safety engineers and versioned like code. The envelope service rejects out-of-bounds writes regardless of what any model asks for.

Drift monitors and gate metrics run continuously. Breaching a gate demotes the agent to advisory automatically and raises an alert. Rollback to the previous signed model bundle is a single operation with in-flight writes drained safely.

Always, and overrides are treated as the most valuable data we collect. Each one is captured with context and reviewed weekly with your process team.

Edge nodes operate autonomously with locally cached models and envelopes. Cloud connectivity is used for training, fleet management and long-term storage, never for the real-time control path.

Three tiers: Cell from $15,000 per month for a single cell or process, Factory from $95,000 per month for a full plant loop, and Enterprise for multi-site groups with annual contract values from $700,000 to $9,000,000 depending on footprint.

Yes. Paid pilots are the only kind we run. Free pilots have no internal owner and no urgency; paid ones get baselined properly and reviewed weekly.

Shadow mode produces measurable findings in weeks. Bounded autonomy on a first wedge typically follows within a quarter, subject to gate performance. [ASPIRATIONAL] pending design-partner validation.

The success criteria are agreed in writing before we start. If they are missed, we say so, and you do not expand. We would rather lose a deal than defend a number that did not move.

You own your plant data and any models trained exclusively on it. Export is available at any time in open formats, including genealogy graphs and evidence logs.

Not across tenants, under any circumstances. Cross-customer learning is limited to explicitly non-proprietary, aggregated signals and is opt-in.

Yes. Full air-gapped deployment is supported, with signed model bundles delivered through a controlled channel.

An immutable, timestamped log of every decision, input, envelope version, model version and human action, exportable in formats mapped to IATF 16949, DOT and UNECE requirements.