Multi-site deployment, custom alloy, geometry and tolerance models, a managed factory-edge fleet and SLAs. Enterprise agreements land between $650k and $8.5M ACV and expand with every cell, module and site.
Built with foundries and die casters melting iron, steel, aluminium, magnesium and copper
[PLACEHOLDER] Design-partner names are illustrative until first references are signed.
A foundry group's hardest problem is not any single plant — it is that every plant solves the same problem differently, and none of the learning transfers. Smelteon makes the loop, the evidence and the models common while keeping each site's IP its own.
A proven wedge at one plant becomes a repeatable deployment pattern at the next: same runtime, same connectors, same autonomy gates, same evidence pack.
Alloy, geometry and tolerance models trained on your programs, not a generic average — with federated priors shared across your sites and never outside them.
Managed Jetson and factory-server fleet with model versioning, staged promotion, offline resilience and NVIDIA AI Enterprise supportability where required.
Response and uptime commitments for the edge runtime, escalation paths that understand shift patterns, and named engineering contacts.
Quality and safety assurance evidence aligned with IATF 16949, ISO 9001, AS9100 and NADCAP audit expectations, generated as a by-product of running.
Multi-year agreements with usage ramps, annual prepay discounts of 15–20%, and outcome components on the highest-value workflows.
A single melt, mold or pour cell on the Cell tier at $15,000 per month, with a baselined scrap, yield, porosity or melt-energy figure and a defined success metric agreed before the pilot starts.
Jetson inference at each cell beside PLC and robot controllers; Triton, Holoscan and RAPIDS ETL on factory servers. The loop survives a cloud outage.
Fully on-premises deployment including training on private DGX for producers whose alloy recipes and programs cannot leave the boundary.
Edge control with cloud training and validation, federated updates and tenant-isolated storage — the default for most groups.
| Cell — $15k/mo | Factory — $90k/mo | Enterprise — custom | |
|---|---|---|---|
| Scope | One melt, mold or pour cell or line | Whole plant, full loop | Multi-site group |
| Agents | One control or inspection agent | All nine agents | All nine plus custom |
| Castwin twin | Read-only plan consumption | Full simulation and promotion | Group-wide, custom geometry families |
| Models | Shared priors | Plant-tuned | Custom alloy, geometry and tolerance models |
| Edge | Single cell node | Plant fleet | Managed multi-site fleet |
| Evidence | Audit log | Audit log plus genealogy | Group assurance packages |
| Support | Standard | Priority | SLAs and named contacts |
| Commercials | Monthly or annual | Annual, usage components | Multi-year, ramps, outcome components |
The buying committee is a plant director as economic buyer, a metallurgist and quality engineer as champions, and IT/OT, quality and safety as approvers. Every one of them gets an answer before a write-back happens.
Network segmentation, OT-safe connectors, SSO and RBAC, no inbound dependency for cell control, and a documented failure mode for every integration.
Casting genealogy, non-conformance handling, grounded and cited outputs, and an assurance-grade audit trail that stands up in an IATF 16949 or AS9100 audit.
Bounded envelopes for every robot and machine action, safety review before any write-back, and a rollback path that is armed for the life of the deployment.
We walk the line with your metallurgist and operations lead, and pick the wedge cell with the clearest, largest, most measurable pain.
Instrument the cell in shadow mode and measure the real scrap, rework, melt-loss, yield and inspection numbers over a representative period.
Compare agent predictions against your own inspection ground truth on a golden set for that alloy and geometry family.
The agent recommends, your engineer approves, and the ROI case is built from real approved actions rather than a projection.
Process windows, rollback paths, escalation rules and robot envelopes reviewed and signed by quality and EHS.
Low-risk cells graduate to bounded write-back. Exceptions escalate. Everything is logged, attributable and reversible.
Adjacent cells, adjacent modules, adjacent plants — using the evidence pack the first cell produced.
$90k
Blended enterprise CAC target
$3.6M
Target LTV at 135% NRR
< 14 mo
Enterprise payback target
15–20%
Annual prepay discount
Yes. Factory-edge inference is on-premises by design, and full on-premises training on private DGX is available for producers with strict IP requirements. NVIDIA AI Enterprise is supported for supportability and controlled model lifecycle management.
Enterprise agreements are custom, landing between $650k and $8.5M ACV, typically multi-year with usage ramps and outcome components on the highest-value workflows such as yield, scrap, porosity and first-pass yield.
An immutable, assurance-grade quality and safety log of every agent action, approval, override and rollback, tied to casting genealogy — designed to serve as IATF 16949, ISO 9001, AS9100 and NADCAP evidence rather than to require separate paperwork.
Start with the plant that has the worst scrap number and the best data. The rest of the group follows the evidence.