Book a plant walkSee pricing

The operations layerunder the wholefoundry.

One factory-edge platform runs melt, mold and core, pour and fill, solidification, inspection, finishing, handling and takt — orchestrated by agents that act inside process windows a metallurgist approved.

Book a plant walkRead the docs
9Agents
6Products
< 60sLine reschedule target

Built with foundries and die casters melting iron, steel, aluminium, magnesium and copper

Ironhaus GroupMeridian CastingsVulcan DuctileAurora Light MetalsKestrel Die CastingNorthfield Foundry Co.

[PLACEHOLDER] Design-partner names are illustrative until first references are signed.

Edge where it must be.Cloud where it helps.

Each cell runs Jetson-based inference beside the PLC and robot controllers. Factory servers run Triton, Holoscan, RAPIDS ETL and the model router. Cloud or private DGX trains and validates. Omniverse mirrors the plant so every recipe is tested before any write-back.

  • Cell edge — Jetson Orin/Thor inference beside PLC and robot controllers
  • Factory server — Triton, TensorRT, Holoscan, model router, historian ETL
  • Cloud / private DGX — training, validation, federated fleet updates
  • Twin — Omniverse foundry line, pouring and solidification simulation
  • The loop keeps running locally when the cloud link fails

Melt. Mold. Pour.Solidify. Inspect. Finish.

Every station is an agent with its own perception, its own process window and its own write-back gate — and every station's outcome is the label that trains the one before it.

MeltMoldPourSolidifyInspectFinish
01Meltchemistry + temp
02Moldsand + cores
03Pourrobotic fill
04Solidifycool + shakeout
05InspectX-ray · CT · vision
06Finishfettle + grind

What the platformactually controls.

Not a copilot over a dashboard. A system of action with bounded write-back into the machines that make the casting.

Adaptive melt control

Furnace melting, alloy chemistry, temperature, degassing, inoculation and nodularization held on target across scrap-charge variation and alloy changes.

Learn more

Sand, mold and core control

Sand properties, compaction, core strength and mold quality controlled early, so geometry and surface soundness are not discovered late.

Learn more

Robotic pour and fill

Ladle attitude, stream, pour temperature, timing and fill controlled per mold and per geometry — the craft step, made repeatable.

Learn more

Solidification and cooling

Cooling curves and shakeout timing controlled in the window where shrinkage porosity and hot tears are actually decided.

Learn more

Defect sensing and prediction

X-ray, CT and vision fusion that finds pores, inclusions, cold shuts and misruns — and predicts them from upstream process signals.

Learn more

Yield, scrap and takt

Yield, melt loss, scrap and line balancing optimized across melt, mold and pour with cuOpt-backed rescheduling when the line moves.

Learn more

Autonomy is earned,never assumed.

01Watch, predict, prove nothing changes

The agent observes the cell and predicts outcomes while the plant runs exactly as before. Baseline scrap, yield, porosity and melt energy are measured against the agent's calls.

  • Zero write-back
  • Baseline KPI capture
  • Golden-set accuracy review
  • Shadow
  • Assist
  • Bounded

Bounded by default,in three lines.

The API refuses to be reckless: a plan carries its process window, its confidence and its explanation, and applying it outside the window is not an option the SDK exposes.

  • Python and REST surfaces, OPC UA and MQTT at the edge
  • Every plan carries window, confidence and explanation
  • Write-back requires an approver identity and lands in the audit log
pour_control.py
# Ask the pour agent for a bounded setpoint move on cell P-04
from smelteon import Plant

plant = Plant("plant-04", edge="foundry-edge-02")
heat  = plant.heats.current(line="ductile-iron-1")

plan = plant.agents.pour.plan(
    heat        = heat.id,
    part        = "crankshaft-4cyl-rev-c",
    mold_state  = plant.agents.mold.state(),
    objective   = "minimise shrinkage porosity at journal 3",
    twin        = "castwin:solidification-v7",
)

# Every move is checked against the approved process window first
if plan.within_window and plan.confidence > 0.94:
    plant.apply(plan, mode="bounded", approver="metallurgist-on-shift")
else:
    plant.escalate(plan, reason=plan.explain())

It has to speakto what you already run.

Connectors to melt and furnace, molding and coremaking, pouring and die casting, inspection, robots and foundry MES — read first, then bounded write-back once the safety review passes.

Induction & holding furnaces

Setpoint read/write, power and temperature profiles, charge and alloy addition records.

Spectrometry & thermal analysis

Heat chemistry, carbon equivalent, thermal-analysis cups and nodularity checks streamed per heat.

Molding & coremaking lines

Sand plant data, compaction, mold quality, core cells and shot parameters.

78%

Target gross margin at scale

135%

Target net revenue retention

12–40

Models served per factory

8–24

Camera or detector streams per line

[ASPIRATIONAL] Scale targets from Smelteon's GPU utilization plan; margin and NRR are business-model targets.

Before youput it on a line.

No. Smelteon is the operations layer above them. We integrate with induction and holding furnaces, molding and coremaking lines, auto-pour and die-casting cells, inspection systems and your foundry MES, and we control them inside windows you approve.

The loop is local. Cell inference runs on Jetson beside the PLC and robot controllers, and the factory server hosts the model router — so perception and bounded control keep running when the cloud link fails.

Typically after shadow-mode baseline, a golden-set accuracy review and a safety review — not on day one. Most design partners spend the first phase proving prediction accuracy against their own inspection results.

See it onyour line.

We start with a plant walk, pick one wedge cell, and baseline what scrap, rework and melt loss cost you there today.

Book a plant walkCompare plans