Smelteon exists because the craft that decides whether a casting is sound — the melter's, molder's and pourer's judgement — is retiring faster than it is being taught. We are building the system that learns it, holds it, and runs it.
Smelteon is a Physical AI company building factory-edge agents that run melting, molding, coremaking, pouring, solidification, inspection, fettling and casting logistics — sensing porosity, inclusions, cold shuts, misruns and hot tears in real time and closing the loop from perception to action.
Melting, molding and pouring are hot, hazardous and craft-bound. The people who hold the judgement are retiring, and the pipeline behind them is thin. Plants are not short of ambition — they are short of pourers.
Grounding, citations, audit, human-in-the-loop and graduated control. An agent earns the right to move a setpoint; it is never granted by a contract.
Every supervised correction trains the system. The value of the platform to the tenth customer is built by the first nine — without their recipes ever leaving their boundary.
One wedge cell with a measurable number beats a platform pitch. Expansion follows evidence, not enthusiasm.
The metallurgist authors the window, approves the action and corrects the model. We automate the work, not the accountability.
Incumbents sell a furnace, or a molding machine, or a pouring cell, or an X-ray system, or an MES. Each is excellent and each is siloed. None closes the loop from perception to melt, mold and pour action — and none can build the cross-plant casting dataset that predicts soundness before the pour.
0–6 months
6–12 months
1–3 years
3–10 years
$16B
TAM — global foundry autonomy
$4B
SAM — reachable segments
$240M
SOM — 3–5 year target
9
Agents in the loop
[ASPIRATIONAL] Market sizing from Smelteon's market analysis; SOM is a 3–5 year serviceable obtainable target, not a forecast.
A cross-plant dataset of porosity, inclusion, cold-shut, misrun and hot-tear outcomes tied to melt chemistry, sand properties, pour behaviour and solidification curves — generalising across alloys, geometries and sites.
Rare defect patterns and new geometries learned at one plant transfer as privacy-preserving priors to every maker running similar castings and alloys.
Foundry-machinery OEM, casting-maker and MES-vendor channels — melt, molding, pouring and inspection vendors shipping Smelteon-ready autonomy.
Factory-edge perception plus autonomous melt, pour and solidification control validated in an Omniverse foundry-line twin that hits soundness before the pour.
Displacement means re-qualifying an alternative through a PPAP and first-article-bound process, and re-hiring scarce pourer talent that the system has already encoded.
More customers → more supervised outcomes → better models → more autonomy → more workflow captured → deeper integration and more proprietary data.
“Simulation told the plant what should happen. Nothing told it what did happen, heat by heat, mold by mold, pour by pour. That gap is the company.”
Pre-revenue, recruiting three to five paid design partners around one measurable wedge workflow: robotic pour and fill control, X-ray and CT porosity detection, or melt-chemistry optimisation. Forward-looking financial and market figures on this site are marked as aspirational.
No. Simulation is one component — the Castwin twin — and it exists to serve the control loop by testing variants before metal is committed. The company is the closed loop: perceive, plan, act, verify, log, retrain.
Because the pain is enormous and measurable, the craft is retiring, the process is physically observable, and nobody owns the closed loop. Casting is one of the last large manufacturing processes still run substantially by feel.
The fastest way to understand Smelteon is to walk your worst cell with us and look at the numbers together.