Sepion
Autonomous separation and standardization that ends fat and protein give-away.
agent.intake_separate
Products
Buy the stage that hurts. Add the rest when it pays.
Five products, one plant-edge runtime, one audit and genealogy backbone, one data flywheel. Sepion sets the composition that Culvex and Aridyn depend on, Vizra inspects the output and feeds each defect back to its cause, and Lactwin simulates the coupled process before every run.
Illustrative scenario: Plant run · skim-milk-powder line + set-yogurt line · shared intake. Not a measured run.
The ecosystem
Five products, one loop
These are one coherent dairy-processing autonomy ecosystem, not separate bets. They share the plant-edge inference infrastructure, the food-safety audit and genealogy backbone, the multi-tenant data plane and one compounding flywheel.
Aridyn
Autonomous evaporation and spray-drying that hits moisture and solubility at minimum energy.
agent.evaporate_dry
Culvex
Autonomous fermentation and culture control for cheese and yogurt that hits pH and texture every batch.
agent.ferment_culture
Vizra
Inline vision and robotic handling that catches contamination, foreign bodies and fill and seal faults before they ship.
agent.defect_inspect
Lactwin
The as-processed dairy-line digital twin that hits target composition, safety and takt before the run.
agent.orchestrator
Composition wedge · agent.intake_separateSepion
Autonomous separation and standardization that ends fat and protein give-away.
The problem
Separation and standardization set the economic floor of every dairy plant: milk is clarified, separated into cream and skim and re-blended to an exact fat and protein target before anything else happens. Today this runs on fixed separator setpoints, periodic lab checks and technologist craft, so composition drifts between samples and processors over-standardize to stay safe, giving away fat and protein on millions of liters.
What Sepion does
Sepion fuses inline NIR and mid-IR analyzers, separator and clarifier telemetry, flow and temperature sensors and lab results into a continuous estimate of fat, protein and solids, then closes the loop on separator throughput, cream-skim split and the standardization blend valves to hold the target at the minimum safe margin. It predicts separator fouling and clarifier bacterial-load excursions before they force a stop.
Key features
- Real-time inline composition estimation fusing NIR, mid-IR and lab calibration
- Closed-loop separator throughput, cream-skim split and blend control at minimum safe margin
- Give-away analytics: liters and dollars of fat and protein recovered per shift, line and product
- Separator fouling and clarifier bacterial-load prognostics with lead time
- Audit trail of every setpoint change and composition outcome
AI capabilities
| Capability | Status |
|---|---|
| Spectral-to-composition regression per milk supply and product, edge-served | design scope |
| Closed-loop standardization policy: shadow, assist, then bounded | bounded stage planned |
| Separator fouling and clarifier load prognostics | design scope |
| Cited reasoning over separation and standardization specs | design scope |
NVIDIA stack
Jetson Orin/Thor at the separator with TensorRT and Triton; DeepStream and Holoscan for analyzer streams; cuOpt for silo and blend sequencing; NIM for packaged inference. DGX/HGX fine-tuning across the design-partner cohort is planned.
- For
- Fluid-milk, cheese, milk-powder and cultured-dairy processors, cooperatives and groups running centrifugal separation and inline standardization.
- Price
- $14,000 per processing line per month, expanding to the Plant tier as adjacent agents activate.
- First release
- Shadow-mode composition estimation on one standardization line from existing analyzers and lab calibration; a give-away dashboard; assist-mode blend recommendations with technologist approval; bounded auto-standardization only after the food-safety and ROI gates.
- In the loop
- Feeds accurate protein and fat into Culvex, accurate solids into Aridyn, composition context into Vizra and the as-processed composition into Lactwin.
Energy and drying wedge · agent.evaporate_dryAridyn
Autonomous evaporation and spray-drying that hits moisture and solubility at minimum energy.
The problem
Evaporation and spray-drying turn milk into powder and consume the most energy and water in the plant. Operators hold conservative inlet and outlet temperatures and feed rates to avoid stickiness, wall fouling, scorched particles and out-of-spec moisture, because a moisture or solubility miss downgrades a high-value batch and a fouling event forces a wash-out. Running conservatively wastes energy and throughput; running hot risks quality.
What Aridyn does
Aridyn fuses inline moisture and NIR, temperature and humidity, feed solids and pressure with a GPU-accelerated CFD and process twin of the evaporator and spray dryer to predict outlet moisture, solubility, particle size and fouling risk in real time, then closes the loop on inlet and outlet temperature, feed rate and solids, and atomization to hold powder spec at the minimum energy and water. It anticipates stickiness and fouling upsets and schedules drying around energy cost and CIP windows.
Key features
- Real-time outlet moisture, solubility and particle-size prediction from inline sensing plus the GPU twin
- Closed-loop inlet and outlet temperature, feed solids and atomization at minimum-energy setpoints
- Stickiness and wall-fouling prognostics with lead time
- Energy, water and CO₂ per kilogram of powder, and drying schedules around energy cost and CIP
- Audit trail of drying setpoints and moisture and solubility outcomes
AI capabilities
| Capability | Status |
|---|---|
| Physics-informed spray-drying and evaporation surrogates | design scope; full CFD-in-the-loop planned |
| Fouling and stickiness prognostics from telemetry | design scope |
| Energy-minimizing drying policy with graduated autonomy | bounded stage planned |
| Cited drying reasoning over specs and safety standards | design scope |
NVIDIA stack
DGX/HGX and OVX for the CFD twin and surrogate training (planned); Jetson Orin/Thor at the dryer; Omniverse and Modulus for the physics-ML twin; TensorRT and Triton for edge serving; cuOpt for dryer scheduling; Cosmos and Replicator for rare drying-upset scenarios.
- For
- Milk-powder, whey, infant-formula and dairy-ingredient processors running falling-film evaporators and spray dryers, and groups with energy and sustainability mandates.
- Price
- $14,000 per drying line per month, expanding to the Plant tier; outcome pricing on verified energy and water reduction and downgrade avoidance.
- First release
- A surrogate moisture and solubility predictor on one dryer from existing telemetry; an energy-per-kilogram dashboard; assist-mode setpoint guidance; then the GPU CFD twin for envelope discovery and bounded auto-drying after the energy, quality and safety gates.
- In the loop
- Consumes Sepion's solids and composition, shares Lactwin's twin backbone, gives Vizra powder-quality context and contributes drying dynamics to the energy benchmarks.
Biology wedge · agent.ferment_cultureCulvex
Autonomous fermentation and culture control for cheese and yogurt that hits pH and texture every batch.
The problem
Cheese and cultured-dairy quality is decided by living biology: cultures acidify milk to an exact pH on an exact curve, and the coagulation, cutting, cooking and texture windows all hinge on it. Culture activity swings with composition, season, phage pressure and inoculation variability, so master cheesemakers watch pH by hand and intervene by feel. The craft that manages this is retiring faster than it can be replaced.
What Culvex does
Culvex fuses inline pH and temperature, milk composition from Sepion, culture lot and vat telemetry to predict the full acidification and texture trajectory in real time, then closes the loop on temperature profile, culture dosing and timing, coagulant addition and cut and cook timing to land target pH, moisture and texture. It detects abnormal acidification and probable phage or contamination events early and encodes the cheesemaker's craft into a reusable per-product model.
Key features
- Real-time acidification and texture trajectory prediction per product and culture lot
- Closed-loop temperature profile, culture dosing and timing, coagulant and cut and cook windows
- Early abnormal-acidification and phage or contamination detection with corrective recommendations
- Culture-lot performance and craft-capture memory that improves batch to batch
- Audit trail of culture actions and pH and texture outcomes
AI capabilities
| Capability | Status |
|---|---|
| Biological curve models for acidification and texture | design scope |
| Anomaly detection for phage, contamination and abnormal fermentation | design scope |
| Closed-loop fermentation policy with graduated autonomy | bounded stage planned |
| Per-technologist craft memory, scoped per tenant | design scope |
NVIDIA stack
Jetson Orin at the vat for trajectory inference; TensorRT and Triton with per-product and per-culture routing; NeMo for dairy-culture reasoning; cuOpt for vat and cook sequencing; Modulus for biology-informed curve modeling (planned); DGX/HGX training (planned).
- For
- Cheese, yogurt and cultured-dairy processors, specialty and commodity makers, cooperatives and groups running batch and continuous fermentation.
- Price
- $14,000 per fermentation line per month, expanding to the Plant tier; outcome pricing on first-pass quality and downgrade reduction for premium cheeses.
- First release
- An acidification-trajectory predictor on one line from existing pH and composition feeds; abnormal-fermentation and phage alerts; assist-mode dosing and temperature guidance; craft-capture memory; bounded control after the quality and safety gates.
- In the loop
- Consumes Sepion composition, shares Vizra inspection context and Lactwin's twin (fermentation is a twin domain), and feeds culture-response data into the shared flywheel.
Perception and robotics wedge · agent.defect_inspectVizra
Inline vision and robotic handling that catches contamination, foreign bodies and fill and seal faults before they ship.
The problem
Inspection is the last line of dairy food-safety defense: contamination, foreign bodies, fill-level errors, cap and seal defects, label faults and package damage have to be caught before product reaches consumers. Most plants rely on spot lab tests, manual visual checks and single-purpose detectors that miss subtle or rare defects, while packaging and palletizing are paced by a scarce workforce, and inspection data is disconnected from the process that caused the defect.
What Vizra does
Vizra runs multi-camera vision and inline sensing at the plant edge to detect and predict contamination, foreign bodies, fill, seal, cap and label faults at line speed, rejects off-spec units and feeds each defect back to the upstream agent that caused it. It also drives Isaac and Jetson robotic packaging, palletizing and material handling, and covers events too rare to collect with validated synthetic defects.
Key features
- Multi-camera and inline vision for contamination, foreign-body, fill, seal, cap and label defects at line speed
- Defect-to-cause feedback to Sepion, Culvex and Aridyn
- Isaac and Jetson robotic packaging, palletizing and material handling
- Synthetic rare-defect generation with Cosmos and Replicator, validated against real data
- Immutable food-safety audit log and batch genealogy for every reject and action
AI capabilities
| Capability | Status |
|---|---|
| Fine-tuned contamination, foreign-body, fill, seal and label vision at the edge | design scope |
| Defect prediction and root-cause attribution to upstream state | attribution in design scope; full causal closure planned |
| Robotic perception, grasp and motion policies for packaging | design scope via Isaac |
| Cited food-safety reasoning grounded in HACCP, FSMA and ISO 22000 | design scope |
NVIDIA stack
Jetson Orin/Thor for 8 to 24 streams per line; Metropolis, DeepStream and TensorRT pipelines; Holoscan for deterministic fusion; Isaac Sim, ROS and Lab for robotics; Cosmos and Replicator for synthetic defects; Triton and NIM for serving; IGX and Fleet Command for the managed edge fleet.
- For
- Every dairy processor with filling, sealing and packaging lines and recall exposure: fluid, cheese, butter, yogurt, cream and powder.
- Price
- $14,000 per inspection line per month, expanding to the Plant tier; robotics add-ons and outcome pricing on recall risk and scrap reduction.
- First release
- Multi-camera contamination, fill and seal inspection on one line with reject control and the food-safety audit log; defect-to-cause attribution; synthetic rare-defect augmentation; Isaac and Jetson robotic handling after safety validation.
- In the loop
- The perception spine: its defect-to-cause feedback links to Sepion, Culvex and Aridyn, and its events feed Lactwin's synthetic rare-fault library.
Twin backbone · agent.orchestratorLactwin
The as-processed dairy-line digital twin that hits target composition, safety and takt before the run.
The problem
Plants commit to a run, a product, a milk supply, a set of setpoints and a schedule, and only discover composition misses, safety-margin problems, fouling or takt bottlenecks after milk and energy are spent. No tool simulates a batch as it will actually be processed across the coupled physics and biology of separation, heat exchange, fermentation, evaporation and spray-drying. Rare faults are also too rare and costly to collect enough real data to train robust models.
What Lactwin does
Lactwin couples GPU-accelerated CFD and process and biology simulation of separation, pasteurization and heat exchange, fermentation, evaporation and spray-drying with the live plant state to simulate a run before it happens, finding recipes and control envelopes that hit composition, safety and takt at minimum energy and water, and handing optimized setpoints to Sepion, Culvex and Aridyn. It also generates validated synthetic rare-fault scenarios that harden every Milkeon model.
Key features
- As-processed batch and line twin coupling CFD with process and biology simulation
- Pre-run recipe and control-envelope discovery at minimum energy and water
- Auto-optimized setpoints handed to the control agents
- Validated synthetic rare-fault generation for every Milkeon model
- Line-balancing, takt, CIP and energy scheduling simulation; what-ifs for new products and changeovers
AI capabilities
| Capability | Status |
|---|---|
| CFD and process surrogates and physics-ML across separation, heat, fermentation and drying | surrogates in design scope; full multiphysics-in-loop planned |
| Recipe and envelope search with multi-objective optimization | design scope |
| Synthetic rare-fault generation with domain randomization, validated against plant data | design scope |
| Twin-versus-actual reconciliation and drift detection | design scope |
NVIDIA stack
DGX/HGX and OVX for multiphysics twins and surrogate training (planned); Omniverse and Modulus for the physics-ML twin; Cosmos and Replicator for synthetic scenarios; cuOpt for scheduling; CUDA, TensorRT and Triton for kernels and surrogate serving; Jetson for distilled on-line envelope checks.
- For
- Multi-line and multi-site processors, cooperatives and groups running coupled separation, pasteurization, fermentation, drying and packaging; process-engineering and new-product teams.
- Price
- Included in the Plant tier at $90,000 per month and in Enterprise agreements; premium tiers for new-product what-ifs and rare-fault data.
- First release
- A single-domain surrogate twin (drying or separation) reconciled against real runs; pre-run recipe and envelope suggestions to one control agent; a validated synthetic rare-fault set; then the coupled whole-line twin and auto-optimization after validation.
- In the loop
- Optimizes Sepion, Culvex and Aridyn setpoints, consumes Vizra's defect events and feeds validated synthetic data and envelopes back to all of them.
Run timeline
The products in one run
Every Milkeon run is an ordered, inspectable sequence. This is run_scn_01, an illustrative scenario: Plant run · skim-milk-powder line + set-yogurt line · shared intake. Scenario complete · both streams inside spec · one human approval at the dryer.
run_scn_01
scenario · not measured
SUCCEEDED
-
01
intake.receive
SUCCEEDED
0.9 s
Plant Orchestrator · Pulled both product specs, the HACCP plan with its pasteurization critical control point, the antibiotic screen for the tanker lot and the standing CIP window from MES and the lab system; locked the target envelope for the run.
-
02
twin.simulate
SUCCEEDED
41 s
Plant Orchestrator · Simulated 24 candidate recipes across separation, pasteurization, evaporation and spray-drying on the Lactwin as-processed twin; ranked them on composition margin, pasteurization margin, energy per kilogram of powder and takt, and discarded every candidate that thinned the pasteurization margin.
-
03
separate.standardize
SUCCEEDED
3 m 05 s
Intake-and-Separate · Fused inline NIR and mid-IR with the latest lab calibration, then trimmed the cream-skim split and the standardization blend valve so skim solids and the yogurt-stream fat sat at target with the minimum safe margin rather than the usual over-standardization.
-
04
pasteurize.verify
SUCCEEDED
continuous
Pasteurize-and-Homogenize · Held HTST at or above 72 °C for 15 s through the holding tube with the flow-diversion valve armed; every critical control point reading was written to the audit log against the HACCP plan.
-
05
culture.vat
SUCCEEDED
4 h 20 m
Ferment-and-Culture · Predicted the acidification curve for the culture lot from inline pH and the upstream composition, held the vat temperature profile and timed the hold so pH landed on the set-yogurt target curve; no abnormal acidification was flagged.
-
06
dry.spray
SUCCEEDED
2 h 10 m
Evaporate-and-Dry · Ran the dryer against the twin's moisture and stickiness prediction; stepped inlet temperature and feed solids toward the minimum-energy setpoint while predicted outlet moisture stayed inside spec and fouling risk stayed low.
-
07
inspect.pack
SUCCEEDED
continuous
Defect-and-Inspect · 16 camera streams across the two packing lines at the plant edge: fill, seal, cap and label checks on every unit, foreign-body and contamination models on the powder stream; two rejects were attributed back to a seal-jaw temperature drift and the packing line was notified.
-
08
cip.schedule
SUCCEEDED
6 s
Hygiene-and-CIP · Placed the pasteurizer and dryer CIP into the cheapest energy window, sequenced wash phases against tank availability and set the release checks that must pass before the next product.
-
09
approve.human
APPROVAL
1 m 48 s
Plant Orchestrator · The twin's lowest-energy candidate wanted a dryer inlet temperature above the site's bounded envelope. Held for the shift dairy technologist, who approved the step with a tighter outlet-moisture alarm.
-
10
release.genealogy
SUCCEEDED
1 m 12 s
Quality-and-Conformance · Both batches released: composition, CCP records, culture lot and pH curve, drying setpoints, inspection rejects and every human decision written to the batch genealogy and checked against spec.
Compare
What is in each plan
Modules are licensed per line on the Line tier and bundled on Plant and Enterprise.
| Capability | Line | Plant | Enterprise |
|---|---|---|---|
| Lines covered | 1 | All on site | All in group |
| Agent modules | 1 of 9 | All 9 | All 9 + custom models |
| Lactwin twin | Not included | Included | Custom twin domains |
| Plant-edge node | Included | Per line | Fleet managed |
| Guarded PLC writes | Scoped tags | All scoped tags | Group policy |
| Batch genealogy | Included | Included | Included + export |
| Autonomy stages | Shadow, assist, bounded | Shadow, assist, bounded | Shadow, assist, bounded |
| Deployment | Plant edge + cloud | Plant edge + cloud | On-prem or private cloud |
| Outcome pricing | On request | Available | Standard |
| Indicative price | $14,000 / mo | $90,000 / mo | $650k – $8.5M ACV |
Tool calls
One log, every product
Every tool invocation, argument and result is written to an immutable, human-readable log, and every reasoning step is expandable. Nothing about a run is hidden from the plant. The log below is the scenario run, not a measured one.
- 05:10:02mes.read_order({ lot: "T-0417", products: ["SMP", "SET-YOG"] })
- 05:10:02ok specs locked · HACCP plan v7 · CCP: pasteurization 72 °C / 15 s
- 05:10:03lab.read_results({ lot: "T-0417", test: "antibiotic_screen" })
- 05:10:03ok negative · lot accepted for intake
- 05:10:04twin.simulate({ candidates: 24, horizon: "shift" })
- 05:10:45ok best recipe #9 · pasteurization margin held · lowest energy per kg
- 05:10:46policy.evaluate({ recipe: 9, stage: "bounded" })
- 05:10:46ok 7 writes permitted · 1 write requires human approval
- 05:10:47plc.write_setpoint({ tag: "SEP1.BLEND_VALVE", basis: "nir+midir+lab" })
- 05:13:52ok skim solids on target · yogurt-stream fat at minimum safe margin
- 05:13:53plc.verify_ccp({ tag: "PAST1.HOLD_TUBE", min: "72 °C / 15 s" })
- 05:13:53ok CCP verified · divert valve armed · readings → audit log
- 05:14:10dose.culture({ vat: "Y2", lot: "C-2291", profile: "set-yogurt" })
- 05:14:10ok acidification curve predicted · hold timed to target pH
- 05:14:12approval.request({ action: "dryer.inlet", reason: "above bounded envelope" })
- 05:16:00ok approved by the shift dairy technologist · tighter moisture alarm set
- 05:16:01plc.write_setpoint({ tag: "DRY1.INLET_T", mode: "min_energy" })
- 05:16:01ok predicted outlet moisture inside spec · fouling risk low
- 05:16:02vision.stream({ cameras: 16, models: ["fill_seal", "foreign_body"] })
- 05:41:18reject: seal fault · unit 1184 · attributed to seal-jaw temperature drift
- 05:41:19defect.attribute({ event: "seal_fault", cause: "pack_line.jaw_temp" })
- 05:41:19ok packing line notified · second reject at 05:43:02 · drift corrected
- 05:41:20cuopt.schedule({ assets: ["PAST1", "DRY1"], window: "energy_low" })
- 05:41:20ok CIP sequenced · release checks armed
- 09:36:40mes.release_batch({ batches: ["SMP-0417", "YOG-0417"], genealogy: "full" })
- 09:37:52ok both batches released · scenario complete
Reasoning trace
Goal, plan, thought, action, observation, exactly as the orchestrator would record it.
GoalReceive tanker lot T-0417, standardize it once, pasteurize onc…
Receive tanker lot T-0417, standardize it once, pasteurize once, and run a skim-milk-powder stream and a set-yogurt stream to spec with full genealogy.
PlanSimulate the whole recipe on the twin first. Lead with composi…
Simulate the whole recipe on the twin first. Lead with composition so the dryer gets the right solids and the vat gets the right fat. Verify the pasteurization critical control point continuously. Let the dryer chase minimum energy only inside the bounded envelope; hold anything beyond it for a human.
ThoughtThe standing recipe over-standardizes both streams to protect …
The standing recipe over-standardizes both streams to protect against lab-sample lag. With continuous NIR and mid-IR estimates the blend can sit at the minimum safe margin. The saving is composition give-away, not safety margin: the pasteurization CCP is never a candidate for optimization.
Actiontwin.simulate(candidates=24) → recipe #9: minimum safe composi…
twin.simulate(candidates=24) → recipe #9: minimum safe composition margin, pasteurization margin unchanged, lowest predicted energy per kilogram of powder.
ObservationRecipe #9's dryer inlet temperature sits above the site's boun…
Recipe #9's dryer inlet temperature sits above the site's bounded envelope. The policy engine routes that single write to the shift dairy technologist and permits the other seven.
ActionExecute recipe #9 at the bounded stage: separation and standar…
Execute recipe #9 at the bounded stage: separation and standardization writes, CCP verification, culture dosing and the vat profile proceed; the dryer inlet write waits.
ObservationApproval granted at 05:16:00 with a tighter outlet-moisture al…
Approval granted at 05:16:00 with a tighter outlet-moisture alarm. Two seal rejects on the yogurt packing line were attributed to seal-jaw temperature drift and corrected by the packing line.
OutcomeScenario complete. Both batches released with composition, CCP…
Scenario complete. Both batches released with composition, CCP records, culture lot, drying setpoints, inspection events and the human decision in the genealogy. No measured plant result is claimed; this is the run the platform is designed to execute.
Autonomy
Four stages, set per loop
A plant does not go from manual to bounded control in one step. Milkeon makes the stage explicit per loop, auditable and reversible at any time, and some decisions never leave the person.
| Stage | What the agent does | What the person does | When |
|---|---|---|---|
| Shadow | Predicts composition, contamination, moisture and takt; writes nothing | Runs the plant as today; compares predictions to lab results | First phase with every design partner |
| Assist | Recommends separation, pasteurization, culture and drying moves with the evidence | A dairy technologist or process engineer approves or corrects each one | After accuracy is proven against the baseline |
| Bounded | Writes low-risk separation, drying and inspection setpoints inside explicit limits | Approves anything above the limits; reviews the shift record | After the food-safety and ROI gates agreed up front |
| Exceptions | Escalates anything outside its envelope, CCP-adjacent moves and release decisions | Always decides these; the agent never does | Permanent |
Accelerated computing
Physical AI, at the plant edge
Milk is perishable and the lines are continuous, so perception has to be local and fast. Milkeon runs GPU inference at the plant edge and keeps training, simulation and optimization in the cloud, in a private cloud or on-prem. Items marked planned are on the roadmap, not running.
Edge perception
8 to 24 camera and inline-analyzer streams per line. Design target: 50 to 150 ms reject and alert decisions, 250 to 1,000 ms composition estimates, all at the plant edge.
Model serving
Defect, composition, shelf-life, process-state and dairy-reasoning models behind one orchestrator, with version routing by product, line and sensor package. Design target: 10 to 40 concurrent models per plant.
Training (planned)
Vision, spectral, time-series and dairy-reasoning models fine-tuned from labeled batch outcomes, operator corrections, lab results and validated deviations. Initial design target: 2k to 10k GPU-hours.
Digital twin (planned)
GPU-accelerated CFD and process simulation of separation, heat exchange, fermentation, evaporation and spray-drying, run as-processed before the batch to find safe, low-energy recipes.
Synthetic data (planned)
Rare contamination, foreign-body, fouling, seal, fill and drying-upset scenarios generated with domain randomization and always validated against real plant events before use.
Optimization
Product sequencing, tank allocation, dryer scheduling, CIP timing, energy and water constraints and line balancing, recomputed on the GPU when the plant is disrupted.
Robotic handling
Packaging, palletizing and material handling for product, packs and pallets, with perception and motion policies built in simulation before they touch the packing hall.
Design targets
Product design targets
The numbers Milkeon is engineered against, from the GPU utilization plan.
50–150 ms
p95 latency for reject and alert decisions at the plant edge
250–1,000 ms
p95 latency for inline composition estimates
8–24
camera and inline-analyzer streams per line
10–40
concurrent models served per plant, routed by product, line and sensors
2k–10k
GPU-hours of training across the first design-partner cohort
50k–250k
synthetic rare-fault scenarios, each validated against real plant data
Design targets from the GPU utilization plan. Nothing on this page is a measured plant result; measurements start when the first design-partner line runs in shadow mode.
From the plant floor
Who this is built for
The three people in a dairy plant whose week changes when the process runs itself, and what each of them needs from Milkeon.
Plant or Operations Director
Owns production rate, quality, loss and yield, energy and water, and food safety. Wants a plant that holds spec at higher rate with fewer interventions, and a number that justifies the spend.
Dairy Technologist or Process Engineer
Owns separation and standardization, pasteurization, culture and drying recipes, process windows and yield. Approves and corrects the agents in assist mode; those corrections train the models.
Quality and Food-Safety Manager
Owns HACCP, FSMA, ISO 22000 and FSSC conformance, composition and contamination conformance, and batch genealogy. Needs every action grounded, cited and written to an immutable record.
IT/OT and sustainability leads approve the deployment. There are no customer quotes on this site yet; design partners will speak for themselves when they choose to.
Integrations
It speaks plant, not cloud
Milkeon reads and writes through the systems already on the floor. No parallel historian, no new HMI, no rip-and-replace. Connectors are built against each design partner's stack first; long-tail integrations are out of scope for the first release.
PLC and SCADA of separators, pasteurizers, homogenizers, culture vats, evaporators, spray dryers and CIP skids
Setpoint reads, guarded writes, interlocks untouched
NIR and mid-IR composition, inline pH, moisture, temperature and humidity
Continuous composition and process state
LIMS results, calibrations, antibiotic and micro screens
Calibration and release evidence
Orders, recipes, product specs, batch genealogy
Targets in, release records out
Multi-camera stations on filling, sealing, capping, labeling and powder lines
Frames, defect events, reject commands
Plant and asset-level meters
Hard constraints for the dryer and CIP schedule
Isaac-driven packaging, palletizing and handling cells
Perception and motion for packs and pallets
SSO and role-based access; SOC 2 on the roadmap
Named approvers, scoped roles
Guardrails
An agent that can move a setpoint needs a leash
Milkeon writes to production equipment in a food plant. Every capability is scoped, every write is policy-checked, every output is cited, and every action is written to an append-only audit log the plant owns.
- Bounded action space. Each agent can only write to an explicit tag allow-list, inside per-tag rate and magnitude limits, at the autonomy stage the plant has set.
- Critical control points are read, never written. Pasteurization time-temperature, flow diversion and release decisions stay with the HACCP plan and the people who own it.
- Policy engine before every write. Autonomy stage, interlock state, product, shift and operator presence are evaluated before a setpoint moves.
- Grounded and cited. Every recommendation carries a citation to the recipe, spec or procedure it relied on. Retrieval is tenant-isolated and permission-aware.
- Immutable audit log. Append-only, exportable and retained on the plant's own storage, aligned to HACCP, FSMA, ISO 22000 and FSSC record-keeping.
- Hard fallback. If the edge node, the network or a model stops, the plant keeps running on its existing PLC and SCADA control and interlocks.
- Tenant and IP isolation. Recipes, supplier data and tolerance models never cross a customer boundary. On-prem deployment available.
Assurance posture
| Item | Scope | Status |
|---|---|---|
| SOC 2 Type I | Cloud control plane | QUEUED Roadmap, first 6 months |
| SOC 2 Type II | Cloud control plane | QUEUED Roadmap, 6 to 12 months |
| HACCP · FSMA · ISO 22000 · FSSC · Codex | Grounding and record-keeping | RUNNING Design basis for retrieval and the audit log |
| Per-tenant isolation | Data, memory, models | RUNNING Design rule |
| On-prem deployment | Recipe-sensitive producers | RUNNING Design option |
Milkeon supports your food-safety records. It is not a certification and it does not replace your HACCP plan.
Enterprise
Standardize autonomy across the group
One policy model, one audit trail and one set of yield, loss, energy and water benchmarks across every plant in the group, with each site's recipes and tolerance models kept private to that site.
- Multi-site plant-edge fleet with shared model governance
- Custom product, composition and tolerance models per site
- SSO and role-based access, named approvers (roadmap, 6 to 12 months)
- Model versioning, validation reports and rollback
- On-prem or private-cloud deployment
- SLAs and outcome-based components on the highest-value workflows
FAQ
Straight answers
The questions plant directors, technologists and quality managers ask in the first meeting.
Only at the autonomy stage your plant has set, and only to an explicit allow-list of low-risk loops inside explicit limits. Every design partner starts in shadow mode, where Milkeon predicts and writes nothing. Assist mode follows, where a technologist approves each recommendation. Bounded autonomy on separation, drying and inspection loops comes after the food-safety and ROI gates agreed up front.
Critical control points are not optimization variables. The pasteurization CCP, the flow-diversion interlock and release decisions stay with your HACCP plan and your people. Milkeon verifies readings, writes them to an immutable audit log and attaches them to the batch genealogy. It supports HACCP, FSMA, ISO 22000 and FSSC records; it is not a certification and does not replace your plan.
Milkeon sits above your existing automation as a supervisory layer. If it stops, the plant keeps running on its current PLC and SCADA control and the interlocks it already has. A Milkeon outage degrades the plant to the way it runs today, not to a stop.
The first phase is shadow mode on one line: Milkeon reads your analyzers, PLC history, lab results and MES while you run the plant as usual, and we measure its predictions against your lab. Recommendations start when accuracy is proven against that baseline, and improve as plant-specific history and technologist corrections accumulate.
Recipes, product specs, tolerance models and supplier data are tenant-scoped and never used to train another customer's models. Retrieval is multi-tenant isolated and permission-aware. On-prem deployment is available for processors protecting sensitive recipes.
You are, the same as with any control strategy. That is why every write is policy-checked, bounded, logged and reversible, every output carries a citation to the recipe or procedure it relied on, and anything above your threshold waits for a named approver. The audit log records the request, the reasoning, the limits applied and the human decision.
As a design partner, on one line, with one hard metric: fat or protein give-away, energy per kilogram of powder, moisture or composition variance, or contamination escapes. We instrument the line in shadow mode, measure the baseline, move to assist mode when accuracy is proven, and expand only when the number holds.
Full multi-module autonomy across the whole plant, long-tail integrations beyond the design partners' stacks, and self-serve onboarding. The first release is one wedge workflow, its connectors, the agent loop with human checkpoints, the review console and the food-safety audit log.
Get started
Pick your first product
Most plants start with Sepion on a standardization line or Aridyn on a dryer. Tell us which number you need to move.