Platform — the full distillery autonomy stack
Caskent connects stills, fermenters, lab instruments, warehouses, bottling cells, and energy systems into one governed operations layer for spirits production.
One system of action from edge to cloud.
Edge runtime closes the plant loop. Cloud training improves policies from labeled runs, sensory feedback, quality outcomes, and rare-fault simulation.
Seven agents. One production loop.
Each agent links to a bounded control area and shares context through the site model.
- 01Explore →
Ferment & Wash
Trajectory control for temperature, pH, gravity, and yeast health.
- 02Explore →
Distill & Cut
Real-time heads, hearts, and tails control.
- 03Explore →
Spirit Quality
Congener, methanol, clarity, and off-note sensing.
- 04Explore →
Cask & Maturation
Cask selection, extraction, oxidation, and angel's share.
- 05Explore →
Blend & Proof
Blend recipes, proofing, dilution, and filtration to spec.
- 06Explore →
Robotic Bottling
Fill, cork, label, inspect, and package with line feedback.
- 07Explore →
Energy & Yield
Steam, heat recovery, yield, feints, and loss optimization.
Closed perceive, decide, act control.
The platform reads plant signals, reasons against site limits, writes bounded setpoints, then records every action for review.
- Plant telemetry and lab labels fused in one site state.
- Policy router selects the safest approved model path.
- Operators can pause, approve, or revert any action.
Prove the run in the digital twin.
Simulate cuts, congener profile, cask maturation, energy, and robot paths before committing policies to the plant.
- Compare candidate runs against the product spec.
- Train against rare faults without risking production.
- Open the digital twin page →
Connectors and integrations.
Signals arrive at the edge. Approved actions write back only inside the operating envelope.
| System | Signals | Control write-back |
|---|---|---|
| Still telemetry | ABV, reflux, plate temperatures, flow, pressure | Reflux, heat, valve setpoints |
| Fermenter | Temperature, pH, gravity, Brix, yeast health | Cooling, nutrient, agitation limits |
| Spectrometer / GC | Congeners, methanol, volatile markers | Quality gates and cut limits |
| Cask warehouse | Humidity, fill, location, tasting, loss | Cask selection and movement plan |
| Bottling line | Fill, cork, label, vision defects | Robot cell and rejection rules |
| Checkweigher / MES | Weight, lot, batch, downtime, OEE | Batch records and line gates |
Architecture built for site reality.
Edge runtime and cloud training
On-site runtime keeps control close to the still, while the cloud improves models from approved records.
Model-agnostic router
Routes between frontier models, fine-tuned open models, and deterministic solvers by risk level.
pgvector, TimescaleDB, RAG
Time-series data, recipe knowledge, procedures, and safety standards stay queryable with citations.
Assurance and trust.
Grounded outputs
Responses cite procedures, recipes, and safety rules used to support a recommendation.
Bounded autonomy
Hard safety and product-quality limits wrap every control policy.
Human checkpoints
Teams choose shadow, assist, or graduated autonomy by unit operation.
Audit trail
Operator approvals, model context, and write-backs are preserved for review.
Connect, simulate, engage.
A staged rollout proves value before scaling across the site.
- Step 12 to 4 weeks
Connect
Attach Caskent edge to one still, fermenter, warehouse zone, or line.
- Step 2Twin fit
Simulate
Calibrate the twin against historical and live production runs.
- Step 3Governed
Engage
Move from shadow to assist to graduated autonomy as confidence grows.
Outcomes tracked from day one.
Illustrative design-partner targets, measured against each site's baseline.
Platform questions.
Does Caskent replace existing PLCs and controls?
No. It sits above existing OT systems, reads validated signals, and writes approved setpoints through controlled interfaces.
Can we start with one workflow?
Yes. Most design partners start with a still, fermenter, bottling line, or quality workflow, then expand once value is proven.
What data is required?
Telemetry, lab results, sensory notes, batch records, and procedures are enough to begin. More labeled outcomes improve the policies.
How is craft preserved?
The master distiller sets the spec, limits, recipes, and approval mode. The platform executes within that envelope.
Request access to the autonomy stack.
Route one still or line through Caskent and measure quality, yield, energy, or assurance impact.