Technology

Kitchen OS — orchestration for real kitchens.

A commercial kitchen is a hard distributed-systems problem wearing an apron: shared resources, hard real-time constraints, physical state that drifts from software state, and no acceptable downtime during service.

The stack

Four layers, from the burner to the cloud.

Most kitchen automation stops at a single machine. BotKitchen is a full operating stack: every stove, dispenser, and robot arm reports to one on-premise orchestrator, and every kitchen reports to one cloud.

The layering matters because the failure modes are different at each level. A robot controller must react in milliseconds. An orchestrator must reason about an entire order. A cloud platform must stay consistent across sites. Collapsing those into one system is what makes single-station automation brittle.

CLOUD

Cloud Platform

Fleet management across regions, sites, and kitchens. Recipes are authored once and distributed everywhere; telemetry flows back up. Role-based access keeps corporate, site, and line responsibilities separate.

ORCHESTRATION

Kitchen OS

The on-premise brain. It sequences every step of every order, coordinates robots and appliances against shared resources, and keeps cooking when the internet does not. Running locally is a reliability decision, not a cost one.

DEVICE CONTROL

Device Controllers

Dedicated controllers for heat, dispensing, grinding, and motion. Each one owns its hardware's real-time behaviour and reports health continuously, so the orchestrator reasons about state rather than assuming it.

HARDWARE

Industrial Hardware

Industrial robot arms, commercial induction stoves, dispensers, cutters, and cleaning systems — standard equipment from established suppliers, integrated rather than reinvented.

Design principles

What the architecture is built to guarantee.

These are the constraints that shaped the system — each one learned from the way kitchens actually fail.

Physical truth first

The kitchen is the source of truth

Software state and physical reality drift apart — a pan moved by hand, a burner left occupied. The orchestrator treats that gap as a safety-critical failure mode and verifies against the world before acting, rather than trusting its own model.

Degrade, don't stop

Local autonomy over cloud dependence

A kitchen mid-service cannot pause for a network partition. Orchestration runs on-premise so an internet outage degrades reporting, not cooking.

Technique, not scripts

Recipes are modelled, not hard-coded

The platform models cooking technique — grinding, tempering, sautéing, simmering — so the same primitives recombine into new dishes. That is what separates a kitchen from a single-purpose machine.

Fail safe by construction

The safe state is the default state

Burners require an active, healthy controller to stay lit. Lose the signal and you lose the heat — the failure mode is designed in, not bolted on.

Engineering

Built for the kitchen environment.

Deployment

On-premise, self-contained

Orchestration runs on hardware in the kitchen with no external service dependencies in the cooking path. Fewer moving parts is both a reliability property and an operational one — a kitchen manager cannot debug a distributed system mid-service.

Observability

Every device, continuously

Stoves, dispensers, and arms are heartbeat-monitored without interrupting an active cook. Health is a continuous signal, not a question asked after something has already gone wrong.

Want the deeper technical picture?

We go into considerably more detail under NDA — architecture, safety design, and how the system behaves under load.

Get in touch