The pond that records itself
ShrimpOS is the software layer on top of the instrumented pond: what is running on the farm today, what is being built, and what has to be true before the next step is taken.
Running on the pilot farm
Continuous monitoring
Five ponds logging dissolved oxygen, temperature, pH and TAN every 30 seconds since late May 2026, mirrored locally on the farm gateway.
Threshold alerting
Operators are alerted when a pond crosses a configured threshold. Running continuously since early July 2026, after a four-week sensor outage on two ponds in June.
Pilot dashboard
Per-pond and per-group views of the sensor record, including the dawn window that the trial is scored on.
ShrimpOS v1
Event detection
Turning raw series into labelled events — a dawn dip, a post-feed swing, an exchange — so that the record reads as a crop history rather than a chart.
Biological model
Trained on our own pond data: environment, growth, pathogen and energy variables on one timeline. Every pilot pond is a labelled training set.
Multi-farm control tower
Productising the pilot dashboard so that several farms, and several crops, are read the same way.
Thirty to fifty ponds, then predictive control
Widen the sensor estate
More ponds on more farms, under the same measurement discipline, before any model is trusted to act.
Forecast
Monitoring becomes forecast when the biological model has enough labelled crops behind it.
Oxygen policy
Forecast becomes an oxygen policy the system executes on the operator's confirmation — a setpoint, not a clock.
One night, as ShrimpOS is designed to handle it
The following is a design walkthrough, not a record of an event. It describes how the pieces are meant to work together on a night when dawn oxygen would otherwise fall.
- 22:40
Dissolved oxygen in one pond is declining faster than its seven-day pattern. Monitoring records it; nothing yet crosses a threshold.
- 01:15
The pond crosses its configured lower threshold. Threshold alerting notifies the operator on duty.
- 01:20
The operator raises the oxygen setpoint for that pond from the app; the change is logged against the pond's timeline.
- 05:30
The dawn window is scored: the pond's dawn mean, the setpoint change and the operator's note sit on one record.
- Later
In build: event detection would label the 22:40 decline itself. On the roadmap: the model would propose the 01:20 change before the threshold was reached, for the operator to confirm.
Design principles
Measured, not estimated
Every number the system shows can be traced to a sensor reading or an operator entry with a timestamp.
Local first
The farm keeps its own copy of its own data on the gateway. Connectivity loss does not lose the record.
Setpoint, not clock
Oxygen delivery is controlled against the pond's state, not a timer.
Operator in the loop
The system proposes; a person confirms. That stays true through the roadmap.
Bring us a pond. We will bring the instruments.
We would rather show you a dataset from your own water than argue about ours. If you run intensive ponds and want one of them measured properly, that conversation takes fifteen minutes.