Life, Cultured by Intelligence
A helical cyanobacterium. An alkaline miniature sea. And an onboard mind that keeps the two in balance. Grown by hand — shared in the open.
How well do you already know these living cultures? Choose the way in that fits your hands.
MAIN — the mind inside the reactor
MAIN reads six sensors — pH, temperature, light, optical density, dissolved oxygen, and TDS — and feeds them into a physics-based digital twin. The twin reasons about the culture the way you would, only continuously: Droop's cell-quota kinetics for nutrient uptake, Steele's curve for the light optimum and photoinhibition, Beer-Lambert attenuation for self-shading, plus bell-shaped temperature and pH responses and a dissolved-oxygen inhibition term. From that, it forecasts growth and proposes the next move — a bicarbonate dose to recover carbon, a nudge of nitrogen, more air, a change in light.
Nothing it proposes acts unchecked. Every action passes a two-layer safety gate — deterministic hard limits first, then a twin simulation that blocks any move predicted to stall growth or push pH out of the safe band. It fails closed. Firmware clamps and a physical emergency-stop sit underneath as independent backstops. The AI proposes; the culture is never at its mercy.
It runs entirely on a Raspberry Pi Zero 2W. Lose the internet and a lightweight rule-based strategy takes over on the Pi, so the culture is never left alone. And through PhycoNet, reactors share what they've learned — the twin's tuned parameters, never your raw data — so the whole network gets better at growing a given strain toward a given goal. All of it open-source, educational, non-commercial.
Read the technical deep-divesA live-style model view — MAIN's six sensors and the digital-twin growth forecast it reasons with before it moves a single pump.
It senses
Six sensors, read continuously: pH, temperature, light, optical density, dissolved oxygen, and TDS. The full state of a living culture, in numbers — analog channels through an MCP3008, an optional ESP32-CAM for a microscope's-eye view.
It predicts
A physics-based digital twin forecasts growth before anything moves — Droop, Steele, and Beer-Lambert doing the reasoning you'd do by eye, only every few seconds, and simulating each proposed dose before it's ever poured.
It protects
A two-layer safety gate hard-limits every action, then simulates it and blocks the harmful ones. It fails closed, backed by firmware clamps and a physical e-stop you can hit with your palm.
It learns
Through PhycoNet, reactors trade tuned twin parameters — never raw data — and every incoming value is bounds- and finite-checked before it touches a live twin. Each reactor grows a little wiser from the whole network.
How MAIN works — in real depth
The engineering and the biology behind an autonomous reactor: the digital twin, the safety gate, the sensor stack, computer vision, and the federated network that lets reactors learn from one another.
The Six Signals MAIN Senses — and What Each Reveals
pH, temperature, light, optical density, dissolved oxygen, and TDS each tell a different biological story about a living culture.
Read the articleThe Digital Twin: Predicting Growth with Droop and Steele
A physics-based model of nutrient quotas and light saturation lets MAIN forecast growth before it ever moves a pump.
Read the articleAlkalinity as a Control Variable
For Spirulina, bicarbonate chemistry is carbon source, pH buffer, and biosecurity all at once — which is why MAIN controls alkalinity, not just acidity.
Read the articleDissolved Oxygen: When Photosynthesis Poisons Itself
In a bright, dense culture, photosynthetic oxygen can supersaturate to levels that inhibit the very growth that produced it.
Read the articleBounded Autonomy: Keeping AI Proposals Safe
MAIN lets an AI propose experiments but never lets it act unchecked — every command clears a deterministic rule gate and a physics simulation first.
Read the articlePhycoNet: Federated Learning Across Reactors
Reactors trade the learned parameters of their digital twins — never raw culture data — so the network converges on a strain’s productive regime faster than any one reactor could alone.
Read the articleInside MAIN's Control Loop: Sense, Model, Decide, Verify, Act
How MAIN closes the loop around a living spirulina culture on a single Raspberry Pi, and why every decision passes through a safety gate that fails closed.
Read the articleMachine Vision for Spirulina: Catching What the Numbers Miss
MAIN's six numeric sensors describe the water; an optional camera describes the cells — and a picture catches the contamination, stress, and morphology that numbers alone smear into noise.
Read the articleMAIN, in plain language
No jargon required. What the system actually is, why it matters, and how it keeps a living culture safe — explained for anyone curious.
Why Spirulina Is a Cyanobacterium, Not an Alga
Spirulina is a photosynthetic bacterium, not a plant-like alga — and that single fact reshapes how you feed, buffer, and protect the culture.
Read the articleWhat Is MAIN? A Plain-Language Guide
MAIN is a small AI that tends a living spirulina culture, reading its water minute by minute and making careful, double-checked corrections so the culture is never left to guesswork.
Read the articleHow MAIN Keeps Your Culture Safe
MAIN's AI is allowed to suggest changes, never to make them on its own — every move clears two independent safety checks, with hardware limits and a physical stop button behind them.
Read the articleBuilt in the open
We build these in the open. Watch the reactors get designed, plumbed, flooded, and grown on YouTube — @OrrBiologicals — where every clean run and every crash makes the cut.
Visit @OrrBiologicalsReserve a reactor
MAIN and PhycoNet are experimental, open-source, and non-commercial — built by hand, one at a time, by a young maker who believes science should be open to everyone. If you want a reactor of your own, or just want to follow the work, say hello.
Reserve yours — email usExperimental · open-source · non-commercial. This opens your email to service@orrbiologicals.com.