How 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.
- MAIN's AI only proposes actions; a separate safety system decides whether any of them actually run
- Every proposed move passes two checks: a fixed list of safe limits, then a simulation that predicts the result
- If anything is uncertain or a check fails, MAIN does nothing — it is built to fail safe, not to guess
- A built-in hardware limit and a physical emergency-stop button work even when the software is wrong
An automated grower that can add chemicals, run a heater, and drive lights is only worth trusting if it cannot harm the culture it looks after. So MAIN follows one firm rule: the AI is allowed to suggest, but it never acts on its own. Every suggestion is handed to a separate safety system that has the final say. The clever part is upstream; the careful part is downstream, and the careful part always wins.
The first check is a plain list of limits — the ranges a Spirulina culture is known to tolerate. Temperature stays in a warm band and is not allowed to climb past roughly 38°C. The water is kept in the alkaline range Spirulina actually likes, roughly pH 8.5 to 10.5. Light stays bright but short of the level that would start to stress the culture. And there are caps on how much can be added at once, so no single dose of bicarbonate or nutrient can be large. Any suggestion that falls outside these lines is rejected instantly, with no debate.
The second check is a simulation. MAIN keeps a running physics model of the reactor — think of it as a simulated copy of your culture that predicts how the real one would respond. Before an allowed action is carried out, MAIN plays it out on the copy first. If the model predicts the culture would stop growing, or that the water would drift out of that safe alkaline range, the action is blocked. It is a test run before anything touches the real thing.
The two checks are deliberately independent, so a mistake in the AI's reasoning cannot switch off the list of limits. And the whole system is built to fail safe: if the simulation errors out or is unsure, the action is blocked rather than allowed through. When in doubt, MAIN does nothing — and doing nothing to a healthy, well-mixed culture is almost always the safe choice.
Behind the software sit two backstops that do not depend on it at all. The reactor's own low-level controller has fixed limits wired in that it will not exceed, no matter what the software asks for. And there is a physical emergency-stop button: press it and everything halts, regardless of what any program wants to do. The human hand is always the highest authority.
MAIN is also built never to leave a culture unattended. If the internet connection drops, it does not freeze up waiting to reconnect. A simple, conservative caretaker runs directly on the small computer inside the unit, keeping watch and making only cautious moves until the connection returns. The culture is never on its own.
None of this leans on the AI being brilliant, and that is the point. Spirulina already helps protect itself — its naturally high-pH home holds most contaminants at bay — so MAIN's real job is to keep the culture in that comfortable, well-lit, well-mixed band and out of trouble, not to push it hard. Safety here does not rest on any one part being right. It rests on several independent parts all having to agree before anything happens at all.