- Document
- AgroNext · Local AI
- Sheet
- 04 / 07
- Scope
- field prototype · product development
Local AI
The decision is made inside the greenhouse.
AgroNext's model runs on an Arduino-based local unit in the greenhouse. It evaluates climate and root-zone readings, the last hour's trend and the plant profile together, and produces separate decisions for irrigation, feeding, heating and the fan.

| Model | GRU time-series |
|---|---|
| Inputs | temperature · relative humidity · soil moisture · CO₂ |
| Window | 12 steps × 5 min (1 hour) |
| Horizon | 30 min |
| Freshness | readings older than 15 s are excluded |
Decision and command are separate
The model's output is a proposal. It cannot reach a relay without passing data validity, operating limits, duration and interval rules. This check runs on fixed rules, independent of the model.
Why local?
The decision never leaves the local network the reading arrived on. An internet outage does not stop control in the greenhouse; the remote panel and NexAI simply wait for sync.
Traceable decisions
Every decision is logged with the readings it used, the model and rule version, the command issued and the device's response. The user sees what was done and why in the panel.