AI agent
Five modes. Three scopes. Inheritance sorts the rest.
PlantHub ships with five AI presets and a hierarchical scope model — tenant, zone, or device. Set once at the top, override only where you need to.
The five presets
Conservative
Errs on the side of doing nothing. Best for delicate plants and slow learners.
- Higher moisture threshold before watering
- Shorter watering durations
- Longer cooldowns
- Fewer daily evaluations
Balanced
The default. Good for most houseplants and outdoor pots.
- Sensible defaults from the plant profile
- Trusts the learning loop as confidence climbs
- Skips watering on rain forecast
Aggressive
For fast-growing plants, productive greenhouses, and hydroponics.
- Lower moisture threshold
- Longer watering durations
- Shorter cooldowns
- More frequent evaluations
Power-Saver
For battery and solar setups, or anywhere LLM API costs matter.
- Adaptive intervals stretch longer between checks
- Compiled offline rules carry the load
- AI only invoked when sensors look unusual
Manual
You decide everything. PlantHub just watches and reports.
- AI is disabled for this scope
- Rules still fire if you have them
- You receive alerts and recommendations but no autonomous actions
Three scopes, one inheritance rule
Settings cascade: device overrides zone, zone overrides tenant, tenant overrides system defaults. You usually set the tenant default and then override only the few devices that need different behavior.
- Device — a single ESP32 node. Useful for a finicky orchid that needs Conservative mode while everything else is Balanced.
- Zone — a group of devices sharing location and environment type (indoor, outdoor, greenhouse). Useful for "everything in the greenhouse runs Aggressive."
- Tenant — all your devices. The default everything inherits from.
Tunable knobs (advanced)
Each preset is a bundle of underlying parameters you can override directly if you want — see API reference for the AI settings endpoints:
- Watering enabled / lighting enabled / ventilation enabled per scope
- Max watering duration (capped by hard safety limit)
- Evaluation frequency (or "adaptive" — let the AI decide)
- Daily evaluation cap (to control LLM spend)
- Autonomous mode — let the AI act, or notify-only