verde·ai
← Back to Verde
advisory agent · prototype
Verde Garden Intelligence Platform

Ask the garden why — not just when to water.

An AI advisory agent for commercial growers. It reads your soil sensors, the forecast, and your water budget — then diagnoses before it prescribes, with a confidence label on every recommendation.

Join the pilot
session · pilot site NSW · 22 Jul 2026
live · grounded in pilot-site sensor, forecast & budget data · try: Zone 3 diagnosis · Saturday rain call · Budget trade-off

Diagnosis before irrigation

Most controllers run schedules. The Advisory Agent reasons over live data through three Verde Cloud tools, grounded by agronomy notes — so a valve fault doesn't get treated like a drought.

get_sensor_history

Daily soil moisture and temperature per zone, plus the irrigation run log and fault flags — the evidence for weather-vs-equipment diagnosis.

get_forecast

7-day temperature, rain and reference ETo from the Forecast Confidence Engine — with confidence values the agent must respect before acting on rain.

get_water_budget

Month-to-date allocation, burn rate and projected overrun from the Water Budget Engine — so plans fit the water you actually have.

grower question → agronomy retrieval → OpenAI model + tool calls → quantified action plan

Held to account by evals

Every build runs against a graded scenario suite before it reaches a grower. Three checks per scenario:

grounding

must_call — the agent has to consult the right tools. No answer about Zone 3 without reading Zone 3's sensors.

correctness

must_mention — the diagnosis and figures the answer has to contain: the valve fault, the soil type, the projected overrun in kL.

tripwires

must_not — hallucination and unsafe-advice checks, like cancelling irrigation on a 55%-confidence rain forecast.