Launching soon · Recruiting pilot growers

Intelligent predictions,
informed decisions

Horticulture's weather-exposed economics, finally met with calibrated intelligence.

The problem

Today's tools miss the nights that cost the most.

And when they do fire, they hand back a useless, unreliable risk score — never a dollar figure, never a recommendation. Not one incumbent runs an AI reasoning layer.

66%

of damaging frost nights, missed by climatology-based tools

Harvora backtest · 6 NZ regions

24%

of a region's crop, lost to a single frost event

NZ Winegrowers 2025

The answer

A thousand futures.
One call you can trust.

Harvora's prediction engine turns raw atmospheric and field data into calibrated weather and agronomic probabilities — then into block-level, dollar-denominated decisions. An AI reasoning layer plans your season, triages what threatens it, and learns every block with compounding seasonal intelligence.

1,000 simulated futures · one block · one nightEnd-state distribution ↓
Damage threshold
6pmMidnight6am
Engine70% breachDecision layerProtect · tonight
94%
Frost nights caught
0.007
Expected calibration error
1,000
Simulations behind every call
0
NZ incumbents with an AI layer
Getting started

Highlight your blocks. We do the rest.

Setup is deliberately light. Upon completion, Harvora quickly gathers the data and the engine gets to predicting on your blocks.

KIWIFRUIT · HAYWARDAPPLES · ENVYCHERRIES · LAPINS$COST ASSUMPTIONSXERO · SYNCED READ-ONLYWEATHERSOILSATELLITEPREDICTING
How Harvora thinks

From probability to decision

Decision pipeline — running
Step 01
Predict

A 1000-sample Monte Carlo engine fuses physics models with machine learning across every block — emitting calibrated probability distributions, not point guesses. Deterministic, auditable, millisecond-fast.

Step 02
Price

The AI-powered finance layer converts each probability into expected value — dollars per block, per decision — in the language growers, lenders and insurers actually use.

Step 03
Decide

Expected-value thresholds and three-tier decision confidence turn priced risk into a clear call: protect or don't, irrigate or wait — AI reasoning shown, drivers attributed.

Step 04
Learn

Every confirmed outcome banks a dollar-valued decision win, builds the block's track record, and retrains the prediction engine and AI models. Intelligence that compounds, season over season.

Prediction areas
Frost
Flagship · 94% of frost nights

Catches 94% of frost nights. Block-level frost probability converted to expected-value thresholds — protect or don't, priced against crop-loss risk. Not a temperature chart, a decision engine.

Water & Irrigation
r = 0.81 · ERA5-Land, 6 NZ regions

FAO-56 evapotranspiration bounded by real S-Map soil data. Deficit trajectory tracks independent ERA5-Land reanalysis across 6 NZ regions (root-zone r = 0.805) — irrigation scheduled when your soil actually needs it.

Disease Risk
Psa · Published models

Published Plant & Food Research disease models combined with ensemble weather, producing block-level infection-risk curves growers and industry can interrogate.

Harvest & Phenology
±2-day harvest timing

Growing-degree-day projections anchored to observed heat accumulation. Harvest windows accurate to ±2 days — labour, packhouse slots, and market timing planned from evidence.

Annual Crop Intelligence
Potato · Open-field tomato

First-class crop-cycle intelligence for potato and open-field tomato. An active crop cycle anchors GDD stage, harvest readiness, irrigation context, disease windows and block economics, with four seven-day blight surfaces running the same 1,000-sample probabilistic contract as the wider platform.

Insect Phenology
Codling moth · Tuber moth · Psyllid

Codling moth, potato tuber moth and tomato-potato psyllid lifecycle timing from crop-specific degree-day models, with uncertainty over biofix, thermal thresholds and forecast weather. Block traps and scouting sharpen the regional signal.

Pollination & Wind
92% bloom · 97% wind days

Bee-flyable-hour forecasts (92% accuracy) for bloom and day-ahead warnings that catch 97% of damaging-wind days — notice enough to supplement hives or bring a pick forward.

Spray Window
96.7% rain-free · dry-guarantee

Finds windows that hold: wind, gust, temperature and a dry margin before and after, checked on the 1-day forecast. Of 335 promised windows across 6 NZ regions, 96.7% stayed genuinely rain-free — a defensible go/no-go on product, labour and compliance.

Soil Scouter
Hardware · Ground truth

Proprietary in-field soil moisture and temperature sensing feeding the prediction engine with block-level ground truth — closing the loop between forecast and soil reality.

Not a weather app.
An AI prediction platform.

Harvora's machine-learning prediction engine is calibrated against what actually happens on your blocks. It forecasts frost, disease, water stress and harvest windows as honest probabilities, not vague icons. An AI reasoning layer then turns every prediction into dollars and a decision: what's likely, what it costs, and what's worth doing about it.

Harvora Intelligence

A 24-feature AI reasoning layer — built for horticulture

Every AI feature exists because of a Harvora prediction. The discipline is grounding-before-generation: deterministic engines find the signal, the AI puts it into words, and a structured fallback ships if it can't. Agentic tool-calling, retrieval-augmented answers, cross-checked responses on high-stakes questions — all sitting on an audited reliability layer, with you making every call.

Explore the intelligence layer
AI Season Planner

A prediction-driven calendar. Spray, irrigation, frost, harvest and more, planned ahead — and kept current as each new forecast sharpens the picture.

Insights Feed

Every night, Harvora reads each block's trajectory and gathers what builds too slowly to trip an alarm — drifts, accumulating pressure, quiet anomalies — waiting in-app by morning.

AI Alerts

The urgent calls, pushed to your phone. The reasoning layer decides what's critical and sends it by SMS and email — AI-driven out of the box, with manual thresholds still there if you want them. Deduped, intelligently bundled and paced around quiet hours and action windows.

Ask Harvora

An agentic assistant grounded in your blocks, your history, your numbers — down to actual costs synced from Xero. A bounded tool-calling loop that retrieves, reasons and runs what-ifs, powerful enough to act — disciplined enough to never act without you.

The rest of the layer
19 features
01Daily BriefingEach morning's decisions, ranked by economic impact.02Nightly Insight ScanFive deterministic detectors sweep every block at 2:30 am.03Decision WinsConfirmed calls valued in dollars, banked per block.04Risk-Weighted PlanningSeason-scale Monte Carlo, priced into revenue impact.05Decision OptimisationA lever sweep — in-chat or overnight — ranking frost, irrigation and harvest actions by dollar outcome. Scenario search, never a forecast.06Pre-Decision ChecklistsBefore you run the wind machine: check X, Y, Z.07Season ReportsWeekly, pre-season, debrief and agronomist-grade PDFs.08Voice LoggingHands-free structured logging from the orchard row.09Inferred LoggingLikely actions detected, confirmed with one tap.10Staff ViewA plain-language day sheet for the crew, one tap to share.
11Analysis CarouselNightly critical, watch and good items drive the dashboard's Today's Priority banner.12Ledger AnalysisAsk reads your synced Xero ledger — summaries, P&L and period comparisons, orchard-scoped.13Account MappingSuggests fixes for unmapped Xero codes so cost overlays ground on real spend.14Conversation MemoryAsk remembers your operation across sessions — “as you mentioned last week.”15Grower-Learning SynthesisProposes durable profile facts from the day's activity, always yours to confirm.16Chart ReadingsExpand any chart for a grounded, cached read of what it shows and why.17Season-State SummaryA weekly on-screen read of where each block sits in its season.18SMS & WhatsAppThe full agent, reachable from any phone — identity-gated and keyword-routed.19Reliability HardeningAn audited layer beneath all 24 features — every answer budgeted, validated and safe against bad input.
Season over season

The longer you grow with it, the better it knows your land

Most software is the same on day one thousand as it was on day one. Harvora isn't. Every season you farm with it, it remembers — and what it remembers, it turns into sharper predictions, quieter alerts, and advice that sounds less like a tool and more like someone who has walked your rows for years.

Season one
It learns your land

Every confirmed frost call, every irrigation, every harvest teaches Harvora how each block actually behaves — where cold air pools, how your soils drain and dry, how heat accumulates row by row.

Season two
It anticipates you

Briefings arrive tuned to how you farm. Thresholds settle around your risk appetite. The question you were about to ask is already answered, in your numbers, for your blocks.

Every season after
It compounds

A track record banked in dollars. A memory of every outcome. Models retrained on your ground truth. The intelligence you build never leaves the orchard — it deepens with it.

Calibration as brand

Published accuracy.
Every prediction.

When Harvora says 30%, it happens about 30% of the time — and we publish the reliability data to prove it. Every surface ships with a measured accuracy stat on real New Zealand data. Calibrated probabilities are trust infrastructure: the foundation growers, lenders and insurers can build decisions on.

See the full proof
94%
Frost nights caught
15–18%
Sharper than the public forecast
0.007
Expected calibration error
r = 0.81
Irrigation vs ERA5-Land · 6 regions
97%
Damaging-wind days caught
92%
Pollination-window accuracy
±2 days
Harvest-timing accuracy
2,103
Python tests passing · 87 optional skips
486
Web tests passing
9 · 13
Crop families · NZ regions
Soil Scouter · Hardware, launching soon

A soil sensor node in every block — ground truth, live

Harvora's own in-field hardware: a self-contained, solar-assisted node with a seven-parameter probe in the root zone, reporting every 15 minutes over WiFi or LoRa. Today the engine forecasts your soil from weather; Soil Scouter lets it measure it — correcting the irrigation model's starting point, feeding the per-block learned models, and providing the real-world labels that unlock the deliberately held-back runoff and infiltration ML.

Seven parameters, one probe

Volumetric moisture (±2%), soil temperature (±0.5 °C), electrical conductivity, pH and N-P-K from a single RS485 probe at root depth — plus ambient temperature and humidity above the ground.

Measured initial conditions

With a reading in the past 24 hours, measured soil moisture replaces the modelled estimate as the irrigation forecast's starting point — cutting the error that accumulates in evapotranspiration estimates.

A season on its own power

Solar-assisted lithium cells, a ~20-second wake cycle every 15 minutes, deep sleep in between. Months of unattended operation in an IP65-sealed enclosure — no wiring in the block.

WiFi first, LoRa always

Readings post over orchard WiFi; where there's no coverage, LoRaWAN carries them out — a single gateway covers roughly 5 km of open farmland.

Field notes, in detail
08 features
0115-minute cadenceConfigurable sampling interval; microamp deep sleep between cycles.02Fresh-data priorityEach new reading flags the block so prediction jobs process it first.03Deficit detection48 hours below the irrigation trigger raises an accumulating-risk finding.04EC anomaly watchA simultaneous conductivity rise across 3+ blocks flags agronomist review.
05Weekly recalibrationModelled-vs-measured error updates each block's learned irrigation response.06Leaf-wetness proxyBlock-level humidity sharpens Psa antecedent modelling beyond regional stations.07Soil panel per blockMoisture trend, EC sparkline, NPK and battery, live in the block view.08Honest by designNPK and pH ship marked indicative — lab confirmation advised, never implied.
The economics

Frost alone is a five-figure decision — per hectare, per season.

Exposure runs to tens of thousands of dollars a hectare. Harvora prices every risk night in dollars and tells you when protection pays for itself — before you start the wind machine, not after. And with read-only Xero integration, the AI reasons with your actual costs and returns — your dollars, not industry averages.

23.4

Damaging frost nights per season, average NZ location

66%

Of frost nights missed by climatology tools

94%

Caught by Harvora at the economic setting

Work with us