Agriculture AI agent · Greenhouse climate
Greenhouse Climate-Control Copilot (Setpoints & Envelope)
Propose temperature, humidity, CO2, screen, vent and lighting setpoints against the crop strategy your grower set, move them only inside a signed envelope, and leave the climate computer's safety logic untouched.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Setpoint moves
- Signed envelope
- After launch
- Agent Care
What this agent does
Runs the strategy a grower wrote
In
- 01
Read each compartment — temperature, humidity, VPD, CO2, radiation, screens and vents — and the weather outside.
- 02
Read the crop strategy: crop, stage, 24-hour temperature target, humidity band, CO2 target and the lighting plan.
Reason
- 03
Work out where the compartment sits against the strategy, and what the next few hours of weather will do to it.
- 04
Fit the change to the equipment that must make it — pipes, vents, screens, lights, fans and the CO2 supply.
- 05
Test each setpoint against the signed band, equipment limits, light-abatement hours and the tariff window.
Decide
- 06
Sort each change into one the envelope covers and one the grower must release, and show the band it sits in.
- 07
Hold the change, or hand the compartment back, when a sensor, a screen or the weather stops agreeing with it.
Out
- 08
Write only the setpoints the envelope covers, one compartment at a time, with the value before and after.
- 09
Draft the climate log: the readings, what moved, what the climate computer did with it and what the grower decided.
Product statement
The agent proposes setpoints and moves them only inside a signed envelope. The crop strategy, the CO2 rule around people and every equipment limit sit outside it.
Example workflow
One setpoint change, end to end
- 1Compartment readAir and plant temperature, humidity and VPD, CO2, radiation, screen and vent position, outside weather
- 2Strategy readThe grower's plan for this crop and stage: 24-hour target, humidity band, CO2 target, lighting and screens
- 3Change proposedWhich setpoint moves, how far, for how long, and what the compartment is expected to do in response
- 4Envelope testedBand, step size, compartment and expiry, against equipment limits, abatement hours and the dosing rule
No human action required
Stages 1 to 4 run without a person in the loop — reading, proposing and envelope-testing finish before the grower is asked for anything. A change the band does not cover stops where it stands.
Written to the climate computer.
Waits for the grower to release it.
Grower
Reads the change, the readings behind it, the band it sits in and what the compartment is expected to do, then decides.
Release · Change the strategy · Hold the setpoint
Released — handed back
- 6Setpoint writtenInto the climate computer inside the band; anything outside it waits for the grower to release it
- 7Outcome evaluatedWritten against achieved, sensor health, breach attempts, alarm latency and grower overrides
Overrides
A setpoint the grower rejects or rewrites is recorded with the reason and counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Setting the crop strategy and its 24-hour targets.
- Signing the unattended envelope, its bands, expiry and stop.
- Clearing a compartment for entry while CO2 is dosing.
- Silencing, bypassing or re-arming a climate alarm.
Automation boundary
Agent acts unaided
- ✓Read every compartment against the strategy the grower signed.
- ✓Propose setpoint changes with the readings and the band behind them.
- ✓Move pre-authorised setpoints inside the signed bands, one compartment at a time.
- ✓Hold, hand back and raise the alarm path when a reading or a limit moves.
Write actions stay inside the envelope agreed in implementation: named compartments, a band, an expiry, and a stop the grower can pull at any time.
- Changing an equipment, wind or structural limit.
- Running lights or screens inside abatement hours.
- Restoring control after a stop or a hand-back.
- Changing envelope bands, stop rules or escalation paths.
Example output
One night's setpoint, annotated
Everything the agent proposes is attached to the reading it came from and the limit it was tested against.
Setpoint output · single proposed change
Illustrative example- Compartment
- House 4, tomato
- Reading
- Humidity above the band
- Proposed change
- Pipe up, screen gap
- Setpoint result
- Written — screen held
- Confidence
- 91%
- Crop strategy
- The grower's, unchanged
As received
The compartment as the aspirated box and the climate computer report it, beside the change proposed.
Evidence used
- Aspirated box, plant height
- Pipe and screen position
- Forecast for the night
Why the screen waited
The gap would have opened inside the abatement hours the site is bound by.
What the score decides
Confidence decides how hard the grower checks the sensor, not how far a setpoint may move.
Action
ReleaseChange the strategyHold the setpointValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every compartment
From the climate computer and the sensors
Apply the strategy that governs it
Use the grower's crop strategy, the signed bands, the equipment limits, the abatement hours and the compartment's own history.
Act on the house before the crop feels it
Humidity drifting toward a disease window, or a screen and a pipe working against each other, is caught while a setpoint can still be changed.
Put the strategy change to the grower
A change the band does not cover, a sensor that has stopped speaking for its compartment, or weather that arrived differently reaches the grower.
Build an evidence trail
Retain the readings, the setpoint before and after, what the climate computer did with it, the alarms raised and what the grower decided — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Climate computers
- Priva
- Hoogendoorn
- Ridder and Argus
- setpoint and alarm APIs
Sensors & measurement
- Aspirated boxes
- plant sensors
- CO2 and radiation
- outside station
Equipment & energy
- Screens and vents
- pipe and boiler
- Lighting and dimming
- CHP and meters
Agent
Greenhouse climate copilot
- Reads the house
- Proposes setpoints
- Writes in the band
Crop & production records
- Crop registration
- scouting and labour
- Strategy plans
- Operations Center
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the climate computer
Each control wraps the one inside it. A setpoint clears every layer before it is written, and the climate computer's own safety logic and the dosing interlock sit outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide how far a setpoint may move
- L3
- Stops what the equipment cannot take
- L4 – L5
- Bound unattended writing and keep the record
- L6
- Hands the compartment back when signals degrade
How Nestack evaluates it
Evaluate the climate the crop stood in — not only the setpoint.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the change the grower opens
- E1Setpoint reconciliationDid the compartment follow the setpoint that was written?
- E2Step-level evaluationDid it read the right sensors, strategy, crop stage and limits?
- E3Tool evaluationDid the write land on the compartment and setpoint it named?
- E4Sensor-health evaluationHas a sensor drifted, or stopped speaking for its compartment?
- E5Breach and alarm latencyWere band breaches stopped, and how fast did an alarm land?
- E6Business outcomeWhich compartments and stages carry the overrides, and why?
Floor — the climate the crop stood in
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycle
Direction of processing →
01 · Sensing / retrieval
2 modes
- GH-01
Sensor drifted or badly placed
A whole compartment is controlled to a wrong reading.
- GH-02
Forecast moved, strategy did not
The house is pre-set for weather that arrives differently.
Stage gathers
Compartment climate, outside weather and equipment state
02 · Strategy
2 modes
- GH-03
Setpoint the house cannot hold
Agronomically fine, past an equipment or structural limit.
- GH-04
Humidity held in a disease window
The band allows an hour of leaf wetness the crop pays for.
Stage proposes
Which setpoint should move, how far and for how long
03 · Envelope gate
1 mode
- GH-05
Dosing continued with people inside
Enrichment runs on while a compartment is being worked.
Stage bounds
Bands, limits and abatement hours before a write
04 · Write / control
1 mode
- GH-06
Command fights the climate computer
A screen or vent move contradicts its own safety logic.
Stage writes
Only what the grower signed, one house at a time
05 · Reconciliation
1 mode
- GH-07
Least-seen compartment drifts
Control degrades where the model has least history.
Stage reconciles
Setpoints written against the climate the crop stood in
- Sev-1 · people or equipment are put at risk
- Sev-2 · the crop stands in the wrong climate
- Sev-3 · control degrades, more to the grower
Affected slices
Deviation sits in the weeks before a crop settles down
A deviation rate that reads acceptably across a site can sit almost entirely in the compartments planted last and those where one sensor speaks for a whole house. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Young crop, first weeks | 6.5% | 3.4× | Review | |
| Compartments with one sensor | 5.2% | 2.7× | Review | |
| Cold nights under screens | 3.6% | 1.9× | Watch | |
| Established fruiting tomato | 1.9% | 1.0× | Normal |
Bar: setpoint-deviation rate lift vs. established-crop baseline · scale 0–4.0× · tick marks the 2.0× threshold
2 of 4 slices over threshold
Evidence-linked improvement
An hour of wrong climate is already in the crop
A setpoint that went wrong is closed out in the house first — hand the compartment back, tell the grower, note it in the crop record.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Compartment by compartment: written-versus-achieved gaps, override rate and alarm latency.
- 02
Diagnose
A sensor, the strategy as read, a band, the write path or the equipment — one of them explains it.
- 03
Improve
The grower signs the changed sensor, band or rule, and the envelope version moves with it.
- 04
Verify
A held-out week of compartment data is replayed hour by hour against the new band.
- 05
Learn
That hour is replayed on every release, and the limit it crossed is written into the standing check.
Learn → Detect
The return edge. Each cycle ends with the envelope re-read — a band widens only when the grower signs a new one, and narrows on our side without asking.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, setpoint logic, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Crop strategy and growth-stage targets.Week 1, Week 2
- 03Compartment and equipment inventory.Week 1, Week 2
- 04Climate-computer and sensor assessment.Week 2
- 05Setpoint proposal logic per compartment.Week 2, Week 3
- 06Screen, vent, heating and dosing logic.Week 3
- 07Equipment-limit, abatement and tariff gating.Week 3, Week 4
- 08Signed envelope, step, expiry and stop.Week 4
- 09Sensor-health and setpoint evaluation.Week 4, Week 5
- 10Write path and alarm integration.Week 5
- 11Climate-log drafting and override capture.Week 5, Week 6
- 12Observability, deployment and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne compartment | ProductionProduction climate-computer link | AdvancedMulti-site / multi-crop |
|---|---|---|---|
| Introduced at Pilot | |||
| Setpoint proposals by compartment and crop stage | ✓ | ✓ | ✓ |
| Crop strategy read from the grower's own plan | ✓ | ✓ | ✓ |
| Sensor-health and representativeness checks | ✓ | ✓ | ✓ |
| Crop strategy left with a named grower | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Equipment-limit, abatement and tariff gating | — | ✓ | ✓ |
| Pre-authorised setpoint moves inside a signed band | — | ✓ | ✓ |
| Climate-log drafting and override capture | — | ✓ | ✓ |
| Observability and evaluation | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-compartment and multi-crop strategies | — | — | ✓ |
| Multi-site, contractors and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on compartment and crop count, climate-computer and sensor integrations, equipment and energy constraints, envelope scope, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your crop strategy and its targets by growth stage
→Crop strategy, growth stages and target mapping
Week 102Compartments, equipment, capacity and the known limits
→Compartment, equipment and capacity inventory
Week 103Climate-computer access and where each sensor actually sits
→Climate-computer, sensor and weather source assessment
Week 204The band you are willing to sign for unattended moves
→Signed envelope, step size, expiry and stop rules
Week 405Weeks that went wrong — a cooked house, a Botrytis run
→Evaluation suite, regression cases and failure-mode testing
Week 406Your alarm paths, dosing rules and entry practice
→Alarm integration and the limits the agent may not touch
Week 507A named grower and a technical service lead to review
→Review workflow, then supervised setpoint moves
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the first supervised setpoint moves and the alarm tests run against them.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Crop strategy, compartments and the automation boundary
W2Climate-computer access, sensors and the first compartment
W3Setpoint logic and screen, vent, heating and dosing coordination
W4Envelope bands, equipment-limit gating and sensor-health testing
W5Write path, alarm integration and supervised setpoint moves
W6Moves run inside the signed band, then Agent Care starts
- Reading the band
- The envelope is written and tested in week 4, before a setpoint moves on its own in week 5. Nothing is written to a live house until it has.
- At the end of W6
- Proposals have run alongside your grower's own strategy, been read by the person who signs it and reconciled against what the compartments achieved, then Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Agriculture AI agent
Build a climate copilot around the strategy your grower already writes.
Show us a season of climate data, the strategy behind it and the band you would sign for unattended moves. We'll agree the envelope first — which compartments, which bands, what expiry and what stops it — then test it against those weeks.
- Typical delivery
- 4–6 weeks
- Deployment
- Your climate-control stack
- Setpoint moves
- Inside a signed band, stoppable
- After launch
- Agent Care available
Nestack Agents · Greenhouse climate control
AGT-AGR-09 · Agent Care available after launch