Source.ag

A high-tech greenhouse is one of the most controlled production environments in existence. Temperature, humidity, light, CO2, water and nutrients are all adjustable, and the crop responds to every one of them.

That control is also the problem. A grower makes hundreds of decisions a week across variables that interact, on a crop that will not reveal whether they were right for another six weeks — and by then the decisions have compounded.

Source.ag builds software for exactly that gap: forecasting what the crop will actually produce, simulating strategies before committing to them, and automating the daily adjustments that used to depend on someone's judgement at seven in the morning.

What Does A Harvest Forecast Actually Predict?

Harvest Forecast predicts yield with dynamic, data-driven adjustment, and understanding why that number matters explains the whole product.

Greenhouse produce is sold forward. A grower commits volumes to buyers weeks ahead of harvest, and those commitments are made on an estimate of what the crop will produce — an estimate that has traditionally rested on experience and counting.

Being wrong is expensive in both directions, which is the part people outside the industry miss. Over-forecast and you cannot supply what you sold, which costs money and costs the relationship. Under-forecast and you have produce with no buyer, on a crop that does not wait.

A model that updates continuously from actual conditions and plant data is doing something a weekly walk-through cannot: adjusting the estimate as the crop develops, rather than at the points a person happened to look.

The honest caveat with any forecast is that it is a prediction, and biological systems produce surprises. A disease outbreak, an equipment failure, an unusual weather run — all of these break a forecast, and none of them are visible in the data before they happen. Better forecasting narrows the error; it does not eliminate it.

The open-field equivalent works from a completely different data source. Taranis flies imagery over acres nobody can walk, because outdoors there is nothing to control and everything to observe.

Greenhouse produce is sold before it exists. Over-forecast and you cannot supply what you sold; under-forecast and you have fruit with no buyer.

Three Crops Is A Deliberate Limit

Source.ag covers tomatoes, bell peppers and cucumbers, and that narrowness is a design decision worth crediting rather than a gap in coverage.

The reason is that crop models are crop-specific in a way that does not generalise. A tomato plant's response to light, temperature and nutrient concentration is not a parameterised version of a cucumber's — the growth stages differ, the fruiting behaviour differs, and the failure modes differ.

A platform claiming to handle every greenhouse crop is therefore either using a very general model that performs poorly on all of them, or has quietly built depth for two and marketing for the rest. Naming three is more honest and more likely to work.

The three chosen are also the right three commercially. Tomatoes, peppers and cucumbers are the dominant high-tech greenhouse crops by volume in the markets where this kind of facility is concentrated, which means the addressable market is large despite the narrow list.

For a grower outside those three, the implication is straightforward: this is not your product yet, and no amount of general capability compensates for a model that has not learned your crop.

Irrigation Control Is Where Software Touches The Crop

Irrigation Control automates precise water and nutrient delivery, and this is the module that changes from advising to acting.

Everything else in the platform produces information a person then uses. Irrigation control produces an action, applied to a living crop, on a schedule nobody is watching — and that is a categorically different level of trust.

The case for it is strong. Irrigation is the most consequential daily variable in greenhouse growing and the one most sensitive to timing, and automated delivery responding to actual plant and environment data outperforms a fixed schedule set at the start of the week.

The case for caution is equally strong and simpler. A watering system that malfunctions damages a crop within hours, and the recovery time is measured in weeks. This is not a domain where a software failure is an inconvenience.

The questions to ask are therefore operational rather than technical: what happens when the connection drops, what the safe default is, who is alerted and how fast, and whether a grower can override immediately from wherever they are. A vendor with good answers has thought about it; one that has not is asking you to find out.

The same discipline appears wherever measurement precedes improvement. An observability platform like Langfuse makes the identical argument about software: you cannot improve what you have not instrumented.

• Confirm the safe default when connectivity or the platform fails.
• Establish who is alerted, how, and how quickly, before automation is enabled.
• Run forecasting alongside your existing method for a full crop cycle before trusting it.
• Judge a forecast on error narrowing over a season, not on any single week.
• Check the retirement notice on Source Plant App if your workflow depends on it.

Reading The Grower List

Source.ag names Harvest House, Thanet Earth, Rainbow International, Agrocare, Mularski Group, Royal Pride, Greenbalance Bryte and Breugem Horticulture among its customers.

That list is more meaningful than a customer count would be, and specifically because of who is on it. These are substantial commercial growing operations rather than pilot sites, and horticulture is an industry where reputation travels fast and nobody adopts software casually.

The geographic pattern is also informative. Several are Dutch operations and Thanet Earth is British, which places the product squarely in the Northern European high-tech greenhouse cluster — the most technically advanced concentration of this kind of growing anywhere.

That matters for a prospective customer outside that region. Software developed against Dutch glasshouse conditions carries assumptions about climate, light levels, structure and growing practice, and those assumptions may not transfer cleanly to a different latitude or a different style of facility.

The Source Plant App is listed as retiring in September 2026, which is worth noting rather than reading anything into. Products being retired is normal; a grower whose workflow depends on one should confirm what replaces it and when.

Piloting Greenhouse Software Across A Season

Agricultural software cannot be evaluated in a fortnight, because the feedback loop is the length of a crop cycle.

The only honest test is running it alongside what you do now for long enough to compare.

The commercial half of a growing business sits outside all of this — a ledger platform such as Puzzle handles the accounting that a good harvest eventually turns into.

1. Start with forecasting only, in parallel with your existing method, and change nothing about how you grow.
2. Record both forecasts weekly against actual harvest, so you have a comparison rather than an impression.
3. Judge on whether the error narrows across the season, not on whether any single week was closer.
4. Add the workspace and dashboards second, once forecasting has earned some confidence.
5. Introduce irrigation control last, on one compartment rather than the whole facility, with alerting configured first.
6. Confirm the failure behaviour by testing it deliberately, rather than discovering it during a connectivity outage.

Source.ag: Greenhouses, Measured

• Harvest Forecast predicts yield with dynamic, data-driven adjustment — the number the whole commercial operation depends on.
• Irrigation Control automates water and nutrient delivery, which is where software touches the crop directly.
• Source Workspace centralises real-time data and KPI dashboards across the operation.
• Three crops: tomatoes, bell peppers and cucumbers, which is a deliberate limit rather than a gap.
• Named growers including Harvest House, Thanet Earth, Agrocare and Royal Pride, with Source APIs and cloud modules for integration.