Open layer in greenhouse / controlled-environment automation — Mycodo, WUR AGC datasets, Common Greenhouse Ontology, OpenAg, and the Priva interface-open reality

international (Netherlands, US, Canada, Bhutan)

Content

The protected-cultivation cell of this field guide is populated almost entirely by closed-commercial or research-stage actors (Haven Greens, ioCrops, Sollum, Hoogendoorn, Vineland, Tencent×WUR). This unit records the open layer that exists beneath and across that cell, verified 2026-08-06 through the opensource-agrifood G-OSA-22 cycle (primary sources cited below; see Links).

Headline: the open CEA layer exists as three partial shapes — open control software at small/maker scale, open cultivation data, and an open but unadopted data standard — while the flagship “all-open” claim failed, and vendor “open” platforms are interface-open, licence-closed.

Shape 1 — Open control software, small/maker scale (Mycodo)

Mycodo (GPL-3.0; kizniche/Mycodo; 3,267 stars; v8.17.0 released 2026-08-03) is an environmental monitoring and regulation system on commodity single-board computers — the closest thing to a general-purpose open “grow computer.” Data stays local by default. Peer-reviewed deployment: Penjor et al. (2022), Bhutan Journal of Agriculture, DOI 10.55925/btagr.22.5108 — a Mycodo-based automation prototype ran 8+ months in real hydroponics structures (NFT, DWC, vertical tower) at ARDC-Wengkhar. Scale ceiling: no commercial-scale CEA deployment documented.

Shape 2 — Open cultivation data (WUR Autonomous Greenhouse Challenge)

The field guide’s “research-stage” note on the WUR/Tencent Autonomous Greenhouse Challenge upgrades with primary evidence: the challenge does release its data openly, across four editions (2018, 2019, 2021, 2024) under persistent DOIs. 4th edition “Dwarf Tomato Timeseries and Images” (DOI 10.4121/fa102772-32db-4b30-bace-12f2016722ce; made available 2025-06-23): 5-minute climate/control-state/weather data, yield measurements, and canopy RGBD imagery from six AI-controlled 96 m² compartments (Bleiswijk). Creators include Silke Hemming (WUR) — the corpus’s first named WUR greenhouse-technology researcher. Open data, not open code or hardware.

Shape 3 — Open data standard, unadopted (Common Greenhouse Ontology)

The Common Greenhouse Ontology (CGO; Apache-2.0; vekadvies/common-greenhouse-ontology, TNO-led) is an ontology of greenhouse components, properties, and measurements, positioned by TNO as the communication standard for greenhouse robotics integration. It is the constructive answer to the closed-data-format lock-in problem — but the reference repo has 0 stars, last pushed 2022-12-06, and no commercial climate-computer vendor is documented as implementing it.

Shape 4 — The failed flagship (MIT OpenAg Food Computer)

The most ambitious “everything open” CEA claim. OpenAgricultureFoundation/openag_brain (GPL-3.0) archived 2018; gro-hardware carries no verifiable licence and was archived 2019; the project closed April 2020; Food Computers were never commercially available. The results record is publicly discredited (NYT 2019; Gizmodo “Theranos for plants”; IEEE Spectrum on refugee-touting of non-functional units; ProPublica). “Would have been open source” is aspiration, not an open layer.

Shape 5 — Vendor “open” reality (Priva, Hoogendoorn, Signify)

Priva’s “open platform” (primary source, priva.com): API + Priva Horticulture Developer Portal, gated by a Priva Gateway + Priva Connected subscription, with data shared through an approved Priva Integration Partners ecosystem (Blue Radix, Source.ag, IUNU). No open-source licence disclosed. CEO Hans Peters (GreenTech Amsterdam 2024): “We want to become the App Store of horticulture.” Data-ownership claims (“all data remains yours”) are vendor-stated, not licence-verified. Hoogendoorn (IIVO + LetsGrow.com; “open software platform” usage in a 2020 industry overview) and Signify (GrowWise) surfaced no open programme. Verdict: interface-open, licence-closed — open system ≠ open licence.

What this unit is doing in the taxonomy

Anchors the openness axis of the on-farm-production-protected cell, which the field guide otherwise documents as vendor-closed. Distinguishes from:

• Haven Greens — closed-loop vendor deployment (lettuce, King City ON).
• ioCrops — closed FTVC™ greenhouse-automation stack (Korea).
• Sollum — closed sun-as-a-service lighting subscription (Montréal).
• Hoogendoorn — closed IIVO climate computer + LetsGrow.com data layer (Netherlands/Quebec).
• Vineland — pilot robotic greenhouse harvester, closed hardware.

Why it matters for talks

• When a talk contrasts “open” vs “closed” agrifood AI, the greenhouse cell now has verified open-layer content: GPL control software exists but stops at small scale; open data exists (WUR AGC); open standards exist but are unadopted. The openness thesis holds for data and hobby control and fails at commercial deployment and data-format interoperability.
• The “App Store of horticulture” quote is the enclosure-by-integration counterpoint to open-standards arguments — worth pairing with the open-interface ≠ open-licence distinction.
• OpenAg is the cautionary case for open-agrifood credibility: open claims without delivery damage the field.
• The Bhutan ARDC deployment shows the open layer’s Global South reach is research-led institutional use, not farmer-led licensed hardware — consistent with the G-OSA-15 finding.

Critical context

• Mycodo deployment evidence is maker/research scale; no commercial-scale CEA deployment documented.
• WUR AGC datasets are open data, not open code/hardware; the algorithms and greenhouse hardware stay with teams/vendors.
• CGO is an open standard with zero verified vendor adoption.
• Priva’s data-ownership claims are vendor-stated, not licence or contract verified (G-OSA-16-adjacent).
• The OpenAg results record is discredited; treat its “open ecosystem” framing as contested.