Yield that feeds security
Every feature ties back to more reliable harvests and traceable care - so communities are less fragile when supply chains wobble.
ABG-V1 reference · Autonomous AI 1.0 · operator dashboard V1
Fresh yield.
Field-ready AI.
From model to leaf.
No Plant Is Average (AI for food security).
yieldAI is for growers who want more harvest per square metre, stronger food security, and less mystery in the tunnel. We bridge digital intelligence to physical care - cameras and soil probes to irrigation, snips, and harvest - using ordinary benches, retrofit X / Y / Z gantries, and table hubs you can reach with a wrench. Prefer refurbished workstations, longer-lived parts, and materials chosen for repair? So do we: the product is meant to be affordable to adopt and transparent to audit, with inference on your LAN so your crop data stays yours.
Food security & traceable yield
AI → soil & stem (not just chat)
Refurb-friendly · partner-ready docs
Local-first · fair access to automation
MCP tools · Postgres · MQTT · Pi Flora soil hub · telemetry ingest · operator dashboard (LAN V1) · AI assistant · ABG-V1 reference rig
Reference rig · Cartesian bridge + 6-DOF arm over crop
No Plant Is Average (AI for food security).
Intelligence belongs in the glasshouse, not only in a slide deck.
We design for fair access to serious automation: local data, repairable hardware, and agronomy you can audit, so more communities can grow dependable fresh food close to where it is eaten.
Mission
Grow more food, waste less, and keep humans in the loop
Climate pressure and urban demand mean we need fresh, local yield without burning the planet on freight - or on endless new silicon. yieldAI is a commercial precision cultivation platform that uses AI where it helps (vision, planning, suggestions) and physical automation where it counts (water, light, motion, harvest logging) - always with records you can show a regulator, a neighbour, or your future self.
The reference path ABG-V1 proves the loop on a real bench: RTSP, BLE soil, PostgreSQL, MQTT motion, MCP-callable tools, and optional on-LAN models so small farms and R&D labs can experiment without renting someone else’s cloud brain by default.
What we stand for
Nature-first engineering, priced for farms that need it
AI → physical world
Models meet real leaves, drippers, and gantries. We design for the wet edge - not dashboards that stop at a browser tab.
Sustainable & circular hardware ethos
Favour repairable prints, refurbished PCs for the mainframe where it makes sense, and routes that stay affordable - so good automation isn’t only for the wealthiest glasshouses.
Growers stay in charge
Local data by default, clear audit trails, no mandatory cloud brain - you see what ran, when, and why. Commercial support when you want it; your narrative stays on your farm.
Closed loop
See · feel · act · log
This is the AI-to-physical-world loop: cameras and models see, probes and bench nodes feel, gantry and pumps act, and Postgres logs - so software suggestions always meet something you can touch, weigh, and eat.
01 · Sense
Perceive
RTSP frames, soil and ambient telemetry, reservoir and climate signals - ingested with scope: plant, bed, or whole surface.
02 · Interpret
Decide
The AI interprets the bench through images, labels, and phone sensor telemetry on the LAN, plus on-LAN models and care tooling. Bench phones target on-device vision with clean and labeled feeds to AI; fixed cameras use mainframe inference. Full vision worker for non-phone RTSP is roadmap. Optional LLM calls the same MCP tools a developer would - no shadow APIs.
03 · Act
Execute
MQTT motion bridges and irrigation manifolds today; treatments, snip, and harvest actuation stay gated by automation mode, human confirmation, and bench qualification before live moves.
04 · Log
Prove
Append-friendly events, job records, yields - powering both compliance narratives and tomorrow's better model (orchestration matures per release).
Field reality
Real tunnels, real timber, real constraints
We design for run-of-the-mill houses and standard plant-bed tables - places where sustainable choices (local timber, reused extrusion, hand-built hubs) matter as much as CAD. The gantry retrofits to your structure: full X, Y, and Z travel over the bed so the arm and end effector reach the whole working volume - not a single-axis toy bolted to one side.
The mainframe lives in a cool, dry room - office, equipment closet, or shed nook - while heat, humidity, and spray stay at the bench. At the table, hubs aggregate sensors, host the table brain the gantry plugs into, and run water, medicine, mechanical cuts, and harvest sequences from commands the mainframe issues over the LAN.
Topology
One database · two latencies
Heavy inference and long-horizon records sit on the mainframe; millisecond-class I/O stays at the edge. Agents and dashboard both read the same truth - Postgres and job semantics do not diverge by UI.
cool · dry
Mainframe
Ubuntu, PostgreSQL, MCP server, telemetry and motion bridges, optional on-LAN LLM, and the yieldAI Agent gateway. Heavy vision on fixed LAN cameras remains roadmap; bench phones already stream and label on-device.
wet edge
Table hub & gantry
X/Y/Z mechanics, arm, sensor fan-in, pump and valve drivers. Executes only what passed policy and mode gates upstream.
crop layer
Beds, slots, plants
The digital twin is not magic - it is the merged view of telemetry, media, care plans, and events you already store. The dashboard makes it legible.
Architecture page → MCP, MQTT motion, job records (orchestration matures per release), and safety modes.
Software
What the product includes
Mainframe & MCP
A stable tool surface for motion, environment, locations, vision, and plants - so any MCP-capable agent on the LAN can reason with real IDs, not hallucinated coordinates.
Table hub & gantry path
Three-axis Cartesian volume over the table, carriage and arm
interface, hub-resident real-time control for irrigation and tooling
tied to the same cmd_id and job records as the UI.
Data plane
PostgreSQL as system of record, MQTT for motion and telemetry bridges, phone RTSP with on-device labels, and a growing fused table scene for the operator dashboard — same ledger as the agent.
Product today
What is shipping in the lab
yieldAI is an integrated bench-scale platform, not a slide deck. The public site stays high-level; engineering contracts and tuning numbers ship under commercial terms. Below is the safe summary aligned with our product catalog as of mid-2026.
Shipped
Autonomous AI 1.0
Local agent gateway with MCP-grounded tools, OpenClaw skills, and on-LAN LLM inference — one honest loop from question to bench data.
Shipped
Operator dashboard V1
LAN greenhouse console: digital twin views, AI assistant with evidence, jobs and tasks, admin settings, and safety mode gates for humans and agents.
In progress
Bench sensing & vision
Android phone camera agent with on-device labels and streaming; Raspberry Pi Flora BLE soil hub with health visibility; telemetry ingest into the same ledger as the dashboard.
In progress
Jobs & orchestration
Job records, human physical tasks, and investigation runners — with deeper motion runners and live gantry moves still gated behind bench qualification.
In progress
Mechanical gantry (ABG-V1)
Retrofit X/Y/Z volume over standard benches, table hubs for irrigation and tooling, and a reference mechanical build proving the wet-edge path.
Planned
Production vision fusion
Multi-camera table scenes and production-grade inference on fixed LAN streams — parallel to the phone path already grounding the assistant.
RVO / WBSO reviewers: a password-protected R&D project tracker lists logged hours, milestones, and capability status for grant reconciliation (no engineering secrets).
July 2026 · release 68
What's new in the lab
This lab update extends honest remote commissioning: dashboard cards and the AI assistant now show the same quality-gate evidence when you work away from the greenhouse LAN — so remote operators see pass counts and thresholds from the latest scheduled lab run without conflicting or invented status.
Honest gate counts off-site
Pi sensor and bench phone commissioning cards now show smoke and pytest pass counts from the latest scheduled lab run when live hub metrics are unavailable off-site — so remote operators see clear quality evidence instead of blank or misleading status.
Cross-lane gate visibility
Sensor and phone quality gates appear together on commissioning cards and hourly maintenance rollups — using the same vocabulary as dedicated cron views when you review the bench remotely.
AI assistant gate alignment
The dashboard AI assistant cites quality-gate evidence that matches what operators see on commissioning cards — so off-site questions about sensor or phone readiness get answers grounded in the same scheduled lab run, not invented live bench state.
Soil hub remote context
Soil hub health and quality-gate context stay visible for remote agronomy reviews when live probe metrics are unreachable — falling back to the latest honest lab snapshot instead of going blank off-LAN.
Operators
A calm, fresh view of the living bench
The greenhouse dashboard (first-party operator UI on your LAN) shares the same truth as the agent: a plant's digital twin, vision timelines, a chat-first AI assistant with evidence-backed replies, and an audit trail for every pour, cut, or move — so sustainability and food-safety stories stay grounded in data, not vibes.
Lab capture
Lab capture
Lab capture
System blueprint
Lab capture
Lab capture
See it move
Local food, local AI
A short product walkthrough: planning, sensing, and actuation on every bench — without mandatory cloud inference.
Watch on architecture page →Twin view
One scroll through identity, stage, care plan, linked pathology references, and the numbers that actually drove the last suggestion - not a spreadsheet zoo.
Vision & context
Bench phones stream and label on-device; the assistant can ground replies in live snapshots and sensor context. Fixed-camera fusion for every table remains on the roadmap.
Audit you can export
Motion, chemistry, harvest: parameters, initiator, outcome. Same rows the agent sees; the difference is typography, not truth.
Full operator UI specification ships with commercial and pilot packages - aligned with the same safety and mode gates as the agent.
Creator
Tobie Alberts
Creator & technical founder · yieldAI
Tobie Alberts is the creator and technical founder of yieldAI, building bench-scale precision cultivation from Delft, Netherlands. With nearly two decades leading edge IoT, responsible AI, and large-scale customer intelligence platforms — including CTO roles at Intent HQ and IoT.nxt — he brings enterprise-grade systems discipline to greenhouse automation: local inference, repairable hardware, MCP-grounded agents, and audit trails growers can trust.
yieldAI is his WBSO-backed R&D path to connect AI models to real leaves, drippers, and gantries — not slide decks. He holds certifications in ethical hacking and computer forensics, reflecting a commitment to secure, transparent design from sensor to harvest.
Documentation
Engineering depth, delivered with the product
yieldAI is documented as an integrated platform: motion and telemetry contracts, MCP tool manifests, roadmap vision slices, and the operator dashboard spec are written to the rigour you expect in a regulated facility, and delivered with commercial packages and partner engagement. Field firmware packages ship per engagement.
Field & hardware master
Greenhouse layout, edge devices, environmental boundaries, and how physical I/O maps into the data model.
Implementation architecture
MQTT motion, telemetry ingest, job orchestration, safety modes, database schema, and release slices - single source of truth for integrators.
Operator dashboard spec
Digital twins, media timelines, suggestions, and audited actions - the human layer over the same Postgres ledger as automation.
Partners
Pilots, integrators, and food security partners
yieldAI is a precision cultivation platform delivered as an integrated offering: hardware integration, software, updates, and optional professional services. We work with growers, integrators, and food security programmes that want dependable automation and traceable operations on the bench.
No Plant Is Average (AI for food security) is our compass: partnerships rooted in real harvests, local resilience, and agronomy you can stand behind.
If you are exploring pilots, OEM or integration, or investment, start with the founder profile below or your usual business introduction channel.
Automation that respects the crop, and your business model.
No Plant Is Average (AI for food security)
Sustainable hardware choices, local inference, and traceable yield, packaged as a product you can stand behind in the market.
Talk to us about pilots