The technology behind your Gardyn

Every Gardyn is a connected device with one or two cameras, a set of sensors and computer vision models that review your plants every night. This post explains how those pieces work together.

Here is the path a single photo takes, from capture to a message in the app and the optimization of your plants’ growth.

Key takeaways

• Gardyn Home has two cameras and Gardyn Studio has one; both capture photos of your plants every 30 minutes.
• Every night, fine-tuned computer vision models process thousands of photos and generate hundreds of results across plants.
• Those results, combined with live sensor data, drive Kelby messages, tasks and your daily Thryve score.
• Kelby also adjusts your Gardyn on its own through germination mode and Smart Watering.
• Ask Kelby answers questions using Gardyn knowledge plus your device data, and cannot change settings on your Gardyn.

Computer vision

Gardyn Home has two cameras and Gardyn Studio has one. They capture photos of the plants every 30 minutes. Each night, the system picks the most useful shots from the last 24 hours and runs them through our computer vision models.

The nightly pipeline

Every night, we process thousands of pictures. They run through fine-tuned models that our team curates and evaluates, and they generate hundreds of results across plants, from a first sprout to a strawberry ready to pick. Those results are what drive Kelby: the messages you receive, the tasks in your app and your Thryve score.

What the models look for

Our models look for a wide range of named conditions, grouped into a few families:

• Growth and harvest: sprouts versus true leaves, thinning, flowering, bolting by plant type and harvest readiness.
• Water and roots: tank level, dry or wet yCubes and parts of the Gardyn that are not getting water.
• Pests: aphids, tracked across several severity levels.
• Device health: dark or washed-out images, a camera still wearing its protective film, offline devices and light levels.
• Biomass and sustainability: computing plant biomass and the sustainability metrics that come from it.

The models are aware of the plants’ cycle: Kelby does not check for harvest readiness on day 3, or for germination on day 80. A new detection is trained on labeled images, checked by our team so it doesn’t send you false alarms, and only reaches your Gardyn once it proves reliable in testing. After launch, our team keeps reviewing results and teaching the models with new examples.

Plant AI since 2021

Gardyn leads plant AI in smart home hydroponics. We took our first steps in 2021 with models that learned to spot one thing at a time, like a sprout or a pest. Since then, we have moved to models that understand the whole image and compare it to many known examples, which lets us teach them new conditions much faster.

Flags

A detection is an observation. A flag is a decision about whether that observation matters. Our flag logic engine reads the vision output alongside live sensor data: water level, pump behavior, temperature and humidity. Some flags fire in real time, the moment your Gardyn reports a change; others come from the nightly vision analysis.

That combination is what makes the system useful instead of noisy. Take wilting. The cameras can see drooping leaves, but the right advice depends on why:

• Wilting and low water: fill the tank.
• Wilting and the pump set to manual: switch watering back to the automatic schedule.
• Wilting and no tank refresh logged in 45 days: refresh the tank.

Each flag weighs the device’s history, its growth stage and earlier flags before it fires. Some lead to a message, some to a task, and some stay internal, like early signs of a pump problem. Our team can reach out by email before your plants feel it, and spot patterns across devices that shape what we improve next. Flags also celebrate. The first sprouts, a successful setup or the first tomato on the vine all trigger their own moments.

The digital twin

Every Gardyn has a digital twin in our cloud. It combines the device’s telemetry (water level, pump activity, temperature and humidity) with its photo history, every flag it has raised and the tasks that have been completed or not.

That twin lets the system reason over time instead of reacting to a single photo. Dry yCubes on day 2 get a check-in; if they are still dry on day 9, our team steps in. Wilting is read differently when the tank has been low all day.

It already knows specific plants, too. Every day, our models look for tomatoes, strawberries and cucumbers as they fruit, and flag them so you know when to pollinate and harvest.

Two jobs for Kelby

Everything above feeds Kelby, and Kelby does two things with it. It lets you know what to do, through messages and tasks. And it optimizes your Gardyn’s settings on its own, so your plants thrive.

What Kelby tells you

Messages

All of this only matters if it reaches you in a way that helps. Kelby speaks in a few different voices, each matched to how sure we are and how urgent it is:

• Celebrate: “You have sprouts, how exciting!” Good news gets as much airtime as problems.
• Nudge: a low tank, open slots inviting algae or plants ready to harvest.
• Intervene: urgent alerts, like wilting plants, flagged clearly so you act fast.
• Ask: when a signal is unclear, Kelby checks in rather than guessing.
• Stay silent: if a signal is weak or only useful to our team, you never see it.

Tasks

When something needs your hands, Kelby turns it into a clear, step-by-step task in the app: thin your sprouts, refresh the tank, harvest your tomatoes. What you complete, and what you leave open, goes back into the digital twin, so the next decision starts from where your Gardyn actually is.

The Thryve score

Alongside messages and tasks, the Thryve score gives you a daily 0 to 100 view of your Gardyn’s care. It combines 10 factors drawn from telemetry, flags and the tasks you complete: tank level and refreshes, watering, light, temperature and humidity, plant food, open slots, sprout trimming, root checks and overgrowth. Each score comes with a short list of the most useful things to do next.

What Kelby does for you

Some decisions don’t need you at all. Kelby, the system, adjusts how your Gardyn runs as your plants grow:

• Germination mode: right after planting, Kelby soaks the seeds, then holds off on water for a few days. It sounds backward. Seeds sprout better.
• Smart Watering: the pump schedule adjusts quietly in the background as plants move from germination to growth to maturity.

The same detections and flags that trigger a message can also change how your Gardyn runs. That is the second half of the loop, and the part you never have to think about.

Ask Kelby

Ask Kelby lets you ask anything in plain language, like “Why are my basil leaves yellow?” or “When should I add plant food?” Before Kelby answers, it pulls the relevant guidance from our own help center and plant guides, then adds what it knows about your device.

That device context is what makes it different from asking a general chatbot. Kelby knows which Gardyn model you have and whether it is online, its latest readings like water level, temperature and humidity, the features on your account, the messages Kelby has sent you, where your tasks stand and your Thryve score. The answer is grounded in Gardyn’s knowledge and shaped for your Gardyn.

For an AI assistant in your home, accuracy matters as much as security. That is why Ask Kelby does not have access to everything yet. Every source it draws on goes through a curation process first: we review the content for accuracy, decide what data is safe and useful to share, and only then open it up. We would rather Kelby give a narrower answer that is right than a broader one that is wrong.

We are now working to bring everything the vision pipeline knows into Ask Kelby: the nightly detections, the flags and the plant history behind your messages, tasks and Thryve score. Each piece goes through the same curation process before Ask Kelby can use it.

One more distinction matters. Kelby, the system, already takes action on your behalf: germination mode and Smart Watering, for example. Ask Kelby, the chat assistant, does not. It answers and guides, but it cannot change anything on your Gardyn. Letting a chat assistant act on a device in your home on its own is not something we consider safe yet, so we have chosen not to ship it.

How it all fits together

The pieces run on two rhythms. In real time, Kelby reacts the moment your Gardyn reports a change, like the water level or the pump. Every night, the vision models review the day’s photos. Together they form one loop:

1. See: the cameras capture your Gardyn every 30 minutes.
2. Understand: vision models scan thousands of photos each night for a wide range of conditions.
3. Decide: flags combine what we see with sensor data, history and your tasks.
4. Act: germination mode and Smart Watering adjust your Gardyn automatically.
5. Speak: Kelby sends a message, creates a task, updates your Thryve score or stays quiet.
6. Answer: Ask Kelby is there whenever you have a question.

Thank you to every Gardyner who has grown with us, and to the team that builds this every day.

Eric, CTO

Frequently Asked Questions

How does Gardyn use AI to monitor plants?

Gardyn Home has two cameras and Gardyn Studio has one. They photograph your plants every 30 minutes, and each night computer vision models review the most useful shots for conditions like sprouting, flowering, harvest readiness, water issues and pests.

What is Kelby?

Kelby is Gardyn’s AI system. It turns what the cameras and sensors detect into messages and tasks in the app, and it adjusts your Gardyn on its own through features like germination mode and Smart Watering.

What is the Thryve score?

The Thryve score is a daily 0 to 100 view of your Gardyn’s care. It combines 10 factors, including tank level, watering, light, plant food and open slots, and comes with a short list of the most useful things to do next.

What is the difference between Kelby and Ask Kelby?

Kelby, the system, takes action on your behalf, such as adjusting the watering schedule as plants grow. Ask Kelby is the chat assistant: it answers questions using Gardyn’s help center, plant guides and your device data, but it cannot change anything on your Gardyn.

Does Kelby send an alert for everything it detects?

No. Each detection is weighed against sensor data, growth stage and device history before it becomes a flag. Some flags lead to a message or task, some stay internal for the Gardyn team, and weak signals are never shown to you.