This gadget uses AI to listen to plants and translate what they say - Fast Company

Open original

Electric Plant, a startup, has built a small sensor that picks up the electrical signals plants emit and uses AI to interpret them as data on hydration and stress. The company's CEO, Justin Mast, describes the device as a kind of dongle.

  • Plants produce electricity, and their electrical signals shift in response to environmental stimuli.
  • The sensor decodes those signals to gauge conditions like hydration and stress.
  • The device is pitched as a simple add-on rather than a complex monitoring system.

I can't access the full article from the URL, and the excerpt you've pasted is only two sentences, so I can only report what those sentences say. Plants generate small electrical signals, and the company behind this device, Electric Plant, claims its sensor can track and decode those signals in response to things like light, water, and temperature. From those readings, it says it can estimate hydration and stress levels in the plant. The one quote here comes from cofounder and CEO Justin Mast, who describes the product as "kind of a dongle." That's a modest framing: it suggests an add-on accessory rather than a standalone gardening system or a finished smart-home product. Beyond that, the excerpt says nothing about how the AI works, how accurate the readings are, what plant species were tested, what the device costs, or whether it's available yet. For a read-it-later decision, the key question is whether the claims have evidence behind them. Plant electrophysiology is a real research field, but turning raw electrical signals into reliable hydration or stress diagnoses is a much bigger claim. If the full article includes test results, peer review, or independent comparisons, that's what would make it worth reading closely. If it's mostly founder interviews and product description, treat the "listening to plants" framing as marketing until proven otherwise.

Ask about this article

is this commercially available? could i build one if not?
Short answer: the Bud sensor from the article is for sale in some form, but the pricing and availability details aren't in what I found. You can build a basic version, though not the AI part without a lot of data. **Is it commercially available?** Yes, at least in the sense that the company is selling it. The article describes Bud, from Electric Plant Co., is a lightweight, low-cost handheld sensor that attaches to plants through two probes. The TechCrunch piece says the translation model, Bombadil, was trained on those devices across about 125 different species. The company claims its sensors are also significantly cheaper than existing tools, which generally cost thousands of dollars. I didn't find a public price or a shop page in these results, so check Electric Plant's site directly for ordering details. Other options exist if you want something you can buy now: - [PlantSonics](https://plantsonics.com/) sells self-contained devices. Its Bamboo M is listed at $459 and reads your plant's electrical signals and plays them as sound — no app, no subscription. It turns signals into music rather than data, so it's more of a novelty than a diagnostic tool. - [PlantWave](https://plantwave.com/) is a sensor that connects to our free app for iOS and Android. - [Vivent Biosignals](https://vivent-biosignals.com/plant-driven-cultivation/) targets growers. It says its sensors use electrodes placed directly on the plant and transfer data directly to our intuitive live dashboard. It's aimed at commercial agriculture, so expect a quote-based sale rather than a retail price. **Could you build one?** You can build a hardware version of the sensing side. The hard part is the AI. The sensing principle is simple. Electric Plant's approach uses electrodes that measure tiny voltage changes. Electrode-based plant monitors are a known DIY project, and Pl
i think id be able to use the PlantSonics technology but pipe it into AI/ML models instead of music. can you find more on that? links instead of full explainations are fine
Your idea is well-trodden. PlantSonics handles the sonification side, but the research community has been doing the ML side for years, and some of the datasets are public. Here's where to look. **Starting points with code and data** - [plant-biosignal-classification](https://github.com/jeremyrenaudin/plant-biosignal-classification) is a GitHub notebook that classifies plant electrophysiology signal quality with KNN, SVM, random forest, gradient boosting, and AdaBoost. It was built on 48 plant traces resampled to 1 Hz, and the data is unlabeled, which makes it a good sandbox for your own labeling work. - The [Zenodo repository for "When Plants Respond"](https://zenodo.org/records/15095523) holds about 444 MB of data from plants outdoors over five months, plus feature selection files. The authors report macro F1-scores of up to 95% in binary tasks. - The [human eurythmy dataset on figshare](https://figshare.com/articles/dataset/Machine_Learning_Detection_of_Plant_Bioelectric_Responses_to_Human_Eurythmic_Gestures_/30227083?file=58324288) comes from a paper that recorded bioelectric activity from 2978 plant samples across three species (basil, salad, tomato) using differential electrode pairs (leaf and soil electrodes) sampling at 142 Hz. That setup is close to what a DIY sensor would produce. **Papers worth reading for method** - [When Plants Respond (arXiv 2506.23872)](https://arxiv.org/pdf/2506.23872) covers electrophysiology and ML for monitoring systems, and it's the paper behind the Zenodo data. - [Early Detection of Water Stress by Plant Electrophysiology (arXiv 2604.28038)](https://arxiv.org/pdf/2604.28038) describes a fully convolutional network deployed directly on a microcontroller, the PhytoNode. That's the closest thing to an edge AI version of your idea. It also makes a useful point: discrete predictions, such as 'healthy' or 'stressed', alone are insufficient for practical decision-making. - [Automated Phytosensing: Ozone Ex
i have a pi 2b and an arduino. are those good enough?
A Raspberry Pi 2 Model B and an Arduino can handle the sensing side of a DIY plant monitor, but not the AI side. The Pi 2B is the weak link if you want to do anything with machine learning on the device. **What the hardware can do** The sensing principle is simple: electrodes measure tiny voltage changes on the plant. An Arduino is well suited to this. It has analog inputs to read the signal, and it's what you'd use for the high-impedance amplification and sampling. The Pi 2B can run a Linux script that pulls readings from the Arduino over serial and logs or graphs them. That setup works fine for collecting data. **Where the Pi 2B falls short** The Pi 2B has a 900 MHz quad-core ARM Cortex-A7 and 1 GB of RAM. It can run small models through TensorFlow Lite or similar, but it's slow and it won't handle anything large. Training a model is out of the question on it. You'd train on a laptop or a cloud service and only run inference on the Pi. The bigger obstacle is data. Bombadil, the model the TechCrunch piece describes, was trained on readings from devices across about 125 species. A model that translates plant signals into meaningful conditions like drought stress needs labeled data from your own plants under known conditions. You'd have to collect that yourself, over weeks, and the results would only apply to the species and setup you trained on. **What a practical build looks like** - Arduino with a high-impedance instrumentation amplifier (an INA128 or AD620 is a common choice) to read the electrode signal. - Two stainless steel or silver-plated electrodes, one inserted in the stem or soil and one on a leaf or the pot. - Pi 2B logging readings and running a basic threshold or simple classifier. - Skip the AI at first. Start with a rule like "signal changes sharply when watered," then add a model only once you have labeled data. Projects like [OpenGrow](https://www.opengrow.org/) and the wider Arduino plant-monitoring community have documented electrode-based setups you can copy. Check the [Raspberry Pi 2 Model B specs](https://www.raspberrypi.com/products/raspberry-pi-2-model-b/) if you want to confirm the hardware details before you buy anything. **Bottom line:** Good enough for a data logger and a basic sensor. Not good enough for the AI translation the
good to know. lets just zero in on data collection. arduino or rasperrry pi 2b? i feel like the pi will be easier to administer and i dont have any other use for it
Short answer: go with the Pi 2B for the data logging, with the Arduino kept as the analog front end. The Pi's ease of administration is the deciding factor, and it's a sound reason. Here's the split that makes sense for data collection: **Why the Pi wins for logging** The Pi runs a full Linux system, so you can schedule collection with cron, store readings in a SQLite database or CSV, and pull data over SSH or Wi-Fi without touching the device. It can also serve a simple graph or push data to a dashboard. An Arduino on its own has no storage beyond a small SD card shield and no network stack without extra hardware, so it's a poor fit for running unattended for weeks. **Why you still want the Arduino** The Pi 2B has no analog-to-digital converter. You can't read the electrode voltage directly from its GPIO pins. Plant signals are in the microvolt-to-millivolt range, so you need a high-impedance amplifier and a proper ADC in front of it. The Arduino's analog inputs (or an external ADC like an ADS1115 connected to the Pi over I2C, which avoids the Arduino entirely) solve that. If you want to drop the Arduino, add an [ADS1115 16-bit ADC](https://www.ti.com/product/ADS1115) to the Pi over I2C. That keeps everything on one board and makes the setup even simpler to manage. The [Raspberry Pi GPIO documentation](https://www.raspberrypi.com/documentation/computers/raspberry-pi.html) covers the I2C wiring. Use a Python script with the `smbus2` or `adafruit-circuitpython-ads1x15` library to read the values and log them. **Practical recommendation** 1. Start with the Pi 2B plus an ADS1115 and your amplifier circuit. One board, one OS, easy remote access. 2. Use the Arduino only if you find the ADS1115 too noisy or want the amplifier's analog reading done more cleanly there. 3. Log timestamped readings every few seconds to a file, and note the watering times and conditions alongside them. That labeled log is the data you'll need later if you ever train a model. The Pi 2B is slow, but data logging needs almost no compute, so its limits won't matter here.
lets go one board. whats the cost for the ADS1115? < $3?
Yes, the ADS1115 is under $3, but only if you buy from marketplace sellers and accept shipping delays. Most reputable retailers charge more. **Budget options (under $3 per unit, plus shipping):** - AliExpress listings for a bare ADS1115 4-channel module show $0.33 and $1.09 prices in the results I found. Some are sold without header pins, so check the listing before you order. - The catch is shipping, which often runs weeks, and the clone boards vary in quality. They're usually fine for logging, but buy a spare if you can. **Established retailers (more than $3):** - [ShillehTek](https://shillehtek.com/products/shillehtek-ads-1115-pre-soldered) sells a pre-soldered module at $9.75 USD. - [ProtoSupplies](https://protosupplies.com/product/ads1115-4-channel-16-bit-adc-module/) lists a 4-channel 16-bit ADC module at $3.95, with stock on hand. - [DFRobot](https://www.dfrobot.com/product-1730.html) sells a Gravity-format version at $8.90, which is handy if you want a plug-in connector instead of soldering. - [Adafruit](https://www.adafruit.com/product/1085) lists its version at $14.95, with a breakout that includes a STEMMA QT/Qwiic connector and an established support community. For a single unit you'll actually use, the ProtoSupplies board at about $4 is the best balance of price and reliability. If you're comfortable waiting, an AliExpress board at around $1 to $2 is fine for this project. Either way, you'll also need a few jumper wires and a Pi 2B I2C connection on pins 3 and 5 (SDA and SCL).
retailers is fine for now. pick the best one. the prices dont bother me at this scale. i want the least headache at this stage. we can zero in later
Based on the conversation so far, the ProtoSupplies board is the pick. It's $3.95, it's in stock, and it ships from an established retailer, so you avoid the weeks-long AliExpress wait and the quality lottery. The ADS1115 itself is a sound choice for this job. It's a 16-bit converter, which matters because plant electrode signals are tiny, and it talks to the Pi over I2C on pins 3 and 5 as the earlier answer noted. Since you want minimal hassle, a few things are worth confirming before you buy: - **Header pins:** Get a version with headers already soldered, or you'll be soldering a board you're trying to avoid fiddling with. The ProtoSupplies listing is the one I'd check first for this. - **Jumper wires:** Buy a female-to-female set for the Pi's GPIO header so you don't have to solder anything. - **Enable I2C on the Pi:** Run `sudo raspi-config`, go to Interface Options, and turn on I2C. Then run `i2cdetect -y 1` and you should see the device at address 0x48. If you don't, it's a wiring or enablement problem, not a bad board. - **Library:** The [Adafruit CircuitPython ADS1x15 library](https://github.com/adafruit/Adafruit_CircuitPython_ADS1x15) is well documented and works on the Pi, which makes it the lowest-friction route to first readings. If the ProtoSupplies listing turns out to be out of stock or lacking headers when you check, the [DFRobot Gravity board](https://www.dfrobot.com/product-1730.html) at $8.90 is the fallback. Its plug-in connector means no soldering at all, which is the trade you're paying for.
anything else before i sent this over to thirdplace?
Before sending this to thirdplace, check these things. The list is about the build itself, since the article covers the Bud sensor and the AI model but not your DIY plan. **Electrode safety and signal quality** Your Pi 2B and the ADS1115 are both 3.3V logic devices, and the ADS1115's inputs must stay within its supply range. Put a protection circuit or series resistor on any electrode that touches a live plant or powered equipment, and never power the plant side from mains. Also confirm the amplifier is isolated or battery-powered if you're using a mains-powered Pi near the plant. **Input range and gain** The ADS1115 has a programmable gain amplifier, so set the gain to match the expected signal. Plant signals in the microvolt-to-millivolt range need a high-gain instrumentation amplifier upstream, because the ADC alone won't resolve them. Expect noise from Wi-Fi, the Pi's power supply, and nearby mains wiring. Shielded cable and a common ground between amplifier and Pi usually fix most of it. **Data logging basics** Set a fixed sample rate and timestamp every reading in UTC. Store the raw voltage, not just a converted value, so you can reprocess later if your calibration changes. Log watering, light, and temperature in the same file, because the labeled log is what you'll need if you ever train a model. **Pi setup** Use a current Raspberry Pi OS image. Pi 2B support for current OS releases is limited, and the Pi 2B has no built-in Wi-Fi, so you'll need a USB Wi-Fi adapter or an Ethernet connection for remote administration. Confirm your OS version still supports the Pi 2B before you flash the card. **Things to confirm with thirdplace** - Whether they've built electrode-based monitors before, and what they'd recommend for electrode material and placement. - Their preferred amplifier chip, since the INA128 and AD620 are both common but have different noise characteristics. - Whether they want the Pi as the only processing unit or plan to offload to a laptop. If you send me the thirdplace question or their response, I can check the specifics against current documentation.
thirdplace is just an ai harness. the same as this, but set up for development. nothing materially changes but the ability to create files and run software etc
Thirdplace is an AI harness, so the main question is whether it changes the plan. For this project, mostly it doesn't. The sensing hardware, the Pi, the ADS1115, and the logging design all stay the same. What changes is where you run code, where you write files, and how much of the Pi-side work you can hand off. **What stays the same** The electrode measurement chain still has to be built and wired by you. An AI harness can write the Python logging script, but it can't put electrodes in a plant or dial in the amplifier gain. The hardware constraints from earlier apply regardless: the Pi 2B reads voltage through the ADS1115, and the signal conditioning happens upstream of it. **What a dev harness adds** Because you can create files and run software, a harness can: - Write and test the logging script before you connect it to the Pi, using simulated voltage data. - Generate the SQLite schema or CSV format and the cron job that runs collection. - Set up a small plotting or dashboard script that reads the log. - Draft a labeling workflow so your watering and condition notes land in the same file as the readings. Thirdplace may not be able to reach the Pi directly, though. Check whether it can SSH into your device or whether you'll be copying files across manually. That determines how much of the deployment it can actually do. **What to watch** The Pi 2B is still the constraint. Keep the code light enough for its 1 GB of RAM, and avoid anything that assumes a newer OS or Wi-Fi hardware it doesn't have. Ask thirdplace to target the Raspberry Pi OS version you've confirmed works on the 2B, and to use the [Adafruit CircuitPython ADS1x15 library](https://github.com/adafruit/Adafruit_CircuitPython_ADS1x15) rather than inventing its own I2C driver. If you give thirdplace the logging requirements from earlier (UTC timestamps, raw voltage stored, watering and temperature in the same file, fixed sample rate), that's a complete enough spec for it to start writing code.
it can. if you have nothign to add im going to move this whole conversation over as a handoff now
Yes, you have nothing else to add, but there's one thing worth flagging before you hand this off: the Pi 2B's I2C and the ADS1115 are fine, but the Raspberry Pi OS support point is the one most likely to cost you time, so put it first in the handoff. Here's what I'd pass along as the handoff summary: **Hardware** - Pi 2B as the logger, with the ADS1115 on I2C pins 3 and 5 (address 0x48), per the earlier answer. - A high-gain instrumentation amplifier upstream of the ADC. The ADS1115 alone won't resolve microvolt-to-millivolt electrode signals. - A USB Wi-Fi adapter or Ethernet, since the Pi 2B has no built-in Wi-Fi. **Software constraints** - Target a Raspberry Pi OS release that still supports the Pi 2B. Current images may not, so verify this before flashing. - Use the [Adafruit CircuitPython ADS1x15 library](https://github.com/adafruit/Adafruit_CircuitPython_ADS1x15) instead of a custom I2C driver. - Keep the code light for 1 GB of RAM. **Logging spec** - Fixed sample rate, timestamps in UTC. - Store raw voltage, not only converted values. - Watering, light, and temperature notes in the same file as the readings. **Open questions for the handoff** - Whether thirdplace can SSH into the Pi or whether files need to be copied by hand. - Electrode material and placement, and the amplifier chip choice (INA128 versus AD620). Send it over.