What is SmartPlant?

SmartPlant is a Node.js library that treats a plant as something to be measured and asked, rather than something to be guessed about. It reads ordinary environmental sensors, the plant's own electrical signals, and — with an LED used as a probe rather than as a lamp — the answers it gives back.

It is not a plant monitor. It joins a plant and a computer into a functional symbiont — a cyborgplant — with shared perception, memory, reasoning, voice and the ability to act.

Its organising rule is a refusal: nothing is inferred from data that is not there. A missing sensor blocks a conclusion instead of being replaced by a default, a low-confidence estimate loses its authority to change anything, and every refusal names both the reason and the instrument that would lift it.

npm install smartplant

import { createPlant } from 'smartplant'

const plant = await createPlant( {
  name    : 'Ivy',
  species : 'Monstera deliciosa',
  sensor  : 'mock',                  // no hardware needed
  ai      : { provider : 'ollama' }, // or gemini / openai / claude / grok
  memory  : { path : './ivy.json' },
  units   : 'metric',                // or 'imperial'
} )

await plant.read()
console.log( plant.status() )
// 😊 84% | Temperature: 🌡️ 21.6°C | Humidity: 💧 58.9% | Soil moisture: 💧 62.1% | Light: 🌞 326lux

The twenty-three layers

The spine — nothing runs without these

Layer	What it does	Why it matters
🌡 Sensors	Ten pluggable drivers behind one read() contract	Your hardware, or none at all
💾 Memory	Persistent readings, care log, species profile, and what the system worked out about itself	Advice builds on history, and the history survives a restart
🫀 Internal states	Five qualitative estimates of what the plant is doing, not what surrounds it	The system stops seeing only sensor numbers
🔁 Co-adaptation	One door every action goes through, and a record of when its own refusals were wrong	The system can be wrong about itself and find out
🦿 Body	Multirate fusion, reflexes, safety limits, personalization	Autonomy that cannot kill the plant — see Symbiosis

Reading the plant

Layer	What it does	Why it matters
🧬 Signals	Electrophysiology: action and variation potentials, circadian rhythm	The plant's own electrical voice
🪞 Self-reference	Electrome fingerprint, internal clock, two-site coherence, VPD-aware blue	Judged against itself, not a population average
🔬 Spectral	🔵🟢🔴 LED as a probe, not illumination	The plant is interrogated, not just listened to
👁 Vision	Classical phenotyping, ONNX models, PlantCV bridge, thermal canopy	See wilting hours before you notice it

Looking after it

Layer	What it does	Why it matters
💧 Watering	Volume from the pot, fraction from the archetype, duration from a measured pump	A dose it can defend, and a check that the water arrived
🪴 The pot	Whether the container is running out, and moving to a bigger one	The pot is part of the body, and it stops fitting
🔎 Experience	A ledger of what resolved each problem, against the base rate of doing nothing	It reuses what worked, without inventing what did not
🔁 Self-correction	Prediction error, identity drift, decaying priors, response hysteresis	It finds out when it is the thing that is wrong

Knowing whether it works

Layer	What it does	Why it matters
🩻 Diagnosis	40 checks over everything wired up, each ending in something to do	Know it works before walking away
🩺 Self-check	Weekly and monthly reviews, plus a technical inspection of the instrument	It watches its own trajectory, and its own sensors
🖥 Dashboard	The plant in a browser: vitals, activity, and what has no sensor	All of the above, visible without writing code

Talking, and to whom

Layer	What it does	Why it matters
🧠 Knowledge	Ontology, forward-chaining reasoner, semantic recall	Conclusions you can audit, not just trust
🤖 AI	Seven providers, structured output, offline fallback	Your model, your keys, your privacy
🗣 Voice	Five first-person personas, emoji scales, ten languages	It talks to you, it doesn't report at you
🗨 Colony	Conversation between plants, 73 skills, and delivery that survives a link failing	They talk to each other, and only about what they measure
🧬 Inheritance	Directed transfer of validated priors between individuals	A new plant starts with what the last one learned

Once it can move

Layer	What it does	Why it matters
📡 Space	WiFi presence and a lidar scan, without a camera and without a map	It knows if you are in the room, and how far the neighbour really is
🚜 Navigation	Constraints a robot stack cannot know, over a delegated ROS 2 planner	A pot is not a delivery robot

Every layer works alone. They compose.

Requirements

Node.js 18+ and nothing else. npm i smartplant installs zero dependencies — the library imports nothing outside Node itself. chalk and enquirer are optional and only make the CLI prettier; without them it runs in plain text and says so. serialport, mqtt and onnxruntime-node are optional peers loaded on demand; Python and ffmpeg are needed only by the tiers that use them.

---