Yellowing lower leaves
Usually overwatering — roots sitting wet, not thirst.
How Herb reads it: Herb checks your watering log first: three waterings this week says overwatering, none in two weeks says the opposite.
Point, shoot, know
What is this plant — and what’s wrong with it?
One photo names it against a real catalogue. Another tells you why its leaves are yellowing — read against that species’ profile and your plant’s own history. And when we can’t tell, we say so instead of picking something plausible.
your photo
Swiss cheese plant
Monstera deliciosa
Monstera deliciosa
Monstera deliciosaConfident
Monstera adansoniiPossible
Philodendron bipinnatifidumUnlikely
Matched against the catalogue — not a label invented on the spot. Nothing resolves? You get an empty result and an honest “we couldn’t place this”.
drag to compare
What happens in that second
A photo becomes a real record
The interesting part isn’t recognising a leaf — it’s what the name is attached to afterwards.
- Read the frame
Vision returns candidate names with how sure it is — and it’s told that a blurry or partial shot deserves a low number, not a confident one. - Resolve against the catalogue
Each name is matched to a real species record — including synonyms, so an old botanical name still lands on the right plant. - Nothing resolves? You get nothing.
No fallback, no “probably a monstera”. An identification that failed returns empty — because a confident wrong name poisons every care tip that follows it.
reading the frame…
· · ·
We changed how sure we sound, not just how sure we are
We used to show a confidence percentage. We removed it — one number was standing in for five different things, and nobody could tell which. Now the certainty is carried by the sentence itself.
Not sure
UnsureHedgedCertain
Not sure
“I can't place this one confidently.”
You get no name rather than a wrong one — and a nudge to try a clearer shot of a leaf.
Possible
“This looks like a Monstera adansonii, but I'd check the leaf holes.”
Hedged out loud, with the thing you can verify yourself named explicitly.
Confident
“Monstera deliciosa — Swiss cheese plant.”
Stated plainly, matched to a catalogue record, with its full care profile attached.
Why this matters more than accuracy scores: a plant app that’s right 95% of the time is only useful if you can tell which 5% to doubt. Percentages don’t tell you that. Language does.
Most apps only have one branch
When a name can’t be resolved, the tempting thing is to return the closest guess — it looks better in a demo. Here’s what actually happens instead.
A photo comes in
Vision proposes names. Each one is looked up in the catalogue — including synonyms.
A name resolves
You get the plant, and its file
The match carries its real care profile, pet-safety record and verification tier straight into your garden.
One tap and it’s a plant you own, on a watering schedule.
Nothing resolves
You get an honest empty result
No nearest-guess, no demo fallback — even if the request itself failed. You’re told we couldn’t place it, and what kind of photo would help.
Nothing wrong gets written into your garden, so nothing wrong gets computed from it.
Yellow leaves mean six different things
That’s why a symptom lookup isn’t enough. The same photo means something different depending on what you did to the plant last week.
drag to browse
Crispy brown edges
Dry air, underwatering, or salt build-up from feeding.
How Herb reads it: Cross-read against your local humidity and the species' own humidity preference.
Dark spots with halos
Often bacterial or fungal, spread by splashing and still air.
How Herb reads it: Flagged as serious when it's spreading — with the treatment steps, not just a label.
Fine webbing under leaves
Spider mites — they thrive in exactly the warm dry air we sit in.
How Herb reads it: If you've had them before, that history is in the read — recurrence changes the advice.
Sudden drooping
Thirst, shock after a move, or roots that can't breathe.
How Herb reads it: Recent repot or relocation on the timeline? That's the first thing Herb points at.
A diagnosis also records itself. Treatments, recovery and relapses land on the plant’s timeline — so the second time it happens, the app already knows what worked the first time.
Point it at the soil. Point it at the room.
Two more reads that turn vague advice into something about your flat — because “bright indirect light” means nothing until you know what your window actually does.
Soil check
A photo of the pot surface
Dry
top layer · needs water within a day
DrynessDry on top
CompactionSlightly packed
DrainageAdequate
Soil intelligence is a Botanist feature
Light scan
A photo of the corner or window
Bright indirect
the band most houseplants actually want
Light classBright indirect
Direct sunAround 2 hours
Verdict for a MonsteraIdeal spot
Included on every plan
Both readings feed the same engine: a light scan becomes the score for that spot on your home map, and a soil read becomes context the next time you ask why something looks off.
A wrong name doesn’t stay wrong on its own
In most apps an identification is just a label. Here it’s the key everything else is computed from — which is exactly why we refuse to guess it.
Wrong species saved
A fern gets filed as a monstera.
Wrong watering
It inherits a schedule built for the wrong thirst.
Wrong grouping
It joins a water group it should never be in.
Wrong everything else
Placement scores and shop suggestions all compute from that name.
So the honest failure is the cheap one
“We couldn’t place this — try a photo of a single leaf against a plain background” costs you thirty seconds. A confident wrong answer costs you a plant, and you won’t know why for a month.
Every record tells you how much to trust it
A catalogue is easy to inflate — copy a taxonomy list and claim hundreds of thousands of plants. We publish the whole backbone and the grade, so a name-only record can never masquerade as a care profile.
S
Fully verified
Every field checked by a human against multiple sources.
0species
A
Cross-verified
Care data confirmed against two or more independent sources.
3,417species
B
Core care
The essentials — light, water, growth — verified and stored.
1,654species
C
Name only
A real, accepted species with NO care data. We say so, loudly.
376,408species
The uncomfortable number is the honest one. Most of the catalogue is tier C — real species we can name but haven’t profiled. We’d rather show you that than quietly generate care data to fill the gap.
Questions, answered the way the app would
How many plants can it identify?
It matches against a catalogue of 381,479 species built on an accepted botanical backbone. What varies is how much we know about each one — thousands carry verified care profiles, and every record shows its grade.
What if it gets it wrong?
You can correct it — the plant is yours, and the species can be changed at any time. Identification suggests; it never locks anything in. And when nothing resolves confidently you get an empty result rather than a wrong one.
Does it work on plants that aren't houseplants?
Yes — garden plants, herbs, trees, succulents and vegetables are all in the catalogue. Care depth is strongest for commonly grown plants, and the record tells you when it's thin.
Do I need a perfect photo?
No, but a single leaf or the whole plant against a plain background beats a crowded shelf. If the frame is ambiguous, you'll be told it's uncertain instead of getting a confident guess.
Is plant identification free?
Yes — the free Hobbyist plan includes 30 AI scans a day, which covers identification and health diagnosis. Soil intelligence is a Botanist feature; light scans are on every plan.
Can it tell me if a plant is poisonous to my cat?
For the species we hold a toxicity record for, yes — with three possible answers: harmful, assessed as safe, or not assessed. We never treat “absent from the harmful list” as “safe”.
Admit one · plant of your choosing
Photograph the one you can’t name
Point the camera at whatever’s been sitting unnamed on your windowsill. If we know it, you get its whole file. If we don’t, you’ll get told that — which is the part most apps skip.