Most plant sites are content farms. They generate a confident paragraph about every species, invent the care advice, and never tell you where a single fact came from. BotanAI is built the other way around: we publish what we can trace to a named dataset, we show the gaps we cannot fill, and we put our own identification accuracy on the table, including the part that makes our model look bad.
That discipline is the whole product. A page with nothing real to say is never generated. Where a fact is missing, the page says so plainly rather than filling the space with prose that merely sounds right. It is the one thing a generator structurally cannot copy.
The reference
We publish 56,125 species, each built from open data with a source, a date and a number on every claim: names and classification from World Flora Online, native range from Kew’s world checklist, flowering curves and climate envelopes computed from research-grade observations, and 395,688 openly licensed photographs, each credited to its photographer. Where a species carries a real herbarium sheet, its accession number and determiner are shown with it.
WFO · names, classification Kew · native range Research-grade obs · flowering, climate Licensed photos · creator-credited
That is not the number of plants that exist. It is the number that clears our evidence gate. Species with too few commercially usable photographs to document are omitted rather than turned into empty pages, though they are still listed by name inside their genus. The full account of what each dataset is, its licence, when we last synced it, and where our coverage is thin, lives on the sources page.
The honesty thesis
Every plant-identification app advertises an accuracy figure. Almost none publish a method, a dataset, or a list of the plants they get wrong, because the honest numbers are not flattering. Fine-grained identification from a single photograph is genuinely hard.
So we measured our own on-device model on 38,949 openly licensed photographs across 2,040 species and published all of it. It gets the species exactly right 55.9% of the time overall. That is worse than the figures competitors advertise, and we show it anyway. The number that matters to you is different: the model only commits when it is confident, on about 34.5% of photos, and when it does commit it is right 89.5% of the time. Knowing when to stay quiet is most of the value. The full method and confusion matrix are on the accuracy page, and they are reproducible, which is the point.
The app
BotanAI is a free Android app with no ads and no sign-in of any kind. There are no accounts and there is no BotanAI server. Your garden, your photos, your growth timeline, your care logs, your AI Plant Doctor history, your wishlist and everything the app calls “community” are stored only on your device. Community features are single-device and local-only: nothing is transmitted to us or to other users. Deleting your data means uninstalling the app or clearing its storage. There is nothing on a server to delete, because there is no server.
Identification is a privacy-first cascade, tried on your phone first and escalated only when it is not sure. It is not three engines cross-checked at once.
- 1 On-device model Google AIY Plants V1, 2,102 classes. Fully offline. Nothing leaves the device.
- 2 Pl@ntNet Tried only when the phone is not confident. The photo and an API key are sent.
- 3 Plant.id The last step, for the hard cases the first two could not settle.
The AI Plant Doctor is a text chat. If you attach a photo, the photo is analysed by Plant.id and only a short text summary is passed to the chat model. Weather-smart care reminders use your approximate location to fetch local weather. The app asks for camera, notifications, coarse location only, and the permissions needed to reschedule reminders after a restart. There is no fine location, no background location, and no advertising id.
Everything core is free: unlimited identification, the AI Plant Doctor, disease and health checks, the light meter, weather-smart reminders, and the growth timeline. The free tier caps the garden at 20 plants. Pro removes that cap for about $2.99 a month or $19.99 a year, billed through Google Play, with the exact price shown in the app because Google Play sets it. No ads, ever.
Where your data goes
When data does leave the device, it goes only as a transient, per-feature call to a third-party processor, and only what that feature needs. Here is every one, and exactly what it receives.
| Processor | Used for | What it receives |
|---|---|---|
| Pl@ntNet my-api.plantnet.org | Identification | The plant photo or photos, plus an API key. Only when the on-device model is not confident. |
| Plant.id plant.id | Identification, health | The plant photo, an API key, and approximate latitude and longitude when available. For identification and health, and to analyse a photo you attach in the AI Plant Doctor. |
| Groq api.groq.com | AI Plant Doctor | Chat text only: your typed message, a short text summary of the identification, and the plant name. No photo, no location, no identifier. |
| Google Gemini, via Firebase AI Logic googleapis.com | Verification, voice | Downscaled photos and text prompts, for PlantProof photo verification, and text only for an optional plant voice that is off by default. Attested by Firebase App Check. |
| Open-Meteo api.open-meteo.com | Weather | Approximate latitude and longitude, to power weather-smart care reminders. |
| Firebase Analytics and Crashlytics, Google google.com | Diagnostics | Anonymous, non-identifying usage events and crash reports. No user id, no user properties, no advertising id. Crash reporting in release builds only. |
| Google Play Billing google.com | Billing | Subscription and purchase state, for Pro. |
Nothing is sold, and there is no advertising id anywhere in the app. The full detail of each call lives in the privacy policy.
What we do not do
We do not have a founder story to sell you, so there is not one here. We do not publish invented awards, press quotes, user counts, or testimonials. The app is newly launched, which is exactly why there are no reviews or star ratings shown anywhere on this site: manufacturing them would be the first dishonest thing we did.
We do not publish care advice or toxicity claims we cannot cite to a source. An earlier version of this site did, and those pages were retired for it. We do not use stock imagery or AI-generated pictures. And we do not claim a community that verifies identifications, because the app’s community is local-only and never leaves your phone.
Corrections
If something here is wrong, we want to know, and we will say what changed and when. Write to support@botanai.app.