On device · Offline · Open source
A pocket field guide that collects what you find.
Point your camera at a plant. PlantDex suggests what it likely is — ranked, with a confidence level — and presses each find into your own illustrated atlas. Identification runs entirely on your phone.
How it works
Four steps, no signal required.
The whole loop works in airplane mode — the model lives on your phone, so there is never a "connecting…" spinner in the field. Nothing about your plant is uploaded to identify it.
-
01
Capture
Fill the frame with a leaf or flower and tap once. Or pick a photo you already have.
-
02
Identify
On-device, you get a ranked list of candidate species, each with a confidence level — never a single, unqualified verdict.
-
03
Catch
Confirm a match and press the species into your collection — kept locally, on your device.
-
04
Browse the atlas
Watch your illustrated dex of the region fill in, one find at a time, with notes, taxonomy and sources.
The mission
An atlas of every plant on Earth — region by region.
Version one covers south-western Europe: around 7,800 species. From there we expand region by region, building toward a guide to the world's flora that anyone can carry, anywhere — and that everyone can help improve.
Open source
Open model. Open data. Reproducible.
The classifier and the data pipeline are public — Apache-2.0 code, CC-BY / CC0 data — so anyone can audit the work, reproduce the build, or extend the guide to a new region. The model is shared under CC-BY.
App repository
The Flutter app, end to end — Apache-2.0.
Data pipeline
Ingestion, licence gates & sketch generation.
Model + cards
Weights, the model card & the data card.
The published model release is live at github.com/michal-giza/plant-journal-model. The app and pipeline repositories will be opened alongside the first public release.
Contribute
Help the map grow.
This is a project, not just an app. Researchers, naturalists and translators can help open up new regions, contribute openly-licensed data, and localise the guide. The long-term vision is a worldwide flywheel where every openly-shared, verified record makes the atlas a little more complete for the next person on the trail.
- Add openly-licensed (CC-BY / CC0) data & occurrences.
- Help bring a new region into scope.
- Translate the guide (English now; Polish & Spanish planned).
Privacy & safety
It identifies. It never advises.
Private by design
Identification runs on your device. Your photos never leave your phone to be identified, your collection is stored locally, and no account is required. The core loop works with no connection at all.
Honest about limits
PlantDex helps you identify plants and learn about them — it always shows several candidates with a confidence level, never a single certain answer. It does not tell you whether a plant is safe to touch, eat, or use. Never forage, eat, or use a plant medicinally based on an identification alone.
Credits & licences
Built on open science.
PlantDex stands on freely-shared biodiversity data and research. App and pipeline code is Apache-2.0; catalogue data is CC-BY / CC0; the model is CC-BY. With gratitude to:
- GBIF
- Global Biodiversity Information Facility — occurrence & range data, and the taxonomy backbone used to reconcile names. CC-BY / CC0
- PlantCLEF 2024 · Pl@ntNet
- The classifier builds on PlantCLEF 2024 / Pl@ntNet work — Goëau, Lombardo, Affouard, Espitalier, Bonnet & Joly. CC-BY 4.0
- Wikimedia Commons
- Openly-licensed reference imagery for species in the atlas. CC-BY / CC0
- Wikipedia
- Species descriptions and reference text. CC-BY-SA