Plant Diagnosis — Plant Monitor (Tiers 1

Plants / Plots

Settings

Kindwise crop.health API key

Tried first for entry-photo diagnosis (below) — specialized for the 23 major food crops. Sign up free at kindwise.com.

Kindwise plant.health API key

Used as a fallback when crop.health has no key or comes back with nothing usable (e.g. a houseplant or ornamental, not one of the 23 food crops). Kindwise issues this as a SEPARATE key from crop.health, even though it shares billing with plant.id.

Kindwise insect.id API key

A third, independent Kindwise key — identifies a bug from a photo of the bug itself, not the plant. Use the "Identify insect (test)" tool below to try it. Calling any of these directly from this page sends the key in each browser request — fine for personal testing, but for anything shared, use the local server's /diagnose proxy instead (below) so keys never leave your machine.

Identify insect (test)

Camera server URL (optional)

Not checked

Either works, with the exact same /upload, /captures and /diagnose endpoints: run node local-server/server.js on a machine on your network (fully offline, e.g. http://192.168.1.50:8787), or point this at your Netlify site's URL (e.g. https://your-site.netlify.app) once its cloud functions are set up — see ai-capability/supabase-schema.sql and netlify/functions/ in the Vrksayurveda project.

Route Plant Diagnosis through the local server instead of calling Kindwise directly Automatically diagnose newly imported camera photos

On by default. If the local server already diagnosed a photo when it arrived (Kindwise key or trained model on that server), the import just uses that result — this box only triggers a fresh diagnosis from the browser for photos the server hasn't diagnosed yet.

Automatically check for new camera photos while this tab is open (every ~90s)

On by default. Nothing to click — new captures (already diagnosed by the server, if it has a Kindwise key or a trained model) appear in the right plant's timeline on their own. The button below still works too, for an immediate check.

Add a plant on the left, then log or import its first weekly entry here.

Tier 3 — Build a Training Dataset

Turns everything logged above into a labeled folder of images, ready to train a plant-photo classifier — this runs right here in the browser (via JSZip), so there's nothing to install for this step. It replaces running export_dataset.js by hand. Actual model training still happens separately in Python (below), since real neural-network training needs more than a browser tab can reliably give.

Label by growth stage Label by top Plant Diagnosis result (weak labels)

Weak labels come from one Plant Diagnosis model's guess, unverified by a person — fine to experiment with, but spot-check before trusting a model trained on them.

Automatic training now runs on the local server, on a schedule. It builds its own dataset directly from camera captures + their saved Plant Diagnoses and retrains a classifier — nothing to click here for that (config.json on the server: autoTrainEnabled, autoTrainIntervalHours, every 24h by default). One limitation: the server can only see what it captured itself, not this browser's growth-stage entries, so scheduled training always labels "by top Plant Diagnosis" (never by growth stage). Once a training run finishes, the server automatically uses that model for future diagnoses — falling back to Kindwise if the local model can't run — and results show up in Outcomes below with no change needed here. If you specifically want a growth-stage-labeled model, use the manual export + Python steps below instead — that path still works exactly as before.

Manual / stage-labeled training (optional, outside the browser)

For a growth-stage-labeled model specifically (the automatic pipeline above can't produce one). Unzip the download above with "Label by growth stage" selected, then from the tier3-training/ folder:

pip install tensorflow-cpu --break-system-packages

python3 train_classifier.py --dataset ./unzipped-folder --output ./model --epochs 10

See AI_Capability_Tier1_2_3_Technical_Deep_Dive.docx for full setup steps and an honest account of what's tested vs. not.

Outcomes — Diagnosis History

Every Plant Diagnosis across all plants, most recent first — camera photos diagnosed automatically by the local server, manual "Run Plant Diagnosis" runs, and (once trained) results from the server's own local model.

All diagnoses Issues flagged only

Access restricted

Plant Diagnosis — Plant Monitor (Tiers 1–3)

Plants / Plots

Settings

Timeline

Tier 3 — Build a Training Dataset

Manual / stage-labeled training (optional, outside the browser)

Outcomes — Diagnosis History