AI Diagnosis¶

With AI Diagnosis you describe your plant's symptoms through a short, guided assistant — instead of writing a free-form chat question, you pick from a curated catalogue. You receive the three most likely causes with confidence, an explanation, and recommended actions, each linked to the matching entry in the Pest Management (IPM) system.

Partially available

AI Diagnosis is usable as its own page AI Diagnosis (/diagnose): you select symptoms from a curated catalogue, optionally add a note, and get an AI-generated top-3 assessment linked to the pest management system. Currently not implemented are a link to a specific plant or planting run, a photo upload directly inside the assistant, and a saved diagnosis history — every session is stateless and is not retained after the answer. There's also no dedicated menu entry yet. The following sections describe today's state in the present tense.

An assessment — not a definitive diagnosis

Every answer is an AI-generated assessment based on the symptoms you selected, not a definitive diagnosis. Please verify the finding yourself before starting any treatment. AI Diagnosis never triggers a treatment automatically, and any existing pre-harvest interval gate (the legally required waiting period between treatment and harvest) remains active regardless.

Full mode and consent required

AI Diagnosis is only available to signed-in users. In Light Mode (anonymous access) the feature is unavailable. In addition, the analysis — just like the AI Assistant — needs all three approval levels (operator, your garden/tenant, your consent "AI access to your plant data"). If any of these is missing, you can still fill out the assistant up to the last step, but the analysis itself then fails with a generic error message (see When the analysis fails).

Prerequisites¶

Signed-in user account (not Light Mode)
AI features must be enabled instance-wide and for your garden (tenant)
Your consent "AI access to your plant data" (ai_tenant_data_access) must be granted — see AI Assistant: Granting consent for details

What the AI Diagnosis Does¶

AI Diagnosis is a structured, symptom-based path to an assessment — you describe what you see on your plant instead of formulating an open question. That lowers the jargon barrier and gives the AI a clearly scoped input format.

How this differs from photo-based recognition

AI Diagnosis is text-based: you select symptoms, not images. That makes it different from Kamerplanter's photo-based features:

Identify Plant by Photo determines the species of an unknown plant from a photo.
Pest Detection by Photo detects pests and typical damage patterns from a photo of an already-known plant.

The context step of AI Diagnosis (see below) includes a hint pointing to the existing photo recognition feature, in case a photo could provide additional context — but AI Diagnosis itself does not accept or evaluate a photo.

Getting Started¶

You currently reach AI Diagnosis directly at the address /diagnose in your browser — there's no dedicated entry in the side menu yet.

The assistant guides you through three steps: Symptoms, Context, and Result. A progress indicator at the top shows which step you're on.

Step 1: Select Symptoms¶

The catalogue contains over 30 curated symptoms, grouped by category (for example "Leaf discoloration", "Visible pests", or "Growth anomaly"). Check every symptom that applies to your plant — multiple selections are possible.

Preview possible causes

Next to many symptoms, a small info icon shows a short hint about common causes when tapped or hovered — that helps you pick the right symptoms even before you submit.

Once you've selected at least one symptom, the Next button becomes active.

Step 2: Add Context (Optional)¶

In this step you can optionally leave a note in the free-text field (up to 2,000 characters) — for example about location, watering habits, or when the symptom first appeared.

Your note is not sent verbatim

For privacy reasons, your free text is not forwarded to the language model. The AI only receives the neutral hint that you added a note — your text itself stays with you.

Below that you'll find a hint with a link to the existing photo recognition feature, in case a photo could provide additional context (see How this differs from photo-based recognition above).

Click Start Diagnosis to trigger the analysis — or Back to change your symptom selection.

Step 3: Read the Result¶

After a few seconds, the three most likely causes appear, sorted by descending confidence.

Understanding the Result¶

Each of the three diagnosis cards shows:

Rank and name of the suspected cause, supplemented with the scientific name when botanically nameable
Confidence as a percentage, color-coded (green = high, blue = medium, orange = low)
Explanation of why the AI arrived at this assessment
Recommended actions as a bulleted list

If the system finds a match in the pest master data, the card additionally links directly to the matching pest detail page and to suggested treatments. If a suggested treatment carries a pre-harvest interval (the legally required waiting period between treatment and harvest, see Glossary), that's clearly marked on the treatment chip with a warning icon and the number of days.

The answer is also embedded in the shared envelope used for all AI answers — AI labeling, expandable source references from the knowledge base, and indicators showing whether your plant data or a cloud provider were used. Details are described under AI Assistant: Transparency.

Use the New Diagnosis button to restart the assistant with an empty symptom selection.

When the Analysis Fails¶

If the analysis fails, the assistant shows a generic error message regardless of the exact cause — whether AI features aren't enabled, your consent is missing, the knowledge base is unreachable, or the AI didn't produce a usable result. In that case, first check the three approval levels under AI Assistant and try again afterward.

From Diagnosis to Treatment¶

If a diagnosis points to a known pest, you can go directly via the View pest details link to the pest detail page and create a pest inspection from there in the Pest Management (IPM) system. If a suggested treatment carries a pre-harvest interval, it's honored on the treatment detail page — the system never triggers a treatment automatically; the decision is always yours.

Frequently Asked Questions¶

Why can't I find AI Diagnosis in the side menu?

The feature doesn't have its own menu entry yet. You reach it directly at the address /diagnose.

Can I attach a photo to the diagnosis?

Not directly inside the assistant. The context step includes a link to the existing photo recognition features — use Pest Detection by Photo for pest detection, or Identify Plant by Photo for pure species identification.

Is this the same as the planned photo diagnosis for diseases and nutrient deficiencies?

No. AI Diagnosis is symptom- and text-based and usable today. A complementary, photo-based condition diagnosis specifically for diseases and nutrient deficiencies is planned as its own feature — the API already exists but is not yet wired into a user interface — see My Plant Doesn't Look Well — What Now?.

Why do I always get a generic error message when something goes wrong?

The assistant doesn't yet distinguish between the possible failure causes (missing approval, missing consent, an unreachable knowledge base, or an unusable AI result) — the same generic message appears in all cases. If in doubt, check the AI Assistant's approval levels.

Is my diagnosis saved so I can find it again later?

No. Every session is stateless: only an anonymized, hashed entry is written to the internal AI audit log (without the plain text of your symptoms or notes), but there is no diagnosis history you could review later.