FloraFang · Support
Support
Email support@ramirezlabs.app and a person will reply. Include your iPhone model and iOS version, and if a specific scan went wrong, roughly what you were pointing at.
Engineering Deep Dive
Building FloraFang: A 4-Layer Defense Against Neural Network Overconfidence →
Why an 86% softmax score is not an 86% chance of being right, how empirical temperature scaling paired with an on device agreement veto eliminates dangerous false reassurance down to 0 of 346, and how multimodal intelligence keeps users safe in the field.
If someone or an animal has eaten something, do not use this app to decide what to do. Call Poison Control at 1.800.222.1222, or for pets, ASPCA Animal Poison Control at (888) 426.4435 or the Pet Poison Helpline at (855) 764.7661. All are answered around the clock. Call first and identify later.
The part that matters is: yes, always. Every hazard identification, for both spiders and plants, and the entire exposure intake checklist along with exporting an exposure incident report, work with no payment, no account, and no trial period, forever. A single one-time purchase (no subscription) optionally unlocks five extras that are genuinely non-essential: the Field Naturalist chat, exporting your personal field log, shareable identification cards, a species life list, and the full species catalog browser. None of them change what a scan tells you.
Because separating individual species often requires a microscope even for an arachnologist, whereas field triage requires knowing whether an organism presents a medical hazard. FloraFang focuses on the key hazard groups: Black Widows, Brown Recluses, and common harmless families (such as jumping spiders, wolf spiders, cellar spiders, huntsman spiders, orb weavers, and tarantulas). For plants, it recognizes a verified list of notorious toxic species and nothing else. When evidence is uncertain, it issues an honest refusal instead of guessing.
Modern machine learning models are notorious for producing high confidence numbers (like 85%) even when looking at unfamiliar noise or computer screens. FloraFang uses a four layer defense in depth architecture to prevent this: (1) an entropy filter that detects out of distribution blur or screen glare and halts guesses; (2) empirical temperature scaling (T = 1.53) with a strict calibrated benign floor (0.86) paired with a secondary model agreement veto that eliminates false reassurance on dangerous species down to 0 of 346 holdout samples; (3) multimodal corroboration requiring on device Apple Intelligence foundation models on iOS 27 or later to verify visual markings (with graceful Core ML fallback on earlier iOS releases); and (4) strict clinical framing that never labels wild organisms as “safe,” but rather “not medically significant,” reminding users that defensive bites can still occur.
Powered entirely on device by Apple Intelligence Foundation Models on iOS 27 or later, the Field Naturalist is an offline conversational assistant built into every saved scan. You can tap quick prompt chips (“Safe way to move it?”, “Typical habitat?”, “Best photo angles?”) or ask custom questions about physical markings and relocation. Crucially, medical advice is blocked at the gate: an automated, deterministic Swift filter intercepts any query mentioning bites, symptoms, or dosages before the model executes, immediately presenting poison control resources instead. It's one of the five extras behind FloraFang's one-time unlock; the hazard identification itself stays free either way.
A refusal comes with what to do next rather than a dead end. It tells you how to get a better photo, or where to take the question. It is a result, not an error.
No, and the app is careful never to say otherwise. FloraFang knows a short list of well-known toxic plants. Most toxic plants are not on that list, so failing to match one is not evidence of anything. Treat an unrecognised plant as unknown, not as harmless.
No. FloraFang will never tell you a plant is safe to eat, at any confidence, and that is a deliberate design decision rather than a missing feature. Ruling out a specific known toxic species is a claim a photo can support. Confirming something is edible is not, and the consequences of being wrong are severe enough that a phone camera is the wrong instrument for the question.
It gets you to poison control quickly and helps you make that call well. The numbers appear before anything else, because seconds spent waiting on a photo to process are seconds not spent dialling. While you're on hold, it collects the things poison control will ask for, including weight and age, which part was eaten, how much, when, and any symptoms, into a single block you can read aloud. That part is useful whether or not the app identifies the plant correctly, which is why it exists.
Fill the square frame with the subject and get close. Cropping tightly to the organism is the single biggest thing you control. A spider on a wall can be two percent of the picture, and at that size the app is mostly looking at stucco. Use the zoom slider rather than moving the phone closer; past about ten centimetres the main lens physically cannot focus. Recent iPhones switch to the macro lens automatically when you get close enough. Tapping the subject on screen locks focus, which helps a lot on a flat wall.
Yes. Identification runs entirely on your iPhone and the app makes no network requests at all. The poison control numbers are built in and work offline, though placing the call itself needs a phone signal.
Deliberately. Wrongly flagging a harmless spider costs you a moment of concern. Wrongly calling a dangerous one harmless costs a great deal more. So a hazard warning has a lower bar to appear than an all-clear does, and the app breaks ties toward caution.
In the Field Log. Scans are saved automatically as soon as an identification completes, preserving the exact photo, timestamp, location, and hazard assessment recorded at that moment. If you prefer not to keep a scan, you can immediately delete it directly from the result screen. Within the Log tab, you can search observations, filter by hazard tier (such as urgent or caution), or switch to the dedicated Exposure Log to review saved exposure intake reports.
FloraFang is not a safety authority and not medical advice. Do not decide whether something is safe to touch, eat, or approach based on this app. If you have been bitten or stung and feel unwell, contact emergency services or Poison Control at 1.800.222.1222 (US) right away. Do not wait to identify anything first.