Forky, answered.

Forky is an AI nutrition coach built as a native iOS app by LOMY SAS in Paris. It uses computer vision (GPT-4o) to identify every ingredient in a fridge photo or every component in a meal photo, then computes calories, protein, carbs, and fat using a per-component USDA-style lookup. It tracks daily macros against personalised targets, imports recipes from URLs, photos, PDFs or text, and coaches users in English or French.

Forky uses a three-pass vision pipeline (base layer, toppings, sauces) plus per-100g macro lookups rather than asking the model for whole-plate totals. Empirically this lifts accuracy from roughly plus or minus 25 percent on naive vision-only estimates to roughly plus or minus 10 to 15 percent on standard plated meals. Users can tap any component to adjust its gram weight and watch the totals update in real time.

You take one to three photos of the inside of your fridge. GPT-4o Vision returns a JSON list of every visible edible ingredient with an estimated quantity and freshness label. Forky upserts that list into your inventory, infers a shelf-life-based expiry date for each item, and surfaces the items expiring soonest in your meal suggestions so you cook the chicken before the broccoli.

Forky is free to install with a 5-day Apple Introductory Offer (free trial) on every recurring tier. After the trial: roughly 5 USD per week, 10 USD per two weeks, 13 USD per month, or 80 USD per year. A one-time Founders Lifetime tier is offered at launch. The free tier includes limited daily scans; Pro removes the scan cap and unlocks the unlimited recipe library and coach voice.

Not yet. The current build is iOS-only via the App Store, targeting iOS 16 and newer iPhones. An Android version is on the roadmap once the iOS app reaches a stable revenue baseline. The marketing site at forkyai.com works in any browser, but the photo-scan flows require the native camera.

Forky photos and macro logs are stored on a self-hosted backend in France under LOMY SAS (SIREN 988 926 325). Meal photos are uploaded to Cloudflare R2 for offsite backup, encrypted at rest, and served via short-lived presigned URLs. No data is sold. The full policy lives at forkyai.com/privacy and covers GDPR rights, subprocessors, and account deletion.

Yes. The import screen accepts four sources: a URL (Forky scrapes the page and structured data), a photo (vision extracts the ingredients and steps), a PDF (OCR plus parsing), or plain pasted text. Imported recipes land in your library with computed macros per portion and an AI-generated hero image that is vision-verified against the dish name before being saved.

Yes. The coach replies can be spoken aloud. By default Forky uses ElevenLabs for natural voice; if that quota is exhausted or the user has no internet, it falls back to expo-speech (iOS system voice) so the call-style interaction still works. Voice mode forces the model to reply in two short sentences so it feels like a phone call, not a wall of text.

Forky is built and operated by LOMY SAS, a French simplified joint-stock company (SIREN 988 926 325) headquartered at 60 rue François 1er, 75008 Paris. Founded in 2025. Support is reachable at [email protected] and the founder is Elie de Rougemont.

MyFitnessPal is a database-search tracker (you type or scan barcodes). Cal AI is a single-photo calorie estimator. Forky combines both with a third pillar: a fridge inventory plus AI recipe generation from what you actually have. Forky also imports recipes from any URL, photo, or PDF and persists per-component breakdowns so users can edit gram weights instead of accepting a black-box total.

English and French. The entire onboarding flow, paywall, coach prompts, and macro tracking interface translate end-to-end. Forky detects the device language on first launch and offers a one-tap switch in Profile. The backend stores the user's preference and respects it for AI replies in the coach.