I'm into nutrition, diet and exercise, and I wanted one app that brings the useful parts together: food logging, macros, weight, goals, what's in the kitchen. The good ones are mostly behind a paywall, and the free ones make you search a database for every ingredient. So I built nomnom to be free and usable by anyone: type what you ate the way you'd say it, and get numbers you can trust. No account, no subscription, no server.
- Type it like you'd say it. "120g grilled chicken, a cup of quinoa and a Greek salad" becomes items, grams, calories, protein, carbs and fat. Grams, pieces, cups, bowls, plates, or a photo of the plate.
- Real numbers, not guesses. The AI only reads your sentence. Calories come from USDA FoodData Central, tables of about 300 dishes across 16 cuisines, and Open Food Facts for branded products. Every item shows its source, and anything uncertain is flagged or asked about instead of guessed.
- Free, with your own key. It uses a free Groq or Gemini key, or a model running on your own computer. Simple meals like "2 boiled eggs and a banana" are read on the phone with no AI at all.
- Learns your real maintenance. After a few weeks of logging and weighing in, it works out what you actually burn and suggests a better target. Nothing changes until you accept.
- Everything in one place. Weight trend and BMI, micronutrients, a weekly recap, meal reminders you can reply to, steps from Health Connect, a home-screen widget, and a dark mode.
- A pantry that counts down. Add "10 eggs, 450 g chicken breast" when you shop, and it goes down as you log. Cooked weights are worked back to raw, and it tells you when something runs low or nears its use-by date.
- Private. Your log stays on your phone. Only what you log is sent, and only to the AI you chose. Keeps itself up to date, and asks before installing anything.
Download the APK from the latest release and open it on your phone. Android asks once to allow installs from your browser or file manager. During setup, paste a free Groq key (the quickest to get) and start logging. nomnom checks for new versions about once a day and asks before installing them.
"grilled chicken and a Greek salad"
│
▼
AI reads the sentence ──► items, amounts, grams, search terms
│
▼
bundled food tables ────► nutrition per 100 g (USDA + regional dishes)
│
▼
app does the maths ─────► totals you can adjust with exact arithmetic
The AI is good at understanding language and unreliable at remembering numbers, so it's only trusted with the first step.
- USDA FoodData Central, SR Legacy: about 7,300 generic foods with household portion weights.
- Dish tables by cuisine: about 300 dishes across Indian, Chinese and Indo-Chinese, Japanese, Korean, Thai, Vietnamese, Southeast Asian, Italian, European, British, American, Mexican, Latin American, Middle Eastern, African and South Asian food. Every table is searched for every user, so takeaway from another cuisine matches properly. Values are typical recipes, cross-checked against USDA's prepared foods.
- Open Food Facts: packaged products, searched when you name a brand ("a Quest protein bar").
Home-style dishes are averages and say so. Each has a Light / Typical / Rich
setting for oil and ghee, which nomnom remembers. Foods you confirm are
remembered too, so repeat meals come out the same every time. Weighed and
counted foods land within about 10 kcal of the reference values in
tool/live/precise_test.dart.
Pick one during setup, or later under You → AI model.
| Provider | Key | Notes |
|---|---|---|
| Groq | Free at console.groq.com/keys | Fastest, and the free tier lasts for daily use. Defaults to llama-3.3-70b-versatile. |
| Gemini | Free at aistudio.google.com/apikey | Defaults to gemini-3.5-flash-lite. The free tier can be as low as 20 requests a day per model. |
| Custom | Optional | Any OpenAI-compatible server: Ollama, LM Studio, OpenRouter, vLLM… |
To use a model on your own computer with Ollama, run
OLLAMA_HOST=0.0.0.0 ollama serve, choose Custom in nomnom and use
http://<your-computer's-wifi-ip>:11434/v1. Phone and computer need to be on
the same network.
- Your log, weights, pantry and goals are stored only on the phone.
- API keys stay in the phone's secure storage and are never written to backups.
- Only what you log (text, or a photo you choose to send) goes to the AI provider you picked. The weekly recap sends summary numbers and food names, nothing else.
Needs Flutter 3.44.
flutter pub get
flutter run
flutter test| Folder | What's in it |
|---|---|
lib/ai |
Provider clients (OpenAI-compatible, Gemini) and the meal parser. |
lib/nutrition |
Food table search, on-phone meal reading, targets, check-in, weekly recap. |
lib/data |
Models, the store over Hive, pantry, reminders, updater, key vault. |
lib/screens |
Setup, Today, review, History, Progress, Pantry, You. |
lib/theme, lib/ui |
The Paper design tokens and shared widgets. |
assets/data |
usda.json and the dish tables by cuisine. |
tool |
Data build and check scripts, screenshot harness, live accuracy checks. |
More for developers
Screenshots of every screen, rendered with the real fonts into build/shots/,
plus the README images in docs/:
flutter test tool/shots --update-goldensLive checks against a real model (they spend free-tier requests):
GEMINI_KEY=... flutter test tool/live/precise_test.dart # weighed inputs with known answers
GEMINI_KEY=... flutter test tool/live/live_test.dart
Q='whole milk|poha' flutter test tool/live/search_probe_test.dartRebuild the USDA table from the SR Legacy CSV download, and check the dish tables against it:
python tool/build_usda.py path/to/FoodData_Central_sr_legacy_food_csv_2018-04
python tool/check_dishes.pyReleasing. Add a ## [x.y.z] section to CHANGELOG.md, set the same
version in pubspec.yaml, then push a tag:
git tag v2.7.1 && git push origin v2.7.1The release workflow checks the version, builds and signs the APK and drafts a
release with the changelog as its notes. Publishing the draft is what makes it
reach users and the in-app updater. Signing uses the ANDROID_KEYSTORE_BASE64
and ANDROID_KEYSTORE_PASSWORD secrets (key alias nomnom). For signed builds
on your own machine, copy android/key.properties.example to
android/key.properties and point it at the same keystore.
nomnom is free. If it helps you, you can buy me a coffee:
Typography: Clash Grotesk (Indian Type Foundry, Fontshare licence) and Instrument Serif (SIL Open Font Licence). Nutrition data: USDA FoodData Central and Open Food Facts (ODbL).
Estimates, not medical advice.
nomnom is free software, licensed under the GNU General Public License v3.0.