Say what you ate. AI extracts calories, protein, carbs, and fat in under 12 seconds. No barcodes. No database. No forms.
Open Salamati and tap the mic icon. No navigation required — the voice input is always one tap from the main screen. The waveform activates immediately, showing Skrypt is listening.
Talk the way you'd describe a meal to a friend. "Grilled chicken sandwich with fries and a Coke" works just as well as "chicken breast, white bread bun, potatoes fried in oil, 355ml cola." Skrypt understands both. Include portion qualifiers if they're useful — "a big portion," "half," "two cups." The AI handles natural language, not just database keywords.
Deepgram transcribes your description in under a second. Claude classifies the domain (nutrition), identifies the foods and portions, and estimates calories, protein, carbohydrates, and fat. The entire AI processing step takes 1–3 seconds on a typical connection. You'll see the structured entry appear on screen with the breakdown.
Review the entry. If the estimate looks right, you're done — the log is saved to your timeline. If you want to adjust a macro or change the food description, tap to edit. Editing is quick and every field is directly accessible. The log is added to your daily totals and your history immediately.
Skrypt's classification model is trained on how people actually talk about food — not just standardized food database names. This means you don't need to speak in "food database" — you can describe meals the way you'd tell a friend.
Estimates are based on typical preparations. Any estimate can be edited after logging. Providing more detail (portion size, cooking method, brand) improves accuracy.
For most people's goals, "accurate enough" is more useful than "exact." Here's what you should expect from Skrypt's voice nutrition tracking:
What Skrypt does well: common whole foods (eggs, chicken, rice, vegetables, fruit), typical restaurant categories (bowls, sandwiches, pasta dishes), and meals you describe with contextual portion cues ("big bowl," "small serving," "half a plate"). The AI also handles cultural food names, cooking method adjustments, and multi-component meals described in a single sentence.
Where estimates are less precise: highly specific packaged foods where exact label data matters (Skrypt doesn't scan nutrition labels), complex restaurant dishes with unknown ingredient quantities, and highly customized meals where the exact proportions are unusual. In these cases, editing the estimate after logging takes about five seconds.
What accuracy means in practice: if you're using nutrition tracking to understand patterns — whether you're eating enough protein, whether you're hitting calorie targets within a 200-calorie range, whether your carb intake correlates with your energy — Skrypt's estimates are accurate enough to generate actionable insights. Research on nutrition tracking consistently finds that the habit of logging (consistency) matters more for outcomes than the precision of any individual entry.
If you need gram-precise tracking from nutrition labels — for competition prep or clinical diet management — you should use Skrypt's edit feature to input exact values, or consider a barcode-scanning app for packaged foods specifically. Skrypt is designed for the other 95% of use cases.
There are three main ways to log nutrition data, each with a different trade-off between speed and precision:
| Method | Speed | Works for | Best for |
|---|---|---|---|
| 🎙️ Voice (Skrypt) | <12 seconds | Everything — whole foods, restaurant meals, home cooking, cultural dishes | Consistency. Logging when life is busy. |
| 📷 Barcode scan | 5–15 seconds | Packaged foods only — anything with a barcode | Exact label nutrition for packaged products |
| 🔍 Database search | 60–120+ seconds | Foods that exist in the database by name | Finding exact entries for known foods |
| ✏️ Manual entry | 120–300+ seconds | Anything — if you know the exact macros | Edited logs, when you have exact values |
The key insight: barcode scanning is fast for packaged foods, but most of what people eat isn't packaged. Home-cooked meals, restaurants, cafés, meals at friends' houses — none of these have barcodes. Voice logging works for all of it at the same speed.
Database searching is slow for any meal that isn't a single, known item. A chicken bowl with three toppings requires three separate searches, three portion selections, and three log entries. By voice: one sentence, one entry, done.
Voice nutrition tracking is where most people start with Skrypt — it's the most immediately useful feature and the one with the clearest time savings. But Skrypt is designed around five health domains, and the real value emerges when you use all of them.
Nutrition affects everything else. What you eat affects how you sleep. How you sleep affects your training. Your training affects your mood. Your supplement routine affects your energy and recovery. Skrypt tracks all five with the same voice-first approach, and the AI can surface connections across them over time.
The "Ask Skrypt" feature lets you query your own health history in natural language: "What was my protein intake last week?" "How many days did I hit my vegetable goal?" "Was my energy better on days I ate breakfast?" These questions require all five domains to be in the same system — which is why Skrypt tracks all of them.
You can start with just nutrition logging and add other domains gradually. Many users start by replacing their food diary with voice logging, then add sleep and supplement tracking in week two, then start mood check-ins when they're curious about patterns. Each domain is additive — they don't require each other to work.
