Meal drafts and confidence

The draft an assistant sends to log_meal, ingredient confidence tiers (0.85, 0.60, 0.30) and when a meal needs review.

In ChatGPT and Claude the host model does the vision: it looks at the photo or reads what the person said and sends log_meal a draft. Gutlog matches each item against the person's own foods, the barcode index and the food database, then portions it, computes nutrients and stores it. Gutlog's own analysis of photos from the web app, the CLI and the API produces the same draft, so there is one contract for both paths. The field descriptions below are what the model reads in the tool schema.

Example

log_meal input

{  "title": "Scrambled eggs on toast",  "meal_type": "breakfast",  "client": "claude",  "items": [    {      "name": "scrambled eggs",      "grams": 120,      "confidence": 0.92,      "portion_confidence": 0.7,      "cooking_method": "pan_fried",      "ingredients": [        {          "name": "egg",          "confidence": 0.97,          "visible": true        },        {          "name": "butter",          "confidence": 0.55,          "visible": false        },        {          "name": "milk",          "confidence": 0.35,          "visible": false        }      ]    },    {      "name": "sourdough toast",      "portion": {        "quantity": 1,        "unit": "slice"      },      "confidence": 0.85,      "ingredients": [        {          "name": "wheat flour",          "confidence": 0.9,          "visible": false        },        {          "name": "salt",          "confidence": 0.75,          "visible": false        }      ]    }  ]}

The meal

* marks a required field. meal_type is one of breakfast | lunch | dinner | snack | drink.

Field	Type	What to send
items *	array	One entry per distinct food or drink on the plate or in the message.
meal_type	enum	Leave out to infer from the time.
eaten_at	string	When it was eaten, ISO 8601 with offset; default now.
title	string	A short name for the meal: "Chipotle bowl", "Breakfast".
note	string	Anything the person said that is not an item (where, how they felt).
client	string	Your client name, e.g. "chatgpt", "claude", "cursor".
model	string	Your model name, for calibration.

Items

One item per distinct food or drink. Send grams when you can estimate the weight from the photo (plate size, utensils, a hand), otherwise a household portion. A readable barcode makes the product's label win over estimates. cooking_method is one of 15 values such as grilled, pan_fried or unknown.

Field	Type	What to send
name *	string	The item as it would appear on a menu or label: "grilled chicken breast", "white rice", "oat milk latte".
food_id	string	A food id from search_foods or a previous log_meal reply, when you already know the match.
barcode	string	Barcode digits if you can read them (UPC/EAN); the product's label then wins over estimates.
brand	string	Brand when packaging shows it.
grams	number	Your estimate of the edible weight in grams. Prefer this when you can estimate it from the photo (plate size, utensils, hand).
portion	object	Instead of grams: "2 slices", "1 cup", "1 medium".
servings	number	Number of the product's servings (with a barcode).
confidence	number	How sure you are of what this item is, 0–1.
portion_confidence	number	How sure you are of the amount, 0–1.
cooking_method	enum
ingredients	array	Ingredients with a confidence each, most certain first. Include hidden fats, sugars, onion and garlic when plausible.
label_text	string	Any ingredient list or nutrition label text you can read, verbatim.

Ingredients

Most certain first. Include hidden fats, sugars, onion and garlic when they are plausible, at 0.30 to 0.60: they matter for calories and for gut tracking, and the person is asked rather than assumed.

Field	Type	What to send
name *	string	One ingredient as a person would say it: "garlic", "olive oil", "cheddar".
confidence *	number	How sure you are it is in this item, 0–1. ≥0.85 you can see or were told it; 0.60–0.85 very likely for this dish; 0.30–0.60 plausible but hidden (cooking fat, butter, sugar, garlic, onion in a sauce): include these, they are asked about. Omit below 0.30.
visible	boolean	You can see it in the photo (true) or inferred it from the dish (false).
ingredient_id	string	Canonical ingredient id if you know it from a previous response; otherwise leave it out.

Confidence tiers

Tier	Confidence	What happens
confirmed	≥ 0.85, or read from a label	Included and counted in the meal.
likely	0.6 to 0.85	Included, counted and pre-selected: one tap or "yes" confirms them all.
possible	0.3 to 0.6	Suggested, not counted. Asked about, highest confidence first.
dropped	under 0.3	Not stored. Leave these out of the draft.

An ingredient's status starts as included (confirmed, likely) or suggested (possible), and becomes confirmed or rejected when the person answers.

Always logged; sometimes reviewed

The meal is always stored: nothing waits for a confirmation.
needs_review is set when the meal's confidence is under the person's threshold (default 0.8), when a possible ingredient is one of their known triggers or allergens, or when a portion looks implausible.
The answer carries at most 3 questions about possible ingredients. Reply with one line of what was logged, then ask them.
Corrections always win over estimates, and every one is kept with who made it and from where.

log_meal result (shortened)

{  "meal_id": "6f1c2e0a-4b7d-4c2e-9a51-0d3b8e7f9a10",  "needs_review": false,  "totals_display": {    "kcal": 352,    "protein_g": 19.8,    "carbs_g": 24.1,    "fat_g": 19.6  },  "items": [    {      "name": "scrambled eggs",      "grams": 120,      "ingredients": [        {          "name": "egg",          "tier": "confirmed",          "confidence": 0.97,          "status": "included"        },        {          "name": "butter",          "tier": "possible",          "confidence": 0.55,          "status": "suggested"        },        {          "name": "milk",          "tier": "possible",          "confidence": 0.35,          "status": "suggested"        }      ]    },    {      "name": "sourdough toast",      "grams": 50,      "ingredients": [        {          "name": "wheat flour",          "tier": "confirmed",          "confidence": 0.9,          "status": "included"        },        {          "name": "salt",          "tier": "likely",          "confidence": 0.75,          "status": "included"        }      ]    }  ],  "questions": [    "Butter in the eggs?",    "A splash of milk in the eggs?"  ]}

Amounts in results are integers in each nutrient's base unit; see nutrients and units. Logging a draft is free and unlimited on every plan.

Docs menu