Nutrition API

Nutrition data in about a second

The model does one thing — food — and runs on a low-latency inference stack, so a typical request comes back in about a second, not the 4–10 seconds a general-purpose multimodal model takes on the same input. That gap is the difference between analysis you can run while the user waits and analysis you have to queue.

In practice, ~1-second responses let you show nutrition the moment a meal is logged, keep an interactive form responsive, and run large backfills without per-call latency dominating throughput. The benchmarks page has the head-to-head numbers.

Send food as text or as an image

Use POST /analyze-food when you already know the food — a logged meal, a menu item, a recipe line. Use POST /analyze-image when you only have a photo and need the dish identified first. Both return the same nutrition schema, so you can mix them in one product without branching your parsing code.

Quickstart

Authenticate with a bearer token and post a food description. Every plan starts with a 7-day free trial.

curl -X POST https://tastyapi.com/analyze-food \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: text/plain" \
  --data "grilled chicken salad, 2 cups"

import requests

r = requests.post(
    "https://tastyapi.com/analyze-food",
    headers={"Authorization": "Bearer YOUR_API_KEY",
             "Content-Type": "text/plain"},
    data="grilled chicken salad, 2 cups",
)
food = r.json()["analysis"]
print(food["nutrients"]["calories"]["amount"],
      food["nutrients"]["protein"]["amount"])

{
  "analysis": {
    "foodName": "Grilled Chicken Salad",
    "servingSize": { "amount": 250, "unit": "g" },
    "nutrients": {
      "calories": { "amount": 320, "unit": "kcal" },
      "protein": { "amount": 34, "unit": "g" },
      "carbohydrates": { "amount": 12, "unit": "g" },
      "fat": { "amount": 14, "unit": "g" }
    },
    "allergens": { "contains": [], "mayContain": [] },
    "dietaryInfo": { "isVegetarian": false, "isVegan": false, "isGlutenFree": true }
  },
  "language": "en"
}

Response fields

Image analysis adds an imageAnalysis object (quality and ambiguity signals). See the full schema in the API reference or the machine-readable OpenAPI spec.

What developers build with it

• Calorie and macro trackers. Let users log a meal in words or a photo instead of searching a database and guessing portions by hand.
• Food diaries and meal planners. Turn free-text entries like "two eggs and toast" into structured, summable nutrition.
• Fitness and weight-management apps. Attach protein, fiber, and calorie totals to workouts and daily goals.
• Restaurant and menu tools. Estimate nutrition for menu items or user-submitted dishes at scale.
• Packaging and label workflows. Generate FDA Nutrition Facts panels from a formulation — see the free nutrition label generator for a working example built on this endpoint.
• Assistants and agents. Give an LLM a single tool that answers "what's in this?" — see the MCP server for a drop-in integration.

Coverage and accuracy

The model is specialized for food — trained on tens of millions of food images and reference nutrition data rather than adapted from a general-purpose model. It handles prepared meals, packaged products, and raw ingredients, estimates portion size from context, and responds in 8 languages.

Values are calibrated estimates, typically within 10–15% of laboratory analysis, and are meant for tracking and informational use — not clinical diagnosis. When a portion is ambiguous, include a quantity in the request (for example, "2 cups") to tighten the estimate.

Authentication and limits

Pass your key as a bearer token on every request. Usage counts against your plan's monthly quota; on an auth or quota problem the API returns a 4xx with a JSON error body so you can surface it cleanly.

{ "error": "Invalid or expired token" }

• 401 — API key invalid, expired, or over its usage limit.
• 402 — no API key supplied, or the subscription needs payment (the body includes checkout details).
• 400 — missing or malformed input.

Quotas by plan are listed on the pricing page; the full request and response reference lives in the API docs.