How It Works
A complete guide to using the SafeBite scanner — written for both humans and AI agents.
The 4 Pillars of SafeBite
1. The Traffic Light
The verdict. A massive Green, Yellow, or Red light gives you the bottom line: Safe to eat, Caution — highly processed, or Avoid — contains harmful additives.
2. The Ingredient Decoder
Unpronounceable names, translated. SafeBite picks the most complex chemical names in the product and explains what they actually are — flagging anything linked to cancer.
3. The Fake Food Check
Is it real? SafeBite looks for artificial colors, synthetic flavors, and highly refined oils to tell you whether you are holding Real Food or Engineered Food.
4. The Approval Loophole
Where is it approved? SafeBite flags ingredients banned in the EU or UK but still allowed in the US, and those cleared through the industry self-affirmed GRAS loophole instead of FDA testing.
For Humans
1. Scan a product
Open the Scanner and either capture the product label with your phone camera (point your camera at the barcode), upload a photo of the label, or type the product name and ingredients manually. If the camera fails, a UPC / barcode text input and full manual entry are available right below the camera controls.
2. What the scores mean
Green — Low concern. Few or no risky additives detected. Fine for everyday use.
Yellow — Use caution. Some additives flagged as moderate concerns. Limit intake or consider a swap.
Red — Avoid. One or more high-risk ingredients flagged. Look for a safer alternative.
3. Reading pesticide & heavy metal warnings
Every flag lists the ingredient, a severity level (high / medium / low), and a plain-language message. Pesticide and heavy-metal warnings highlight ingredients linked to residue or contamination concerns — these often matter most for children, pregnancy, and frequent consumption. The AI analysis below your score suggests safer swaps and certifications to look for (e.g. USDA Organic, Glyphosate Residue Free).
For Agents
SafeBite is a frontend application — there is no public JSON API to POST to. Agents should (1) discover the app at /llms.txt (see also /api-docs), (2) submit scans through the scanner flow at /scanner, and (3) parse the response shape below.
Submitting a scan
A scan payload includes product_name (required), ingredients (recommended for an accurate score), plus optional brand, category, and upc (barcode context). SafeBite does not resolve a UPC to a product on its own — pair the barcode with the ingredient list.
Parsing the response
{
"product_name": "Organic Oat Cereal",
"overall_rating": "Green",
"score": 7.4,
"assessment": "green",
"cancer_risk_ingredients": [],
"banned_in_eu": false,
"gras_loophole": false,
"is_engineered_food": false,
"reasons": [
{
"type": "positive",
"severity": "positive",
"message": "Whole-grain oats listed as the first ingredient."
}
]
}Example response — sample data, not a live result.
Fields: overall_rating (the traffic-light verdict: "Green" | "Yellow" | "Red"), cancer_risk_ingredients (names flagged as cancer-linked), banned_in_eu (EU/UK-banned but US-allowed), gras_loophole (industry self-affirmed GRAS), is_engineered_food (artificial colors / synthetic flavors / refined oils), plus score (0–10, higher is safer), assessment, and reasons[] (typed flags with severity and a plain-language message). Scores come from SafeBite's local risk dataset, modeled on EWG, CSPI, FDA, EFSA, and USDA-style science — not live agency feeds.
