Built by CipherGuard Labs

Is it OKako?

A camera, a strict food list, and an instant verdict — built so eating out doesn't require a spreadsheet when the rules genuinely can't bend.

Actual output — not a mockup

Recommended

Grilled salmon with lemon & olive oil

Matches approved ingredients directly.

Modify

Mixed greens, honey mustard dressing

Dressing contains added sugar.

swap → oil & vinegar

Avoid

Roast chicken, garlic mashed potato

Potato is a restricted starch.

swap → grilled asparagus

The problem

Restaurant menus and packaging labels don't come with an answer key. Most food-tracking apps solve this by estimating macros from a dish description — which introduces exactly the wrong kind of uncertainty when someone is following a strict, medically-guided dietary protocol rather than casually watching their carbs.

OKako doesn't estimate. It matches.

How it works

  1. Capture

    Point the camera at a menu, or at a packaged food's ingredients panel.

  2. Read

    Apple's Vision framework extracts every line of text directly on the phone. No photo is ever uploaded — only the recognized text leaves the device.

  3. Match

    A fast keyword pass checks each item against an approved / restricted / conditional list. Most items resolve instantly, at no API cost.

  4. Reason

    Anything ambiguous — an unfamiliar prep style, an ingredient synonym the keyword list hasn't seen — falls back to Claude, constrained to classify strictly from that same list and to suggest substitutions drawn only from it.

Built with

iOS
SwiftUI, on-device OCR via Vision framework
Backend
FastAPI + PostgreSQL, deployed on Railway
Classification
Hybrid keyword matcher with a Claude API fallback, called only for ambiguous items
Data model
List-matching, not nutrition calculation — no macro database or licensing required

Why list-matching, not calculation

A ratio computed from a vague menu description carries real error — hidden oils, an unlisted sauce ingredient, a portion that varies by kitchen. For a protocol with no room for drift, that's the wrong kind of confidence to hand someone. Matching against a clinician-owned list is slower to build and less impressive on a spec sheet, but every verdict traces back to a rule a person can actually see and edit — which is what a strict protocol calls for.