How Feast works

A useful food record is built, not merely recognized.

Feast begins with a fast first draft, gives corrections the same status as model inference, and uses history for questions a single meal cannot answer. These guides explain the workflow, the evidence boundary, and the failure test for every capability.

01Observe

A photo, text, voice action, or shared record supplies evidence. None is a direct measurement of every ingredient or gram.

02Reconcile

Correct ingredients, brand, preparation, portion, time, or photos and reanalyze. Contradictions deserve review, not concealment.

03Learn

Use repeated records to study nutrition and sleep patterns, then separate an interesting association from an individual causal claim.

1 · Build a defensible meal record

Capture evidence, then add what the image cannot know.

These guides cover the core estimation loop: photograph, review, correct ingredients and amounts, and preserve enough context for the record to remain useful later.

2 · Interpret the history

Move from one noisy estimate to patterns worth questioning.

A single meal can be wrong in several ways. These guides explain nutrient views, longitudinal coaching, sleep associations, and how to ask follow-up questions without turning a pattern into a diagnosis.

3 · Put Feast where the habit happens

Reduce friction without removing judgment.

Apple integrations, widgets, shortcuts, Watch actions, and shared meals can shorten the path into Feast. They do not make incomplete context complete, so each guide keeps the human review step visible.