What it actually is

Carbohydrate low enough to shift the body to running mostly on ketones — usually under about 50g a day, which is stricter than most people expect and is the reason it's frequently done accidentally wrong. It's a genuine medical therapy for some epilepsy, done under supervision. For weight, the long-term trials put it roughly level with other approaches, with adherence as the deciding factor. The first fortnight commonly involves feeling distinctly unwell, which is expected and temporary.

Why this is hard to keep track of

  • The threshold is stricter than most people assume, so weeks get spent near it rather than under
  • The first fortnight feels awful and nobody tells you how long it should last
  • Macro spreadsheets are abandoned inside two weeks, taking the record with them
  • Eating out is genuinely hard, and it's where the record goes quiet

Worth recording

  • What you ate, in plain words
  • How the first two weeks felt, day by day — this is the part worth having later
  • Energy, sleep and headaches, which move early
  • The meals out and the slips, honestly. They're the most informative entries you have

Cadence

A sentence a meal, and a note on how the day felt

In practice

Say it once, or photograph it

The entries most worth having are the ones made at the exact moment nobody wants to fill in a form. A sentence — typed, dictated, or replaced entirely by a photo of the plate — is a much lower bar, and it carries the context a calorie field would have thrown away.

Keto Chat

Day nine. Headache most of the morning, better after lunch. Eggs and bacon, tuna salad, pork chop with cauliflower.

Called 5 tools

  • log_meal(items: ["eggs", "bacon"], meal: "breakfast")
  • log_meal(items: ["tuna salad"], meal: "lunch")
  • log_meal(items: ["pork chop", "cauliflower"], meal: "dinner")
  • log_symptom(type: "headache", severity: 2, part_of_day: "morning")
  • get_history(metric: "symptoms", window: "14d")

Logged. Headaches on six of the last nine days, all in the morning, and the severity you've recorded has dropped from 4 to 2 across that run. Energy is up from an average of 2.8 in days one to four to 4.1 in days six to nine.

Every figure here was counted from entries you made. The model chose which tools to call — it doesn't score the meal, estimate calories from a photo, or decide what any of it says about you.

And then ask

The question the log can answer

Months of entries hold real answers to the things you've been guessing at, and reading them properly has always meant a spreadsheet and a free evening — which is exactly what nobody has.

Ask in plain language. The answers come back with the counts behind them, including when those counts are too thin to mean anything yet — which is the more useful answer far more often than people expect.

Asking about the pattern Chat

Is the tiredness lifting, or am I just getting used to it?

Called 3 tools

  • get_history(topic: "keto", window: "6m")
  • find_pattern(group_by: "context")
  • compare_periods(metric: "entries")

It's lifting on your own numbers. Week one averaged 2.9 for energy with headaches on five days. Week two: 4.3, headaches on three. Week three so far: 5.1, none. Your pre-keto baseline in the log was 5.4, so you're close to it. Sleep went the other way in week one and has recovered too.

Counts and averages over your own entries, nothing more. Where the data is confounded or too thin, the answer says so instead of rounding it into a conclusion.

What it does

Set up around keto

The same mechanics as everything else here, pointed at what actually matters for this one.

Log the protocol as you're living it

What you ate, in a sentence, plus whatever this particular protocol turns on — a window, a training day, a threshold. No spreadsheet to maintain.

The parts that drift

Windows stretch, thresholds creep, and the rule you're following in month three is rarely the one you started with. Drift is obvious in a record and invisible from inside it.

Whether it's actually doing anything

The question every protocol dodges. Answered here by comparing your own weeks against each other, rather than against a claim from a website.

Straight about the evidence

Where the research is strong we say so, and where a popular claim doesn't hold up we say that too. You can get the marketing version anywhere.

The weeks it slips are the useful ones

Adherence is what separates protocols that work from protocols that don't. The days it came apart, and what was happening on them, are the most informative entries in the whole log.

Something to review with someone

Several of these interact with medication. A real record of what you ate and how it went is what makes that a conversation rather than a guess.

The reason a lot of people start

Something to actually bring to the appointment

You get fifteen minutes, and the first question is always some version of "so what does a normal week look like?" — which nobody on earth can answer accurately from memory.

What this gives you: A record of what you actually ate and how it went, which matters more here than most places because this is a protocol a clinician may want to review.

  • Export the date range you were asked about, as a file you can print
  • Photos alongside the entries, which say more than any description
  • Counts and timings rather than an impression of the worst week
  • Nothing shared anywhere unless you choose to share it

What it won't do

  • Score a meal, or sort food into good and bad
  • Estimate calories from a photo and present the guess as a number
  • Set you a calorie target, a macro target, or a weight
  • Run a streak you can break, or tell you that you've fallen behind
  • Comment on your body. The photos here are of food
  • Diagnose anything, or suggest that a pattern is a diagnosis

It's a notebook that can do arithmetic and answer questions about itself. Deciding what any of it means is a job for you and the people advising you.

Early access

Start with what you're already trying to hold in your head

Tell us what you've been trying to keep track of and we'll set it up with you — recording the thing that's actually useful, at a cadence that survives a bad week.

  • Founding accounts lock their price for life
  • No card, no trial clock, no notification designed to make you feel behind
  • Export everything you've logged, photos included, whenever you want

No card required. Unsubscribe any time.

Keep the record without it becoming another job

Founding pricing is locked for accounts opened during early access.

Founding accounts lock their price for life. No card required.