Method

How NINE thinks

The rules behind every reading, score and nudge in NINE. The short version: a home app should make a better record for your vet, not pretend to replace one.

What does NINE read from a clip?

Only what a clip can show. After you film, NINE sends the clip to an AI model for review and files what it sees straight into the record:

  • Coat condition
  • Skin marks, described by how they look, never by what caused them
  • Posture
  • Eyes
  • Mouth
  • Breathing effort
  • Stool and the litter box
  • Crying in the litter box
  • When a clip plainly shows them: a fast breathing rate at rest, the number on a scale, body shape, bones that stand out, a filmed scruff test, eating or drinking, and hiding

Each reading is labelled AI and carries its confidence. Only a medium or high confidence is filed; a low-confidence reading files nothing, and NINE asks you instead. You can correct any reading. A sign such as a matted coat can also be kept as your own answer; a breathing count, a scale’s number, body shape or a moment at the bowl stays the AI’s until you log your own.

What does NINE never read from a clip?

  • A calm breathing count. Only a fast count at rest is filed, with its number; a count of 30 or under is never shown as a number. Your own hand count is the one to trust.
  • Weight from your cat’s size. Only the number on a scale in the clip, with your cat alone on it.
  • Body shape through a long or fluffed coat, or anything that has to be felt. A clip shows only bones that plainly stand out, or a scruff test filmed with the release in frame.
  • How much your cat ate or drank in a day, or whether energy is lower than usual. A clip shows a moment at the bowl, or a cat hiding.
  • Mobility and vomiting. You log them as you see them.

Can an AI reading make things look better than they are?

No. An AI reading can raise attention: it can put a signal on watch, or tell you to call your vet. It can never lower or clear a warning, close a follow-up or quiet a reminder. Your own answer always wins its slot.

When the AI notices nothing in a clip, NINE says only what was in view, for example “Seen walking · nothing noted by AI”. That line never counts as reassurance: it counts only as a signal checked toward the score, and it never lowers a warning or lifts a score out of its band.

How is the NINE score made?

The score sums up the last seven days from all nineteen signals, checked by you or read by the AI.

  • The worst signal standing sets the band: call now (0 to 9), vet today (10 to 39), watch (40 to 69) or calm (70 to 100).
  • Inside a band, the number moves with how many of the nineteen signals were checked in the last seven days. Checking more never lifts a score out of its band.
  • Fewer than three checked signals show “too few checks” instead of a number.
  • An open question, such as a check-back you have not answered, keeps the score from reading calm.

It is a summary of your own record, not a health grade or a veterinary assessment.

How do I count my cat’s breathing at rest?

With the hand count: while your cat is asleep, tap once for each breath for 30 seconds. NINE saves the count as breaths per minute on the chart.

One count says little; a pattern over days says more. If your vet has given you a number to watch for, theirs is the one that counts. The hand count is free.

What happens when something looks off?

NINE raises it where you are: the signal turns to watch or to call your vet, with the signs to look for. A sign that means call your vet now is never filed quietly; NINE asks you to look.

If it feels urgent, the emergency check is one tap away, and its answer is always the same: don’t wait for an app, call your vet.

How reliable are NINE’s readings?

NINE makes no accuracy claims, and it is built so that you never have to take a reading on trust. AI readings can be wrong, so each one is labelled AI with its confidence, a low-confidence reading files nothing and asks you instead, and you can correct any reading in a tap.

The rules only work one way: an AI reading can raise attention, but it can never lower or clear a warning, so a missed sign never turns into reassurance. The AI never estimates weight from size and never shows a calm breath count as a number, and the readings from a clip have not been checked against vet-scored or hand-counted clips, so treat a breathing count, a scale’s number or body shape from a clip as a prompt to look, not a measurement. Your own hand count and your own scale are the ones to trust.

We test changes to the AI against a labelled set of cat videos. That is an internal check, not a clinical study, so we do not quote it as an accuracy figure.

Where do my clips go?

To your private record, and to an AI model for review. NINE sends the clip, or a smaller copy, with your narration and your cat’s name to OpenRouter, which passes it to the model provider. Breathing clips and paperwork photos are not sent for review.

We do not sell your data, show ads, or train AI models on your cat. You can export or delete everything.

Does NINE tell me what is wrong with my cat?

No. Wellness monitoring, not veterinary diagnosis. NINE describes what it sees and what you log, such as a matted coat, a squint or a changed posture, and never names a disease. What it means is your vet’s call.

If your cat seems unwell, don’t wait for an app: call your vet.

Back to the homepage