Mura AI review: what it actually does, and what it does not

Written by the people who build it, so weigh it accordingly. What is in the app, verified against the current build, what it cannot do, and the two features that are switched off because they do not work well enough to ship.

If you searched for Mura AI you probably want three things answered: what it is, whether it works, and what the catch is on something free.

This is written by the people who build it, so weigh it accordingly. The useful thing we can offer is not praise, it is specifics, including the parts that do not work yet.

What it is

Mura is a free mental health app at mura.today. You talk to it, by voice or by typing, and it responds in a structured way drawn from cognitive behavioral therapy: naming the thought, checking it against the evidence, looking for a reading that fits the facts better.

Around that sit a set of screening tools for physical things that produce symptoms people read as purely psychological.

What is actually in it

Verified against the current build, not a roadmap:

Conversation, by voice or text, with a mood check-in before and after

Thyroid screen, a 30 second camera scan using face mesh landmarks to measure eyelid position, blink rate and asymmetry

Gut health screen, 13 questions covering function and drivers

Testosterone screen and nutrition screen, questionnaire based, scoring risk of specific deficiencies

Sleep tracking, which records through the night on your phone's microphone and reports breathing pauses, snoring and estimated respiratory rate

Anemia eye-scan, a photograph of the inside of your lower eyelid

Camera vitals, heart rate, heart rate variability, breathing rate and a stress score, read from tiny color changes in your face

Breathing tools, 4-7-8 and box breathing

You can use it as a guest before signing up.

What it is not

It is not a licensed clinical service. It does not diagnose anything and it cannot prescribe. For severe depression, psychosis, or active thoughts of suicide, it is not a substitute for human care and nobody should treat it as one.

The screens are screening. They are built to raise a question worth asking, not to answer it. A thyroid screen flagging "possible" means ask for a TSH test, not that you have a thyroid condition.

And an AI is not a therapist. It is available at three in the morning, it does not get tired of you, and it will not be shocked. Those are real advantages. It also does not know you, cannot notice what you are not saying, and has no stake in your life. That ceiling is real.

The limits we would rather you heard from us

Voice-based depression scoring is built but not running. The model is there, it was trained on 35,000 real users, and it does not fit on our server. It needs about 1.75GB for a single pass and the machine has 1,967MB in total, so it is switched off rather than half working. If you see voice analysis described anywhere as a Mura feature, this is the honest status.

Blood oxygen was removed entirely. Measuring it needs an infrared channel at 940nm and phone cameras do not have one. Ours returned the same floor value every time, so it is gone. Any app claiming to read SpO2 from a normal camera is showing you a number it cannot measure.

Camera vitals are noisy. They need decent light and a still face. The reading is gated on a signal quality check and returns nothing rather than a guess when the signal is poor, which is the right behavior but means it sometimes gives you nothing.

The eye scan is not a blood test. It looks at the color of the tissue inside your lower eyelid, which pales when hemoglobin is low. It can flag a possibility. Ferritin, from a lab, is what actually answers the question.

What free means here

No payment, no card, no trial that converts.

More usefully, here is a specific thing we do with your data. The app's web addresses name conditions: /thyroid, /gut-health, /testosterone, /sleep. Standard analytics would record the full address of every page you open, which would pair a persistent identifier with a health topic. So the analytics address is pinned to the site root in two separate places in the code. Our own analytics cannot tell which screen you opened. That was a deliberate decision and it costs us product insight we would otherwise have.

Who it suits

People who want to think something through at an hour when nobody is available

People on a waiting list who want something in the meantime

People who suspect something physical might be involved and want a cheap way to form the question before spending money on tests

People who find it easier to be honest with something that cannot judge them

Who it does not suit

Anyone in crisis, who needs a person

Anyone wanting a diagnosis, which it cannot give

Anyone who wants to be told what to do. It is built to help you work out what you think, which is slower and less satisfying than advice

The verdict

It is free, so the question is not whether it is worth the money. It is whether it is worth the time.

A single conversation will tell you whether this way of thinking suits you, and the screens take a few minutes each and occasionally surface something that changes the picture entirely. What it will not do is replace a person, and we would rather say that here than have you discover it at a bad moment.

General information only, not a diagnosis.

This article is for general information and is not a substitute for professional care. If you are in crisis or thinking about harming yourself, please contact your local emergency services or a crisis line. See crisis lines.