An AI chat for OCD, sticky thoughts and endless checking

You’ve been down this rabbit hole before.

You know how it ends. The thought comes back, you ask, you get an answer, it holds for ten minutes, and you’re back. Meet Maybe, an AI chat built for that exact loop. It won’t hand you the answer, however many times you ask. It stays with you while the urge passes, and its name is the only answer it will ever give.

Start the assessment How it works

Seven days free. Card up front, first charge on day seven, cancel any time before then. If you do pay and it isn’t for you, 30 days to get it back.

Sealed on your device before it’s sent. We never hold the key, so nobody here can read a word.
Built on NICE guidance and the International OCD Foundation, and cited claim by claim.
Opens in any browser and installs to your home screen. No app store, and nothing in a purchase history.

Maybe3:12am

You: I know I’ve asked this before. I just need to be sure it didn’t mean anything.
Maybe: You’ve asked, and you noticed you’re back. That’s the loop doing its job, so I won’t answer it. I’m not going anywhere, though. How loud is it right now, 0 to 10?
You: An 8. It feels so real this time.
Maybe: It always feels real. That’s the one thing it’s reliable about. Let’s stay with the 8 for a few minutes without doing anything about it, and watch what happens.

An example, not a transcript. Real conversations are encrypted before they leave your device, so there is no transcript we could show.

If you recognise most of this, you’re in the right place.

A thought you never asked for, back again
Hours of googling, or asking ChatGPT, that only ever buy you ten minutes
Asking someone you love the same question, worded slightly differently
Checking, counting, praying or replaying it until it feels right, then once more
An answer that would settle anyone else. Not this
A memory that keeps resurfacing to ask whether it was your fault
Steering around the places, people or news that set it off
Lying awake at 3am wondering if you're overreacting, or making it all up

You don’t need a diagnosis to be here, and you don’t have to decide whether it’s “bad enough”. The whole of Maybe is built on published clinical guidance for intrusive thoughts and obsessive-compulsive disorder. It works the same either way.

See what it actually does.

Four things, drawn the way they work rather than described. None of it is a screenshot of your data, because there is no readable copy of anyone’s for us to screenshot.

The chat

It will not answer the question

You pick which one you need. Ride It Out is for the middle of a wave, short replies and no picking the thought apart. Learn & Practice is for the calmer hours, where the useful work happens.

Neither will settle it for you. Ask a second time and the refusal comes back the same way it came the first time, because a freshly worded no is fresh material to analyse.

Ride It Out

It’s loud right now.

You: I just need to be sure it didn’t mean anything. Then I’ll stop.
Maybe: I’m not going to answer that one. Answering is the hook. I’m not going anywhere, though. How loud is it right now, 0 to 10?

Learn & Practice

A steadier moment.

You: Does it still count as checking if I only do it in my head?
Maybe: It counts. Same job, quieter, and harder to notice. Want to put that one on a ladder?

Practice

A ladder, and you climb one rung

This is the exposure part, and it is what published guidance puts first: take on the smallest version of the thing you have been steering around, let the feeling come down on its own, then take the next one up.

The chat builds the ladder out of what you have told it and shows every change as a card you accept or decline. When a rung is done you say how it went, in three words. There is no score and no streak, because a streak is one more thing to check.

Front door checkingSmallest rung first

Stand at the locked door for a minute without touching the handle
Done, and it went: Easier than expected
Leave the house without going back to check
Done, and it went: About as expected
Let someone else lock up
Next one up
Leave for the whole weekend without a last check

An example ladder, not anyone’s. Yours is sealed on your device like everything else, so there is no real one we could show you.

Learn

Three lessons that show their sources

Why thoughts stick, watching instead of wrestling, and facing it on purpose. Four or five minutes each, in plain words, with a short check at the end.

Every clinical claim carries its source in the sentence that makes it, from NICE guidance and the International OCD Foundation. The marks are real links: tap one and you land on the guideline itself.

Why thoughts stick · about 4 minutes
Watching, not wrestling · about 4 minutes
Facing it on purpose · about 5 minutes

The opening line of lesson one, as it appears

Nearly everyone gets intrusive thoughts.NICE CG31 IOCDF

Limits and safety

Built to be used less

You set five ceilings on your own use before you start. Tightening one takes effect straight away. Loosening one waits 24 hours, so the version of you that is spiralling can’t overrule the version that set it. The server holds the line, not the app on your phone.

Get help now sits at the top of every screen, locked or unlocked, and opens crisis lines that answer at any hour. Nothing decides when you have earned it, and opening it is never recorded.

Conversations a day3
How long one runs30 min
Gap before the next one60 min

Tightening any of these lands now. Loosening one waits 24 hours.

Get help nowEvery screen, always

Three steps, and the first one takes a few minutes.

Start with the assessment
A few questions about how the loop shows up for you, and how much of this language you already know. Nothing is diagnosed, nothing is scored, and you are never asked to call yourself anything.
Get your ladder and your lessons
What you said shapes the first practice ladder and the worked example in each lesson, so the three of them are about your situation rather than a stranger's. The chat proposes the ladder as a card you accept or decline.
Then it is there when it hits
Open it in the middle of a wave, at whatever hour that is. It will not answer the question, it will not get tired of saying so, and it holds you to the limits you set when you were calm.

Start the assessment

Seven days free. Card up front, first charge on day seven, cancel any time before then. If you do pay and it isn’t for you, 30 days to get it back.

No results to show you yet. This is what it’s built on.

Getting the answer brings the feeling down faster than waiting it out, and the next spike lands higher than the last. That’s why it never feels finished. Replaying it in your head counts as checking too.

The whole app is built around the other line: staying with the feeling long enough to watch it fall on its own. That is the practice published guidance puts first, and it is where the lessons get both their shape and their citations. Explained from scratch here.

Check, and it drops faster now but returns higherDon’t, and it peaks once, then fades

What we can’t tell you

There are no recovery rates on this page, no percentage who got better, and no study of our own. We haven’t run one, and a number we made up would be worth less than nothing to you.

No independent clinician has reviewed this product yet either. Our terms say that plainly, and the sentence changes when one has. What we can point at is the guidance underneath it: NICE CG31, the treatment guidance the NHS works from, and the International OCD Foundation’s plain-language material. Every claim in a lesson carries its source in the sentence that makes it.

The evidence page walks each claim against the source it came from, and says which parts are not proven. The sources themselves are free to read whether or not you ever sign up.

What it won’t do, and who does it better.

Not because it’s being withholding. Because every one of these is the thing that feeds it, and a chatbot is the most patient answer machine ever built. It will never get tired of you, which is the problem.

If you’ve been asking ChatGPT and it keeps helping for about ten minutes, that’s what this is. More on why that happens.

Tell you the thought means nothing about you
Say what you are or aren't
Help you work out whether it really happened
Give you a fresh answer because you asked a fifth time
Read you generic advice off a script

Maybe compared with a general AI chat and with seeing a therapist
What is being compared	Maybe	A general AI chat	A therapist you can see
Will it answer the question?	No, and it says why each time	Yes, every time you ask. That is the trap	No. Leaving it unanswered is the work
Who can read what you write	Nobody here. Sealed on your device, and we never hold the key	The provider, under whatever its data settings say	Your therapist, which is the point of going
When you can reach it	Any hour, from any browser you can open	Any hour	Your appointment, and often a waitlist before the first one
What it is built on	NICE CG31 and the International OCD Foundation, cited in the sentence that makes the claim	Nothing written for this	Training, supervision, and years of doing it
Can it hold you to a limit?	Yes. You set the ceiling, and loosening one waits 24 hours	No	They set the pace with you
What it cannot do	Notice you are in trouble. Nobody here is watching, because nobody here can read it	Stop feeding the loop	Be there the moment it hits

If you can get to a specialist who treats this, go. A therapist trained in this way of practising is the best thing there is, and nothing on this page replaces one. This is built for the hours they can’t be in, and for everyone still on a waitlist, priced out, or not ready to say any of it to a person yet.

The longer comparison sets us against four other options, every claim sourced from their own published documents, including the places they beat us.

There are no reviews here, and no invented ones either.

We haven’t launched. There are no users to quote, no star rating and no download count, and making some up would be a strange way to start a product about not being handed comfortable answers. Two things sit here instead: the person who built it, and a guarantee that means you don’t have to take his word for any of it.

“I didn’t build this as a business idea. I’ve been the one awake at 3am, asking the same question worded slightly differently in case this time the answer held. It never did. What helped wasn’t a better answer, it was learning to stop needing one. That’s what Maybe is, for whoever is awake right now.”

Luke, who built Maybe

If it didn’t help, take the money back.

Seven days free before anything is charged. If you do pay and it wasn’t for you, email us within 30 days and we send it back. You don’t have to explain, you don’t have to prove you used it, and nobody will ask you to stay.

This is the thing people spend money on when they are not doing well. It should be the easiest money in the world to get back.

Built so we can’t read you.

This loop runs on the thing you have never said out loud. You can’t say that to something you think is being read, and a tool you lie to is worthless. So the encryption here isn’t a badge on a feature list. It is the thing that makes the product work at all.

Everything you write is sealed on your device with AES-256 before it is sent, and the key never leaves that device. We don’t hold a copy, we can’t recover it for you, and we couldn’t hand over a conversation if we were asked, because we don’t have one to hand over.

Every product in this space says “encrypted”; who holds the key is the question that separates them. Ask it of anything you are considering, including us.

Messages, conversation titles, the memory it keeps and your practice ladders are all ciphertext in our database. There is no readable copy anywhere on our side.
A test runs a whole conversation, then searches every collection for the words that were in it. It fails if it finds any.
The chat route writes no logs and keeps no cache, and a second test fails if anything on that route reaches a host outside the model provider.
No ads, no advertising partners, and no profile of your symptoms, because there is nothing readable here to build one from.

One exception, said plainly rather than buried: writing a reply means the model reads that message. It is retained by no one, and it is the only moment your words exist in the clear anywhere but your own device.

The cost of this is real and you should know it before you start. Nobody here is watching over you. If you were ever in danger, we could not see it and could not reach out. An app that watches over you has to be able to read you; this one chose not to be able to.

Exactly what we can and can’t see

Questions worth asking before you start.

Will it tell me the thought means nothing about me?

No. That is the answer you have been chasing, and you already know how long it holds. It names what is happening instead, and stays with you while the feeling comes down on its own.

Do I need a diagnosis?

No, and you are never asked to call yourself anything. It works the same whether you have a diagnosis, a suspicion, or neither.

Can you read what I write?

No. Your words are sealed on your device before they are sent, and we do not hold the key. There is one honest exception: writing a reply means the model reads that message, retained by no one. The privacy page lists the rest of what we can and cannot see.

What does it cost, and when am I charged?

Seven days free, with a card up front. The first charge lands on day seven, and cancelling before then costs you nothing. After that you have 30 days to ask for a payment back, with no questions and no reason needed.

Is there an app to download?

No, and no store listing either. It runs in the browser you already have, and your browser can add it to your home screen so it opens like an app. Nothing about it lands in a purchase history, and the card statement shows a neutral name.

Is this therapy?

No. It is structured practice built on published clinical guidance, and it is not therapy, not a diagnosis and not a crisis service. If you can get to a specialist who treats this, go. This is built for the hours they cannot be in.

Wherever you can open a browser.

There is nothing to download and no store listing. It opens in whatever browser is on the machine in front of you, phone or laptop or the desktop at work, and your browser can add it to your home screen so it opens like an app.

A purchase history is a list other people can see: a shared account, a family plan, a phone somebody borrows. Nothing about this lands in one, and the card statement shows a neutral name.

It does need a connection. There is no offline mode, and that is a decision rather than a gap: an offline copy means a readable copy of your conversations sitting on the disk, which is the one thing this whole design refuses.

The next one is coming. This time, you don’t have to answer it.