The 2am Problem: Why Mental Health Apps Fail Exactly When You Need Them

Here is a list of places people have told us they opened a mental health app.

A night bus. An aeroplane at cruising altitude. A hospital corridor with concrete walls. A car park after a difficult conversation. A rural road at midnight. A basement flat with one bar. A phone at 4% with data throttled because the month had run out.

Now here is what most mental health apps do in those places: nothing.

Not degraded. Nothing. A spinner, a retry button, or a blank screen. Because most of them are a website in a costume, and a website needs the internet.

Why the overlap is not a coincidence

It would be one thing if bad connectivity and acute distress were independent. They are not, and the correlation runs in a direction that should worry anyone building in this category.

Distress happens at night, when you are more likely to be somewhere other than your home wifi.

Distress happens in transit, on trains through tunnels, on planes, in cars, in the specific limbo of travelling somewhere you do not want to go.

Distress happens in institutional buildings, which are made of concrete and have famously terrible reception: hospitals, courts, police stations, care homes.

Distress correlates with precarity, and precarity includes throttled data, prepaid plans that have run out, and phones that are three years past a battery replacement.

None of this is exotic. The moments when a person most needs something to be there are systematically the moments when connectivity is worst. Building a product that requires a network is therefore not a neutral technical decision; it is a decision to be absent during a predictable subset of the moments the product exists for.

What we did about it

Version 1.8.5 shipped full offline support. Specifically:

All nineteen Toolkit practices work with no connection. Breathing, grounding, the panic and anger and rumination sequences. No network call, no degraded mode, no "some features unavailable" banner.

All seven guided meditations are on the device. Audio, not streamed. Five companion narrators, seven languages, between 2:56 and 4:47 each.

Your history is readable offline. Everything you have written and everything the companions have said. Conversations live in an encrypted local database rather than a cloud table, which is a privacy decision that turned out to have this as a side benefit.

Widgets keep working, because they are drawing on local data.

Details are on the offline page.

The part we cannot solve, stated plainly

The companions do not work offline. The conversation, which is the centre of this product, requires a connection, because the model runs remotely.

We are not going to bury that. A device-sized model that could hold five distinct personalities, seven languages, and a memory of your last six months would not fit on a phone at any quality worth having, and shipping a much worse companion so we could claim offline conversation would be a marketing decision rather than a product one.

So what ILTY actually offers with no signal is: everything except the talking. That is a real and honest limit, and it is the reason the Toolkit exists in the shape it does. The offline feature is not the conversation minus the network. It is the thing to do when the conversation is unavailable, which turns out to be the same thing you need when the conversation is available but you cannot form sentences.

Why almost nobody does this

Three reasons, and only one of them is respectable.

It is genuinely more work. Offline means local state, sync conflicts, cache invalidation, storage limits, bundled audio, and testing a matrix of connectivity states rather than one. Every feature costs more. This is the respectable reason and it is real.

Streaming is easier to control and to meter. If the content lives on your server, you decide who gets it and you can change it after the fact. If it is on the device, it is on the device. For a business model built on metering access, offline is actively inconvenient.

Nobody measures it. No app store listing shows it. No review mentions it, because reviews are written on wifi. It never appears in a comparison table. A feature that only matters in the worst moments is invisible in every context where people evaluate apps, which means the incentive to build it is close to zero.

That last one is the actual explanation, and it is why we wrote this post. offline mental health app has no measurable search volume at all, which we checked. Nobody is looking for this until the moment they need it, and in that moment they are not searching, they are finding out.

How to check any app, in ten seconds

Do this now, with whatever you have installed, rather than finding out later.

1. Open the app once on wifi so it is warm.
2. Turn on aeroplane mode.
3. Open it again.

Then note which of these you get: full function, partial function with a clear indication of what is unavailable, or a spinner. The third is the common result, and the second is rarer than it should be.

Worth doing for whatever you would actually reach for at 2am. Including ours.

What to do when nothing works

Since this post is about the failure case, the honest ending is not a download link.

Save 988 in your phone as a contact. In the US, call or text. Voice calls work on signal so weak that data will not move at all, which is a genuinely useful property and the reason a phone number beats an app in the worst conditions.

Learn one thing that needs no device. The physiological sigh is two inhales and a long exhale, and it is in that post for free. 5-4-3-2-1 grounding needs nothing but your eyes. Neither requires a battery, an app, or a subscription. Knowing one of them by heart is worth more than any app on a night bus with a dead phone.

Write the number on something physical. The person you would call. Somewhere that is not the device that might be at 4%.

An app that works offline is better than one that does not. Something you have memorised is better than both.

Frequently asked questions

Does ILTY work offline? Partly, and the split is clean. The Toolkit's nineteen practices, all seven meditations, your full conversation history and the widgets work with no connection. The companions do not, because the model runs remotely.

Why can't the AI companions work offline? An on-device model small enough to ship would be dramatically worse across five personalities, seven languages, and long-running memory. We would rather name the limit than ship a hollow version of the main feature and call it offline support.

Do the meditations stream or download? They are on the device. Seven meditations, five companion voices, seven languages, 2:56 to 4:47 each. No connection needed to play them.

Where are my conversations stored? Encrypted locally: iCloud Documents with full file protection on iOS, Room over SQLCipher with a Keystore-wrapped key on Android. Not sitting in a cloud table with your name on it. More on the privacy page.

Which other mental health apps work offline? Test them yourself with aeroplane mode; it takes ten seconds and is more current than anything we could publish. In our hands-on testing, anything conversational needs a connection for the conversation. What varies is whether the rest of the app survives without one, and most of it does not.

---

ILTY's Toolkit and meditations are on your phone, not on our server, and they work at 30,000 feet and in the hospital corridor. The conversation still needs signal, which we would rather tell you here than have you discover at 2am.

Download ILTY, then turn on aeroplane mode and see what is still there.