A keyboard by Escaped Intelligence
RiskCheck
Message Tone Checker for iPhone
A private iPhone keyboard that shows how your message may come across, before you send it: rude, heated, unprofessional or hard to follow. It runs entirely on your iPhone, so nothing you type ever leaves it.
- English
- iPhone
- Runs on the device
- No Full Access needed
- No account
What RiskCheck is
A second look before you hit send
Most messages that go wrong weren’t meant to. A quick reply to a colleague reads as curt. A text written while annoyed sounds angrier than you felt. An email that made sense in your head buries the one thing you needed.
RiskCheck is a text message tone checker built into an iPhone keyboard. It checks the tone of your texts, emails and chats while you type, in any app where you can use a third-party keyboard. It doesn’t write for you. It shows you how your words are likely to land, and leaves the decision with you.
That is the idea behind it: better judgment, one message at a time.
How it works
Four signals, right above the keys
As you type, a slim strip above the keyboard reads your whole message, not just the last word. Each signal stays neutral while things look fine and lights up when it sees a reason to. Tap one to see why.
Risk
Could this hurt someone, or you?
Looks for wording that reads as rude, insulting or threatening, including the kind you might regret once it’s in writing.
Emotion
Heated, calm or warm?
Shows whether the message reads as angry or annoyed, neutral, or warm and appreciative, which is easy to misjudge in a hurry.
Professional
Would you send it to a client?
A 1–5 score for how appropriate the message would be for a colleague or a customer, so a casual text doesn’t slip into a work email.
Clarity
Will they get the point?
Flags very long sentences, walls of text and shouting in capitals, the things that make a reader skim past what matters.
You stay in charge
It points things out. You decide.
RiskCheck shows you first how a message may land, because a rewrite fixes one message, while noticing why it sounded wrong helps with the next hundred. When you do want help with the wording, it can suggest a calmer version, and you choose whether to use it. A firm message is sometimes exactly the right one.
- Nothing changes without you. RiskCheck never sends your text, and it replaces it only when you tap Replace on a suggested rewrite. A signal is information, not a block.
- Context still matters. A blunt line to a close friend is not a blunt line to your manager. Only you know who is reading.
- Quiet by default. Signals appear only when there is something worth a second look, so the strip doesn’t nag on every sentence.
A calmer version, written on your iPhone
Tap the wand above the keys and RiskCheck suggests a calmer version of your message, shown next to yours: the same request, deadlines and facts, without the heat. It is written entirely on your iPhone, by Apple’s on-device language model (part of Apple Intelligence), with RiskCheck’s own model as a backup. Your draft never goes to a server, ours or anyone else’s.
That privacy has a price, and we would rather say so. A model small enough to run on a phone is far less capable than the cloud models other keyboards send your text to, and a keyboard gets less memory and less of the screen than an app. So the rewrite is bounded by your iPhone’s hardware and by what iOS allows a keyboard to do:
- It needs Apple Intelligence. iPhone 15 Pro or later, with Apple Intelligence turned on. On other iPhones the four signals work as usual; the rewrite isn’t offered.
- Read it before you replace. In our tests, about one rewrite in ten changed something that mattered, such as a deadline or who pays whom. RiskCheck warns you when a number, a name or a “not” goes missing, but it can’t catch everything.
- It works with what it can see. It rewrites the text the app shares with the keyboard, which in some apps is only part of a long message. Sometimes it finds nothing to change, and the first rewrite after opening the keyboard can take a few seconds.
For neurodivergent people
Real-time feedback on how a message may land
Tone in text is hard for everyone: no face, no voice, no pause to read the room. For many neurodivergent people, including many autistic people and people with ADHD, working out how a message will come across takes extra effort, and a misread can cost more than it should.
RiskCheck gives you a second opinion while you type. The four signals show how a typical reader may take your message: heated, too blunt for work, or unclear. You see it before you send, not after someone replies.
- It describes, it doesn’t correct you. Communication differences run both ways. RiskCheck shows how a message may be read; whether that matters is your call.
- Private. Everything runs on your iPhone, so nobody sees what you check or how often.
- Calm to use. Four small signals above the keys. No pop-ups, no sounds, nothing that interrupts your typing.
- Honest about its limits. It catches openly rude, heated or unclear wording best. Subtler cues, like sarcasm or passive aggression, are still being worked on.
A real keyboard first
Built to type on all day
A tone checker is only useful if you actually keep the keyboard switched on, so RiskCheck is built to be a good everyday keyboard before anything else.
- Word predictions from an on-device language model that reads your whole sentence and suggests the next word, or finishes the one you started.
- Autocorrect when you press space, including the classic missed space bar (“butbthen” becomes “but then”).
- A familiar layout sized like the system keyboard, with key previews as you type.
- A space bar that knows what you meant: a tap that lands between a letter and the space bar is decided by the word you’re typing.
Privacy
Runs 100% on your iPhone
Most AI keyboards send what you type to a server to analyze it. RiskCheck doesn’t, because it can’t: the language model and the tone checks live inside the keyboard itself. The optional rewrite uses Apple’s on-device model, which runs on your iPhone too.
RiskCheck doesn’t ask for Full Access, the iPhone setting that lets a keyboard reach the internet, so iOS itself keeps the keyboard offline. It has no networking code anyway. So your work emails, private texts and late-night drafts stay on the phone you typed them on.
How we test it
Measured, not guessed
Every signal is checked against thousands of labeled messages from public research datasets, kept apart from the ones it learned from, before it ships. We pay as much attention to false alarms as to misses: a tone checker that flags every firm but polite email is one you’ll switch off.
Honest status: RiskCheck is good at noticing openly rude, insulting or threatening wording today. The subtle cases, like passive-aggressive politeness or a message that is technically fine but cold, are harder, and we’re still training for them. We’ll say so on this page until they work.
Where RiskCheck is going
From one message to how you communicate
A single message is a snapshot. The patterns are more useful. These are the next things we plan to build, each one optional:
- Your communication style over time. See whether your messages are getting calmer, warmer or clearer from month to month.
- Where you’re different. Spot the outliers, such as tense text messages but polite emails, or one app where you’re much sharper than everywhere else.
- When it happens. Notice whether the heated messages cluster late at night, on Mondays or right after meetings.
- A reminder of what you sent. Short summaries of your recent messages, so you can recall what you promised and to whom.
The privacy rules these features will follow
- Off until you turn it on, feature by feature
- Kept only on your iPhone, never uploaded
- Scores and patterns by default, not the text of your messages
- You can see everything stored and delete it at any time
- Typing and tone checks keep working with every insight switched off
What will never happen
- No reading by us, an employer or anyone else
- No selling or sharing of data
- No training of AI models on your messages
- No ads
Questions
Frequently asked
How do I check the tone of a text before sending it on iPhone?
With RiskCheck, you just type. Switch to the RiskCheck keyboard in any app, write your message, and the strip above the keys shows whether it reads as risky, heated, unprofessional or unclear. There is nothing to copy, paste or upload. Without a tone checker, the best manual test is to reread the message as the person receiving it, ideally a few minutes later.
Can it tell me if my message sounds rude?
Yes, that is the Risk and Professional signals’ job. If a text or email reads as rude, insulting or impolite for work, the strip shows it while you’re still typing, so you find out before the other person does, not after.
Does RiskCheck read or store my messages?
It reads what you type in order to check it, the way any keyboard has to, but only on your iPhone. The text is never sent anywhere. RiskCheck has no servers that receive messages, the keyboard contains no code that connects to the internet, and without Full Access, which it doesn’t ask for, iOS doesn’t let it connect at all.
Does it need Full Access?
No. RiskCheck doesn’t ask for it at all. Typing, predictions, all four signals and the calmer rewrite work without it, which means iOS itself keeps the keyboard offline.
Does it work offline?
Yes. The language model, word predictions and tone checks all run on the device, so RiskCheck works the same in airplane mode.
Is RiskCheck helpful for neurodivergent people?
It can be. Many autistic people and people with ADHD say that judging how a text or email will come across takes real effort. RiskCheck gives instant, private feedback while you type: whether a message reads as heated, too blunt for work, or unclear. It doesn’t tell you how you should communicate. It shows how a typical reader may take a message, and you decide what to do with that. Everything runs on your iPhone, with nothing sent anywhere.
Can it tell if I sound passive-aggressive?
Not reliably yet. Openly rude, insulting, threatening or angry wording is what it catches well today. Polite-but-pointed phrases like “per my last email” depend heavily on context, and we’re still working on them. This answer will change when that work ships.
Is it a grammar checker or a rewriting tool?
It is a tone checker first. It has autocorrect and word predictions like any good keyboard, and when you ask, by tapping the wand, it suggests a calmer version of your message. The rewrite runs entirely on your iPhone with Apple’s on-device model, so it is more limited than cloud rewriting tools: read it before you use it. It isn’t a grammar checker.
How is RiskCheck different from Grammarly?
Grammarly is a broad writing assistant: grammar, rewrites, tone detection and more, processed on Grammarly’s servers, which is why its keyboard needs Full Access and an internet connection. RiskCheck does one job, tone and clarity, entirely on your iPhone, including its optional calmer rewrite. We compared Grammarly, QuillBot, Wordtune and Tonely in our guide to the best AI keyboards for iPhone.
Which apps does it work in?
Any app where iPhone allows third-party keyboards: Messages, Mail, WhatsApp, Slack, Teams, Gmail and so on. iOS always switches to its own keyboard for password fields and some phone-number fields, and an app can choose to block third-party keyboards.
Is it available on iPad or in other languages?
RiskCheck starts with English on iPhone. iPad support is planned. Other languages will follow once the English checks are good enough.
When is it available, and how much will it cost?
RiskCheck is in development and testing now. Pricing will be announced with the App Store release. If you’d like to try it early, write to us.
Want to try it before launch?
We’re looking for people who write a lot of messages at work and want a second pair of eyes that never sees their text. Tell us a little about how you’d use it.
RiskCheck is made by Escaped Intelligence, the makers of OwlyMath. iPhone is a trademark of Apple Inc.