AI Tone Changer
Rewrite text in a formal, casual, friendly, or confident tone.
This tool sends the text you enter to an AI provider to generate a response. Don't paste secrets or personal data. No files are uploaded.
Pick one of eight registers — Professional, Casual, Friendly, Confident, Empathetic, Formal, Enthusiastic or Diplomatic — and the message you pasted is rewritten to match it while still saying the same thing.
How to use it
- 1Paste the message you want to re-pitch.
- 2Choose the target tone; Professional is the default.
- 3Press Rewrite, compare it against your original, and copy it only if it still says what you meant.
Example
- Input
- your invoice is late again, sort it out — Tone: Diplomatic
- Output
- The same request in courteous business English: a polite reference to the overdue payment and a clear ask.
Only the register is meant to move. The model is instructed to preserve meaning and key facts, so a chased invoice stays a chased invoice — but read it before sending, because tone changes can soften a deadline into a suggestion.
What happens to your data
Pressing Rewrite posts your message and the chosen tone to our /api/ai route, which relays them to an AI provider — the text leaves your device, which is the trade for having a model rewrite it. Nothing is stored on our side, but do not paste anything you would not put in an email to a stranger.
Last updated August 2026
The message is written, it is factually right, and you still have not sent it. Read back, it lands harder than you meant — or so softly that the request inside it has gone missing. That gap between what you meant and how it reads is what a tone change is for, and the job is narrower than it sounds.
Decide first whether you have a tone problem or a content problem, because only one of them is fixable by rewording. Changing register keeps the facts and the ask and moves the voice around them. If the message is missing a date, aimed at the wrong person, or buries the request in the fourth paragraph, a warmer version is a better-sounding version of the same broken message — that is a rewrite for structure, which is a different job.
The mistake that costs most is losing the ask. Politeness is largely made of hedges — when you get a chance, no rush, whenever suits — and a rewrite reaching for warmth adds them by default. The instruction behind this asks the model to preserve meaning and key facts, and a deadline that survives as a suggestion has technically been preserved. Check the result for the date, the amount and the consequence specifically, rather than reading it for whether it sounds nice.
Two ceilings are worth knowing before you paste. One submission is capped at 40,000 characters and anything longer is refused rather than trimmed, while the reply is capped at 2,000 tokens — long enough for an email, not for a document.
How it works
Toolvore builds one instruction from the two boxes — rewrite the following text in a professional tone, with your message underneath — and posts it to an internal /api/ai route, which forwards it to Anthropic's API and streams the answer back a few characters at a time. The model works under a one-line brief: match the requested tone, preserve meaning and key facts, return only the rewritten text. Nothing else travels with it — no earlier attempt, no recipient, no house style, no note about which words are load-bearing — so all of that is inferred from the text itself, and the inference is weakest where it matters most: unfamiliar names, jargon it decides to translate, figures sitting inside a sentence it decides to restructure. It is not deterministic either, so the same message and the same register give two different rewrites and there is nothing to reproduce. The reply is rendered as Markdown, which is worth knowing if your original used asterisks or hashes for something other than formatting, and Copy hands over the raw text the model wrote rather than the formatted version on screen.
Common use cases
- Softening a chasing email about an overdue invoice before it goes to a client
- Turning terse notes typed in a hurry into an update the wider team can read
- Making a complaint firmer without letting it turn into an argument
- Rewriting a rejection or a piece of bad news so it reads as considerate rather than curt
- Taking the exclamation marks out of marketing copy for a formal audience
- Re-pitching the same announcement once for a board and once for a group chat
- Checking whether something drafted while angry can be said calmly
Frequently asked questions
How do I make an email sound more professional without sounding cold?＋
Professional reads as cold when the warmth is taken out and nothing replaces it. The fixes are small: open on a line that acknowledges the person rather than the demand, use their name once, and state what you need in a plain active sentence. Contractions are not unprofessional — removing them is most of what makes a rewrite sound stiff. If a result comes back correct but frosty, the usual cause is that every hedge went at once. Put one back by hand; a single line of context normally thaws the whole message.
What is the difference between a formal tone and a professional one?＋
Formal is about distance and grammar; professional is about clarity and competence. A formal message avoids contractions and slang, keeps sentences complete and addresses people by title — it suits complaints that may be read by a third party, legal notices and writing to institutions. A professional message can say we will, or thanks for the nudge, and still be right for a client you speak to weekly. Formal is the safer default for a stranger, professional for someone you know. Choosing formal for a familiar colleague tends to read as a signal that something is wrong.
How do I chase something overdue without sounding rude?＋
Separate the fact from the feeling. The fact — an amount, a date, an agreed term — belongs in a short sentence with no adjectives, and it is not rude however plainly it is put. The feeling is what reads badly: again, still, as I already mentioned. Cut those and the same chaser becomes businesslike. Keep one deadline and one consequence, because a polite rewrite tends to soften both, and a chaser with neither is a reminder to ignore. Close with a specific ask by a specific date rather than at your earliest convenience.
Does changing the tone change what a message means?＋
Only the register is meant to move, and the brief behind it says to preserve meaning and key facts — but tone and meaning are not cleanly separable in English. Hedging changes obligation: I need this by Friday and it would be great to have this by Friday commit the reader to different things. Warmth changes attribution: your team missed the deadline becomes the deadline was missed, which is politer and blames nobody. Both are fair rewrites of one sentence, and both say something different. Read the result for the ask, the date and who is responsible.
Where does my text go when it is rewritten, and is it kept anywhere?＋
This is not one of the tools that runs on your device. Your message and the chosen register are posted to an internal /api/ai route, which passes them to Anthropic's API and streams the reply back. The route holds the text only for the life of the request — no database write, no transcript — but it does read your IP address to enforce a limit of six runs a minute per address, and the model provider receives the text under its own terms. A stiff email to a supplier is fine; a client contract under a confidentiality clause is not.
Why is the rewrite different every time I run it?＋
Language models sample rather than look things up, so one message and one register produce different wording on each run; there is no stored answer to hand back. Nothing here pins it down either — the request carries your text and a single instruction, with no seed and no memory of the previous attempt. That has a practical consequence: each run clears the result panel and replaces what was there, so a version you liked is gone the moment you press the button again. Copy anything worth keeping into a scratch document before trying a second register.
How much text can be rewritten at once?＋
One submission is capped at 40,000 characters, and anything longer is refused with a message naming the limit rather than being quietly trimmed. The reply has a separate and much lower ceiling of 2,000 tokens — tokens being the sub-word pieces models count in — so a long input can be accepted and still come back unfinished, ending in a bracketed note that it hit the length limit. Feed it a message rather than a document. Long rewrites also drift: the register is applied most firmly at the start and loosens as the text runs on.
Is it safe to paste an email somebody else sent me?＋
There are two risks and they are different. The first is privacy: the text leaves your device, so someone else's confidential message is theirs to protect rather than yours to paste. The second is that received text can carry instructions aimed at a model — a line telling it to ignore its brief, sometimes hidden in a signature block or in white text. The system prompt here tells the model to treat everything you supply as material rather than as directions, which reduces that risk without removing it. If a result answers a question you never asked, that is what happened.
