Terms of Service — Redrafty

Effective date: 20 August 2026

These Terms are an agreement between you and Redrafty ("we", "us"), the operator of the Telegram bot @RedraftyBot (Redrafty) and its Mini App (together, the "Service").

By starting the bot, you agree to these Terms. If you don't agree, don't use the Service.

Quick summary

This box is a plain-language summary. It is not a substitute for the full text below, but nothing below contradicts it.

Detect AI costs credits, charged by the length of the text and nothing else. It is much cheaper than Rewrite. Like the other tools, Detect sends your text to our external AI provider for the check.
The other four tools cost credits too. All five tools — Detect, Rewrite, Summarize, Explain and Feedback — send your text to an external AI provider in the United States for processing, and so will any similar tool we add.
You know the price before you send anything, and the quoted number is a ceiling. The rate per word is printed on each tool's card; multiplied by your text it becomes the price of the job. We never deduct more than that. If the work turns out cheaper, you pay less.
We cannot promise that any text will pass any AI detector. Nobody can.
Our Detect result is a probability, not a verdict. AI detectors get things wrong, in both directions.
You decide what to do with the output, and you are responsible for following the rules of your school, university, employer or publisher.
Credits are not money. They don't expire, but they can't be transferred or turned back into cash.
If our system fails, we restore your credits automatically. We don't refund money.

1. What the Service does

The Service processes text you send to it. There are five tools.

🔍 Detect AI — costs credits, priced by length.
You send text, we estimate how likely it is that the text was written by an AI. The check is done by a language model of our external AI provider, on the same terms as the other tools.

Detect is charged by the length of the text — 0.3 credits per word, that is 3 credits for every 10 words, rounded up to the next whole credit and with no minimum charge. How a word is measured across writing systems is in §5.3. The rate is the same in every language. Detect has no settings at all. Texts shorter than about 100 words are not checked and not charged — there is not enough signal in them for an honest answer. Between roughly 100 and 250 words you get a band rather than a number, for the same reason. Re-checking the same text is a new, paid job, charged by length at the same rate as the first check. We also apply a rate limit (roughly 30 checks per minute per user) and a maximum text length, to keep the server responsive.

✍️ Rewrite — costs credits.
Rephrases your text in one of six modes: fix, paraphrase, restructure, shorten, creative, expand.

📋 Summarize — costs credits. Condenses a long text into bullet points, a retelling, or detailed notes.

💡 Explain — costs credits. Explains a difficult passage in plain language.

📊 Feedback — costs credits. Says how your text is likely to read to a chosen audience — a lecturer, a reviewer, a recruiter, an editor. It comments on the text; it does not rewrite it.

How the tools work. Your text is sent to an external AI provider for processing — a third-party company in the United States — for all five tools, Detect included; the editing tools send it paragraph by paragraph. This is unavoidable — it is how the tools work.

One quality level. All tools run in one mode; there is no model to choose and nothing that unlocks with a purchase.

Where you choose things. The tool and its settings are chosen in the Mini App, once. In the chat there is nothing to choose: you send text and the bot runs the tool you last made active.

You know the price before you send anything. The rate per word is printed on each tool's card, so the price of a job is known from its length. There is no separate screen asking you to confirm the price: see §5.3 for what the number means.

More detail on what is stored, for how long, and what leaves our server is in our Privacy Policy at redrafty.com/privacy.

2. Who can use the Service

You need a Telegram account, and you must follow Telegram's own Terms of Service.
You must be old enough to use Telegram where you live. If you are under 18, use the Service only with permission from a parent or guardian.
One person, one account. Creating extra accounts to farm free credits or referral bonuses is not allowed (see §6).
We are not affiliated with, endorsed by, or partnered with Telegram, any AI-detection company, university, or publisher. Any product names mentioned belong to their owners.

3. What we do NOT guarantee

This is the most important section in this document. Please read it.

3.1 We do not guarantee that your text will pass any AI detector

Not Turnitin, not GPTZero, not Originality.ai, not your professor's tool, not any tool that exists today or ships tomorrow. We do not sell "undetectable" text and we will never claim we do.

Detectors are updated constantly and each one behaves differently. Anyone promising "100% undetectable" or "guaranteed to bypass [any detector]" is making a claim they cannot support.

3.2 AI detectors are unreliable — including ours

Commercial detectors advertise 98–99% accuracy. Independent testing puts real-world accuracy closer to 65–90%, depending on the text, the length, and the language.

Two failure modes matter to you:

False positives. Genuine human writing gets flagged as AI. This happens more often to people writing in English as a second language, and to technical or academic writing, because both tend to use even, predictable sentence structures. If English is not your first language, be aware that any detector — ours included — is more likely to flag your real work.
False negatives. AI-written text gets marked as human.

3.3 Our Detect result is a probability, not a verdict

What we show is a likelihood — never a yes/no answer, and never a claim about who wrote the text. On short texts we do not even show a number: between roughly 100 and 250 words you get a band, because a precise-looking figure drawn from that little text would be false precision. Under about 100 words we refuse to answer at all and charge nothing. That is a signal to look at, not proof of anything.

Do not use our Detect result to accuse anyone of anything. Do not treat a "low likelihood" result as clearance. It isn't.

3.4 Our tools do not guarantee any academic or professional outcome

Running text through Rewrite or any other tool does not guarantee:

that a detector will score it as human,
that a teacher, reviewer, editor or employer will accept it,
that a grade, a submission, a publication or a job application will succeed,
that the meaning, facts, citations or formatting of your text survive unchanged.

Always read the output before you use it. Automated rewriting can change nuance, break quotations, alter numbers, or introduce errors. The final text is yours, and checking it is your job.

3.5 This is not advice

The Service is a text tool. It is not legal advice, academic advice, career advice, or professional advice of any kind. It does not create any adviser–client relationship.

3.6 The Service is provided "as is"

We work to keep it running, but we do not promise uptime, uninterrupted access, specific speed, or that any particular model, mode or feature will remain available. We depend on Telegram and on third-party model providers, and they change things without asking us.

To the extent the law allows, the Service is provided without warranties of any kind. Nothing here removes rights you have under mandatory consumer law in your country.

4. Your responsibility

You keep all rights to the text you send. You are responsible for it.

By sending text, you confirm that you have the right to send it and to have it processed — that it's your work, or that you have permission.

Rules that apply to you are your responsibility. Schools, universities, employers, journals and clients set their own rules about AI tools. Those rules vary enormously: some allow AI assistance with disclosure, some allow it for drafting but not for submission, some ban it outright. Some require you to declare it. Check the rules that apply to you before you use the output. We can't check them for you and we don't know what they are.

We do not encourage anyone to break the rules of their institution or employer, and we don't market the Service as a way to do so. What you do with the output is your decision and your responsibility, not ours.

You are also responsible for reviewing the output for accuracy before you use it anywhere that matters.

5. Credits and payment

5.1 How payment works

Payments are made only with Telegram Stars (XTR). This is a Telegram requirement for digital goods and services in bots — we cannot accept cards, PayPal, bank transfers, crypto, or any other method, and we will never ask you for card details. If anyone claiming to be us asks you to pay another way, it isn't us.

You buy Stars from Telegram, then spend them in the bot on a credit pack. Telegram (and, on iOS and Android, Apple and Google) handle the purchase of the Stars themselves under their own terms.

5.2 What credits are

Credits are a prepaid unit of processing inside this bot. They are not money and have no cash value.

Credits do not expire. They stay in your balance.
Credits are not transferable. You cannot send them to another user or another account.
Credits cannot be converted back into Stars, money, or anything else.
Credits only work inside this Service.
New users receive a one-time free starting balance. Free credits, bonus credits, referral credits and loyalty credits are gifts. They follow the same rules as bought credits and can be withdrawn if they were obtained through abuse or a reversed payment.

5.3 What things cost

Cost depends on text length, the tool, the strength setting, and the language of the text.
The answer is billed together with the text you sent. That is why 📋 Summarize is cheap and 💡 Explain is not: one returns much less than you sent, the other several times more.

Why the app shows a range, not a single number. Underneath, we bill for the amount of processing your text actually needs, and different languages need very different amounts for the same number of words. Some languages need more than twice as much processing per word as English; a few need considerably more. That is a property of the text itself, measured automatically. It is not a surcharge, and it is not applied to you because of where you are or what language you speak — the same rule applies to everyone.

Because we cannot know what language your text is in until you send it, the tool cards show a range:

the lower bound is the English rate — the cheapest case, and always reachable;
the upper bound is the rate for the language your app is set to.

Both bounds are true, and neither can be exceeded. If your app is set to English, the range collapses to a single number. Where our measurement for a language is less certain, the upper bound is written as "up to".

The range is an estimate, not a price. It exists so you can compare tools and modes before you have any text.

The price is calculated from your actual text, and the quoted number is a ceiling. It is the rate for your text's language multiplied by its length. We will not deduct more than it, ever — that is not a check we run afterwards, it is how the charge is computed. If the work turns out to need less, you pay less, and the difference is not held back. There is no price confirmation screen anywhere in the Service, because there is nothing left to confirm about the price: the rate is on the card before you send, and the price follows from the length of what you sent. The bot does stop and ask in two cases that are not about price: when the text is too short for the tool to give an honest answer, and when the text exceeds the size limit for one request. No tool has a minimum charge — a short job is billed for what it is, not rounded up to a floor.
Detect is the exception: it is billed by size exactly, at a flat rate of 0.3 credits per calculated word, rounded up to the next whole credit, with no minimum charge. A "calculated word" is a length unit normalised for the writing system — the number of characters that make up one word differs between scripts, and the bot accounts for that, so the same amount of text costs the same in Korean, Arabic and English. The rate itself is identical in every language and every mode.
There is a maximum size per request — 10,000 words, the same for every user from the first message. The model tier you select does not change this limit. Longer texts can be split, or processed in part using the truncate button the bot offers on the size warning.
Running a text again is a new, paid job. Asking for another variant, changing the mode, style or strength, and re-checking the same text with Detect each start a fresh job, priced from the text and the settings you give it.
Pack sizes and credit prices may change for future purchases. Credits already in your balance are not affected.
Bonus credits added to a purchase of yours — your loyalty tier and the one-off welcome bonus for arriving through an invite link — are capped at a maximum of 25% on top of that purchase. Referral earnings, milestone rewards and starting credits are not attached to a purchase of yours and are not subject to that cap.

5.4 Referrals and ambassador rewards

Referral, milestone and loyalty rewards are never paid in money, unless a separate written arrangement says otherwise. Most are credits; five of the six milestone rewards are not credits at all but a discount on a model tier for a fixed number of credits. That discount has no cash value either. No reward of any kind changes the size limit for one request — see §5.3.
The referral bonus is paid on pack S and larger. The XS pack does not pay one.
The referral share is paid on every qualifying purchase a friend makes, for as long as they keep buying — not only on their first.
The rate is tiered and rises with the total ⭐ your referred friends have bought: 10% below 3,000 ⭐, 15% from 3,000 ⭐, 20% from 9,000 ⭐. The rate applied to a payout is the one in effect when that payment settles; later tier changes do not retroactively adjust past payouts.
If a friend's payment is refunded or reversed, it stops counting toward your total, which can lower your rate for future payouts. Credits already paid out for a reversed payment may be withdrawn (see below).
Stacked bonuses on a single purchase are capped at 25% (see §5.3), so a welcome bonus advertised as "+500 credits" pays less on a small pack where 500 would exceed that cap.
Both kinds of invite are ordinary Telegram links — a friend's referral link and an ambassador's link. Opening one attributes your account on your first start, automatically, and that attribution is permanent. Each account carries exactly one of them: the first link you arrive through is the one that counts, and later links leave it as it is.
We can withhold or reverse a reward if the purchase behind it is reversed, refunded, or made through an account that breaks §6.
Ambassador arrangements — revenue share, cash payouts, a personal invite link — exist only where we have agreed them with you in writing, individually. Nothing on this page or in the bot creates one, and the terms of any such arrangement are the ones in that written agreement.

5.5 Refunds

Spent credits are not refunded. Once a job runs and you receive a result, the credits are gone — including when you don't like the result. "The rewrite wasn't good enough", "the detector still flagged it", "I changed my mind", or "I picked the wrong mode" are not technical failures.

Disagreeing with a Detect score. A check is charged, and its result is a probability estimate, not a verdict — see §3. A score you disagree with is not a technical failure: the tool delivered exactly what it promises, an estimate. It does not entitle you to a refund, and there is no appeal procedure for it. Technical failures are different and are refunded automatically, without you asking.

We do not refund money. Not for unused credits, not for a changed mind, not for packs bought by mistake.

One exception, and it is not money: an undelivered purchase. If you paid for a pack and the credits never arrived, write to us and we return the Stars — to your Telegram Stars balance, not as money. Buying the Stars themselves happens in Telegram and, on a phone, in Apple's or Google's store; that transaction is not ours and we cannot reverse it.

Goodwill credits. We may occasionally add credits at our discretion — for example when a result was clearly poor and no automatic restore fired. That is a gift, not a right, not a refund, and not a precedent.

5.6 What happens when something breaks — automatic credit restore

If our system takes your credits and fails to deliver a result — the bot crashes, the model API errors out, the request times out, the reply comes back empty — your credits are restored automatically. You don't have to write to support, and there is no form to fill in. The bot tells you:

⚠️ Couldn't process the text. Credits returned: 2,494 💎

This is a credit restore, not a money refund. No money moves. It's automatic because it's the only genuinely fair reason to ask for anything back, and we'd rather it just work.

If credits were taken and you got nothing, and the balance was not restored within a few minutes, contact us (§10) and we'll fix it.

5.7 Reversals by Telegram

Payments travel from you through Apple/Google and Telegram to us. Telegram operates the payment and can reverse a Stars transaction under its own rules. If a payment to us is reversed, we remove the credits from that purchase and any bonuses granted because of it. If your balance has already been spent, it can go to zero, and we may suspend the account.

5.8 The 21-day hold

Telegram holds funds from Stars purchases for 21 days before we can withdraw them. This is between us and Telegram and does not affect you. Your credits appear in your balance immediately after purchase and are usable straight away.

5.9 Language

The language of the app and the language of your text are independent. You can set the app to any of the languages we offer and still send text in any language. We do not require, check, or restrict a match between the two.
The app's language setting affects buttons, hints and messages from the bot. It does not restrict what you may send, and it does not change what your text costs — cost follows the language of the text, not of the interface.
All five tools reply in the language of your text. The bot detects that language itself. The answer language is not a setting: what you write in is what you get back.
The interface exists in a smaller set of languages than the set we can process and price text in. Those are two separate lists: a language missing from the interface is still processed and billed correctly.
Our AI detection was tested mostly on English and, on smaller samples, on Spanish, German, French, Portuguese, Hindi and Hinglish. For other languages and scripts it may be less accurate. We say so here and in the in-app Help rather than hiding it; we do not attach a disclaimer to each individual result. See §3.2.
Support replies in English. You may write to us in any language.

6. Acceptable use

Don't use the Service to:

Break the law, or produce content that is illegal where you or your readers are.
Create harmful content: harassment, threats, hate speech, content sexualising minors, content promoting violence or self-harm, or instructions for causing harm.
Deceive people at scale: phishing, scams, fake reviews, fake identities, impersonating a real person or organisation, disinformation campaigns, or academic material sold to others as original work.
Spam: generating bulk unsolicited messages, comments, emails or DMs, or content for spam operations.
Process other people's personal data without a lawful basis and their knowledge — don't paste someone's private information into the bot.
Automate abuse: scripting or botting the Service, scraping it, hammering the Detect endpoint, or running the Service through automated pipelines without our written permission.
Work around limits: creating multiple accounts to collect free or bonus credits repeatedly, faking referrals (referring yourself through second accounts), manipulating loyalty or ambassador rewards, or otherwise gaming the credit system.
Resell or share access: selling, renting or sharing your account, reselling credits or output as your own service, or rebuilding the Service as a competing product using our outputs.
Attack the Service: reverse engineering, probing for vulnerabilities, injecting malicious files, or trying to extract our prompts, models or infrastructure details.
Upload malware or files designed to break our parsers.

Occasional heavy legitimate use is fine. This section is about deliberate abuse.

7. Suspension and ending access

We can limit, suspend or close access to the Service when:

you break §6,
your usage threatens the stability or security of the Service,
a payment is reversed or found to be fraudulent,
we are required to by Telegram, a model provider, or the law.

How we handle it. For minor or accidental issues — hitting a rate limit, one weird request — we'll usually just throttle or warn you. For deliberate abuse, fraud, or anything in the "harmful content" list, we suspend immediately.

What happens to your credits. There are two different outcomes, and the difference matters:

While access is limited or suspended, your credits stay on your balance. A throttle, a warning or a temporary suspension does not touch them. When access is restored, they are still there.
When an account is closed for abuse or fraud, unused credits are lost. Because credits have no cash value and are not refundable, there is nothing to pay back. That's the actual consequence, and we'd rather say it plainly than bury it.

If we shut the Service down entirely, we'll announce it in the bot with as much notice as we reasonably can. Unused credits cannot be converted to money in that case either — see §5.2 — but we'll give you time to use them.

You can leave any time. Press 🗑 Delete my data in the Mini App to erase what we hold, stop using the bot, and block it. Unused credits are not refunded.

8. Your text, your data

The full picture is in the Privacy Policy at redrafty.com/privacy. The short version, because it belongs in the deal you're agreeing to:

All five tools send your text to an external AI provider (United States; the same one reads photos and scans). It states that conversation content is not retained by default on the models we use, and that retained data is not used for model training without express permission. None of the models we use are covered by the exceptions its policy makes to that default. We rely on the provider's published policy; we don't control its systems.
We store the text you send for up to 24 hours so the next-step buttons under your result keep working — the 🔍 check offered after a Rewrite. When those 24 hours are up the buttons stop working, and that is why. Every check is charged, including a repeat check of the same text.
We cache processed paragraphs under a hash for up to 7 days (fast cache) and 30 days (cold cache), then they are deleted automatically — unless the identical paragraph is sent again, which restarts the clock. The cache is keyed by the content of a paragraph, not by who sent it: there is no link from a cache entry back to your account.
We do not store your original text in our main database. Job records keep a hash of the input, token counts and cost — not the text.
We never train or tune any model on your texts or your results — not our detector, not anyone's — and not even with your permission: there is no switch for it. Your text is used only to complete the request you asked for.
Seeing, copying and deleting what we hold are buttons. On the Mini App's Privacy screen: 📄 What we store about you, 📦 Send a copy of my data to the chat, and 🗑 Delete my data, with a confirmation screen and a "this cannot be undone" tick. Deleting erases your support requests, your Telegram username, time zone and activity hours, the hashes of your texts and the link between your account and our detection records — immediately, with no time limit on our side. Your credit balance and purchases stay, because that is your money, and payment and credit records are kept with them: we need them for accounting, for the Telegram hold, and to resolve disputes. They contain amounts, dates and Telegram payment IDs, never your text. Processing records without text stay until they expire. Shared cache entries stay as well: they are keyed by the content of a paragraph rather than by account, nothing in them points back to you, and they expire on their own within 30 days. Encrypted backups of our database are kept for up to 8 days, so deleted data leaves them within that time. To have the account closed as well, write to us (§10); the bare account record with your purchases stays, because payment records must belong to an account.
A note on the chat. The bot does not delete messages from your Telegram conversation. To clear it, use Telegram's own "Clear history" — that clears the chat on your side. Deleting what we hold is the button above, and the two are not substitutes for each other.

9. Changes

To the Service. We may add, change or remove features, models, modes and pack contents. Credits already in your balance keep working.

To these Terms. We may update these Terms. If a change materially affects you, we'll notify you in the bot before it takes effect. The "Last updated" date at the top always shows the current version. Continuing to use the Service after a change means you accept the new version. If you don't accept it, stop using the Service — see §7 on what happens to unused credits.

10. Liability, law and contact

Liability. To the extent the law allows, we are not liable for indirect or consequential losses — including lost grades, rejected submissions, disciplinary outcomes, academic penalties, lost income, lost work, or reputational harm — arising from your use of the Service or from any output it produces. Where liability cannot be excluded, it is limited to the value of the Stars you spent in this bot in the 6 months before the claim.

Nothing in these Terms limits liability for fraud, or for anything that cannot be limited under the law that applies to you. If you are a consumer, your mandatory local consumer rights are unaffected.

Your consumer rights. Nothing in these Terms takes away the mandatory consumer protections of your country of residence, and you may bring proceedings there.

If something goes wrong, please talk to us first — most issues are fixed in a message.

Contact:

In the bot: send /support — the same form as in the Mini App, and it works when the Mini App does not
Contact form: redrafty.com/contact — if you can't use the bot

We aim to reply within 24 hours, and within 48 hours over a weekend.

By using @RedraftyBot you confirm that you have read these Terms, and in particular §3 — what we do not guarantee.