PrismHelper — AI moderator for Telegram groups
The AI moderator for Telegram groups. It reads meaning and context, not just a list of banned words.
7 days of Pro, free · No card. No auto-renewal.
Order starts with Prism.
Strict where it matters. Fair everywhere else.
Reads meaning, not keywords.
Three layers check every message: dictionaries, a local neural network and AI that sees the conversation around it. It tells an insult from a discussion, a quote from an opinion, a joke from harassment.
Example“Don’t be an idiot, read the pinned post.”
Insult“Just finished The Idiot by Dostoevsky.”
Book title
Proportionate, never trigger-happy.
A step-by-step ladder replaces instant bans: a warning first, then mutes from 10 minutes to 8 hours. Bans are kept for serious violations, and strikes expire with time.
Warning10 min30 min1 h3 h8 hDefault penalty ladder Every decision explained.
The bot names the broken rule and the reason. Members can appeal with one tap, and admins see a full log of every action.
ExampleAppealPrismHelperbot
Message removed.
Rule: insulting a member.
Penalty: warning. Next time: 10-minute mute.
Three layers. One calm chat.
Every message passes three layers of checks, from instant to thoughtful.
- Instant
Dictionaries
Text normalization and fuzzy search catch typos, merged words and disguised characters: Latin look-alikes, digits, spaces and punctuation.
- 1d10t
- idiot
- f r e e m o n e y
- free money
- A fraction of a second
Toxicity neural network
Runs on the bot’s own server. Understands slang and meaning that no word list covers.
- “nice catch, thanks”
- toxicity 0.03
- “nobody wants you here”
- toxicity 0.86
- Borderline cases
AI with context
Reviews borderline cases with the previous messages in view and reads the chat in batches of up to 12 messages, catching hidden aggression, sarcasm, link-free scams and agitation. Works in any language. Doubtful cases get a second look from a more accurate model.
12 messages, one request
Messages and scores are illustrative.
AI advises. Your rules decide.
Penalties come from your rules and the penalty ladder, never from a neural network. A single dictionary match never leads to a ban: the member gets a 24-hour mute and the admin gets a card to confirm.
Anonymized before the AI sees it.
Names become tags. Links, phone numbers, card numbers and emails are hidden. Messages without violations are never stored, and the data lives in the EU (Netherlands).
In the chat
Mark: DM me at +1 555 0142 or mark.k@mail.com, details at bit.ly/x9w
What the AI receives
[User 1]: DM me at [phone] or [email], details at [link]
What it catches
Eight families of violations, 53 rules. Each one can be switched on or off, capped at a maximum penalty and described in your own words for the AI.
Profanity, insults, threats, harassment
Including disguised spellings.
Spam and advertising
Channels, chats, bots, services, links, channel forwards, referral schemes and “easy money”.
Scams and phishing
Fake Telegram services and people posing as admins.
Malicious files
Harmful attachments.
Prohibited content
18+, shock content, drugs, weapons, extremism, calls to self-harm.
Privacy violations
Doxxing, de-anonymization, leaked private messages, harvesting members’ data.
Flood
Caps lock, sticker spam, mass mentions.
Joining and membership abuse
Spam bots, fresh spam accounts, admin look-alike names, multi-accounts, ban evasion, raids.
Jokes, idioms, memes and words used literally are not violations.
All 53 rules
Communication12
- Profanity without a target
- Insulting a member
- Remarks about family, nationality, language or religion
- Harassment
- Threats, intimidation, blackmail
- Wishing death, pushing towards self-harm
- Provocation and incitement
- Insulting the admins or the bot
- Discrimination
- Political and religious agitation
- Passive aggression
- Banned words (your own list)
Flood and order9
- Flood
- Excessive caps
- Sticker spam
- Voice and video messages
- Off-topic
- Begging
- Night mode
- Abusing bot commands
- The chat’s own rules
Ads and spam8
- Links
- Ads for channels, chats, bots and services
- Forwards from channels
- Referral schemes and dubious earnings
- Selling accounts, fake engagement, ban bypass
- Ads in a name or profile
- Mass mentions
- Banned websites
Personal data and safety8
- Disclosing personal data
- De-anonymization
- Publishing private messages
- Scams
- Phishing
- Impersonating admins
- Malicious files
- Harvesting members’ data
Content6
- 18+ material
- Shock content
- Drugs, weapons, extremism, calls to suicide
- Pirated content
- Dangerous misinformation
- Unmarked spoilers
Profile and participation7
- Captcha for newcomers
- Offensive or promotional name
- Name imitating an admin
- New account with spam-bot signs
- Raid
- Penalty evasion
- Multi-accounts and fake engagement
Reports and appeals3
- False reports
- Appeal abuse
- Dodging a mute by editing messages
Choose how strict your chat is.
Casual talk is fine. Only gross personal humiliation, profanity aimed at a person and insults about relatives count as violations.
Light arguments and teasing are fine. Profanity aimed at a person is not.
No profanity and no insults of any kind. Links and channel forwards are limited by default only in this mode.
This update is damn good.
AllowedAllowedRemovedlol, you’re such a nerd
AllowedAllowedRemovedShut up, you f***ing idiot.
RemovedRemovedRemovedDocs are here: github.com/prism/docs
AllowedAllowedLimited
Fair by default.
Penalties
- A warning first, then mutes of 10 min, 30 min, 1 h, 3 h and 8 h.
- Bans only for serious violations: threats, death wishes, scams, drugs, 18+, malicious files, selling accounts.
- Minor violations never escalate to a ban.
- Strikes expire: by default one is removed for every 7 days without violations.
- Addressing someone directly, repeating after a warning or an especially humiliating form doubles the penalty. A clean record softens it.
- Admins can change any step or give a rule a fixed action: ladder, delete only, warning, 1-hour mute, 24-hour mute or ban.
Explanations in your chat’s voice
PrismHelperbot
Message removed.
Rule: insulting a member.
Penalty: warning. Next violation: 10-minute mute.
Let’s keep it kind.
Your message broke the “insulting a member” rule, so it was removed.
This one is a warning. Next time it’s a 10-minute mute.
That one didn’t pass the vibe check.
Rule: insulting a member.
You get a warning. The next one comes with a 10-minute timeout.
Appeals
- Warnings and mutes are shown only to the offender by default, and the notice disappears after a minute.
- One tap on Appeal. On Pro and above a more accurate AI model reviews it first; on Start it goes to the admins.
- A first slip can get a soft personal reminder instead of a penalty.
- If a re-check finds a mistake, the penalty is lifted, the strike is returned and the explanation is corrected.
- /report report a message
- /me my strikes
- /rules chat rules
Built for the people who run the chat.
- Owner panel
- Every chat you run, with its plan, members, weekly violations and AI checks left.
- Settings in private chat
- Rules, strictness, the penalty ladder, strike expiry, captcha, anti-raid, night mode, welcome message and lists.
- Rules in plain words
- Describe a rule and the AI enforces it: 3 on Start, 5 on Pro, 7 on Team, 10 on Business and Business Ultra.
- Lists
- Your own blocklist of words and sites. Allowlists for words, sites and members.
- Night mode
- During the hours you set, messages are deleted without strikes.
- Log channel
- Every decision the bot makes is posted to a separate channel.
- Moderators
- Helpers without Telegram admin rights get bot commands, immunity and permissions you grant one by one.
- Moderation cards
- Appeals, reports and raids go to the owner, the moderators or both.
- Newcomer protection
- Captcha, limits for the first 24 hours, anti-raid and an account-age check.
- Conflict summary
- A short recap of what happened, written for the admin.
Also: copy settings between chats, a setup wizard, chat statistics, and alerts at 80% and 100% of AI checks and 3 days before the plan ends.
Bot commands
- $ /ruleschat rules
- $ /memy strikes
- $ /reportreport a message
- $ /legaldocuments
- $ /mydatawhat is stored about me
- $ /deletedatadelete my data
- $ /paysupportpayment questions
- $ /checktest a text without penalties
- $ /warnwarning
- $ /mute /unmutemute, unmute
- $ /ban /unbanban, unban
- $ /kickremove from the chat
- $ /forgiveremove strikes
- $ /strikesstrikes
- $ /mod /unmod /modsmoderators
- $ /settingssettings
Three steps to a calmer chat.
Add the bot
Add @prism_moderator_bot to your group.
Give it two rights
Delete messages and ban users. Nothing else is needed.
Open the panel
In a private chat with the bot, tap Panel and pick your chat. A base set of rules is already on.
The “Add to chat” link pre-selects exactly those two rights.
Simple plans. No auto-renewal.
Each plan covers 30 days and is paid inside the bot. Every new chat gets 7 days of Pro, once, for free.
Start
₽299/ 30 days
For a small chat. Full protection; appeals go to the admins.
- chats
- 1
- AI checks
- 15,000
- own rules
- 3
Pro
Free for 7 days₽699/ 30 days
Start, plus AI-reviewed appeals, soft reminders, your own ladder and advanced settings.
- chats
- 1
- AI checks
- 35,000
- own rules
- 5
Team
₽1,199/ 30 days
Pro for up to three chats with a shared pool of checks.
- chats
- up to 3
- AI checks
- 65,000 shared
- own rules
- 7
Business
₽3,999/ 30 days
Pro, the most accurate AI model and a shared blocklist for up to five chats.
- chats
- up to 5
- AI checks
- 125,000 shared
- own rules
- 10
Business Ultra
₽7,999/ 30 days
Business terms for up to ten chats.
- chats
- up to 10
- AI checks
- 250,000 shared
- own rules
- 10
Compare all features
| Start | Pro | Team | Business | Business Ultra | |
|---|---|---|---|---|---|
| Chats | 1 | 1 | up to 3 | up to 5 | up to 10 |
| AI checks per month | 15,000 | 35,000 | 65,000 shared | 125,000 shared | 250,000 shared |
| Dictionaries, neural network, blocklist | |||||
| AI reads every message, any language | |||||
| AI re-checks dictionary decisions | |||||
| Captcha, anti-raid, strikes, penalty ladder | |||||
| Own rules | 3 | 5 | 7 | 10 | 10 |
| Appeals reviewed by | Admins | AI | AI | AI | AI |
| Soft reminder, conflict summary | |||||
| Own ladder, night mode, welcome, allowlists, log channel, explanation style | |||||
| Most accurate AI model for every decision | |||||
| Shared blocklist across chats |
- What is an AI check?
- One request to the AI: a batch of up to 12 messages, a borderline case, a dictionary re-check, a report or an appeal. In a quiet chat a batch is usually one message, in an active one up to twelve, so a plan covers at least as many messages as it has checks.
- Out of checks?
- The bot stays and keeps working on dictionaries, the neural network, captcha and anti-raid. AI resumes with the next period. Unused checks don’t carry over.
- Raid-proof budget
- Each chat has a daily limit of checks, so an attack can’t burn the monthly pool in one day. On Team and both Business plans the pool is split equally between your chats.
- Switching and leaving
- Switching plans converts the time left into days on the new one. Without a plan, captcha, anti-raid and admin commands keep working.
Questions
Still have a question?Ask support · @prismhelpermanager
How is PrismHelper different from regular anti-spam bots?
It looks at meaning and context, not only at matches against a word list. That means fewer unfair penalties and violations caught that dictionaries miss. Every decision comes with an explanation and can be appealed.
Does the AI punish people on its own?
No. The AI gives an assessment; the penalty is chosen by the system from your rules and the penalty ladder. A ban based on a single dictionary match is impossible.
What if the bot gets it wrong?
The member taps Appeal. Admins can lift any penalty, and /forgive returns strikes.
Does the AI receive personal data?
No. Before analysis, members’ names are replaced with neutral tags, and links, phone numbers, card numbers and emails are hidden. Messages without violations are not stored.
What does the bot store, and where?
Chat and member IDs, violations and strikes, settings, payment records and the text of violations, which is kept for 90 days and then reduced to a record without text. Everything is hosted in the EU (Netherlands). Members can see their data with /mydata and delete it with /deletedata.
What permissions does the bot need?
Only two: delete messages and ban users.
Does it work in languages other than Russian?
Yes. The AI works with any language on every paid plan. The dictionaries are tuned for Russian.
What happens when the trial ends?
If no plan is paid, the bot stays in the chat: captcha, anti-raid and admin commands keep working. Moderation and the AI switch off.
Is there auto-renewal? How do I pay?
No auto-renewal. A plan is paid for 30 days inside the bot: Panel → your chat → Plan. You get a reminder three days before it ends.
Can it just delete messages, without penalties?
Yes. Any rule can be set to “delete only”.
Does it work in channels or private chats?
No, only in groups and supergroups.
Can I get a refund?
You can cancel a plan at any time. Unused days are refunded within 10 days, minus the 3% payment fee and AI costs beyond the used part of the period. If you paid and the plan wasn’t activated within 24 hours, you can ask for activation or a full refund.
Your chat, without the noise.
Add PrismHelper, give it two rights, and the first 7 days of Pro are on us.