Connect a bot through Telegram Business mode and it gets a view of the chat that a normal bot never has. Incoming messages from the client, yes, but also the messages sent from the business account itself. That second stream is the useful part and the dangerous part at once.
Useful, because a bot that cannot see what the human said will happily repeat an answer the operator gave thirty seconds earlier, or contradict a discount that was just agreed. Dangerous, because a naive handler treats every message it receives as something to respond to, and one of those messages is now its own operator. The result is a bot answering the human, the human answering the bot, and a client watching a company argue with itself in their DMs.
The echo
There is a sharper version of the problem. When the bot itself sends a message on behalf of the business, that message can come back in as an update from the business account. The bot's own words return to it looking like input. If nothing stops the loop there, the bot answers itself, and the second answer comes back too. A conversation like that escalates in seconds and there is no natural end to it.
So the first job of the handler is not to reply. It is to work out who sent this and whether it deserves a reply at all: the client, the human operator, or the bot's own output coming back around. Only the first of those is input.
The pause is the rule, not the fallback
When Scalegram sees a message sent by the business account that it did not send itself, it stops replying in that conversation. Not for one message. For the whole conversation, until the pause expires or you release it.
That is a deliberately blunt rule and it is the correct one. A human typing into a chat is the single clearest signal in the whole system. It means someone has decided this conversation is worth their own time, and there is no interpretation of that which ends with "the bot should keep going". Any cleverness here, any attempt to work out whether the human's message was a full answer or just a greeting, is cleverness spent on being occasionally wrong in public.
The same instinct runs through the rest of the product. Anything irreversible waits for a person at an ask-me checkpoint, and the operator always outranks the assistant.
A paused conversation is still a live conversation. The pause stops the bot from writing, it does not close the thread or move the contact backwards in the pipeline.
What the pause does not touch
Pausing is about writing, not about record keeping. While the bot is quiet the contact stays where it is, tags stay on it, and the recency of contact updates as normal, because that is how the follow-up rules know the thread is alive. What is not happening, then or ever, is transcript storage. Scalegram keeps no message content at all: the contact record and a link back to the real Telegram chat, nothing else. So the pause is not hiding anything from the system, because the system was never reading the conversation to begin with. That design is covered in the Telegram CRM piece.
Resuming without a surprise
The awkward case is the operator who replies once and disappears. The client answers an hour later, and now nobody is talking to them. Scalegram resolves this with a timed pause rather than a permanent one: after a quiet interval with no further human message, the bot may pick the conversation back up. You can also release it explicitly.
Two things make that safe. The bot reads the current state of the contact before it writes again, so it resumes from the stage the human left it in rather than from where the flow was interrupted. And nothing that grants access, an invite or an upgrade included, resumes on its own. Those still stop and ask.
Why this is not solved with a prefix
The tempting shortcut is to tag every bot message with a hidden marker and ignore anything carrying it. Markers are part of the answer, and Scalegram does track what it sent so it can recognise its own output. But a marker only identifies the bot's messages. It says nothing about the operator's, and the operator is the one whose message must trigger the pause. You need both: recognise your own output so you do not loop, and recognise a human's output so you get out of the way.
The related failure, the same message arriving twice from a retried delivery, is a different guard entirely and is covered in how duplicate replies are prevented.
The operator's side of it
In practice the rule is easy to live with because it matches what people already expect. You open a chat, you type, the bot goes quiet. On a team, per area permissions decide who can write into a conversation at all, at read, add, write and delete level, so the pause is triggered by people who are supposed to be there. And when a bot has been paused often in the same kind of conversation, that is a signal about the flow rather than about the bot. Something in that branch is not good enough yet, and the humans have been quietly patching it.
"The first time we watched a bot reply to something the operator had just typed, it was funny for about four seconds. Then you realise the client is watching two versions of you disagree."
— Alex Onta, Executive Director, SINGUARD
Key Takeaways
- Telegram Business mode shows a bot the messages sent by the business account too, which is how a bot can end up answering its own operator or its own output.
- Scalegram recognises its own sent messages so the echo does not loop, and treats a human message in the chat as a hard stop on replying.
- The pause covers the whole conversation rather than a single message, and it is timed so a thread does not go silent if the operator steps away.
- Pausing stops writing, not record keeping: stages, tags and recency stay current, and message content is never stored either way.
Frequently Asked Questions
What exactly makes the bot stop replying?
A message sent from the business account that the bot did not send itself. That is read as the operator taking the conversation, and the bot stops writing in that chat rather than trying to judge whether the human finished.
Does the bot ever start again on its own?
Yes, after a quiet interval with no further human message, and it reads the current contact state before it writes so it resumes from where the human left things. Anything that grants access still waits for approval.
Can the bot end up talking to itself?
Not in Scalegram. It records what it sent and recognises its own output when it comes back as an update, which is the guard that breaks the echo loop before a first reply is ever generated.
About the Author
Alex Onta is an Executive Director at SINGUARD. He built eTrader, the terminal, the mobile apps, eTrader Broker, Copytrading, Business and Community, along with the worldwide clustered-server infrastructure it all runs on, with his brother Roman Onta helping on the design, and he leads that division today. Together with Roman he builds the Prop Firm CRM, the Broker CRM, Scalegram and CopySignals, and the two of them carry worldwide compliance, payment processing and international business structuring side by side. He lives and works in Dubai for most of the year. Meet the executive duo leading Singuard's five divisions.