Script builder

AI assistant for Telegram: runs the conversation instead of a manager

Build the script from blocks and the AI takes the conversation from there: it qualifies the request, works through objections and drives to the target action. It writes from a regular account, in plain living language.

Try it for $9.99 See the conversation

7-day trial for $9.99, then from $90/mo. We never charge automatically and never store your card.

What it looks like from the other side

An ordinary direct-message conversation. On the right — what the script is doing at that moment.

Dmitry

Dmitry, good afternoon. I saw your comment in the “Logistics and Freight” chat — about orders from messengers getting lost between dispatchers. That is exactly the piece we cover. Is it still a problem?

Alright, suppose so. What exactly do you do?

marked: qualified

We pull orders from every messenger into one window and route them to dispatchers automatically. How many trucks do you run?

40 of our own plus hired ones

Then this is exactly your size. I can send a short video of how it works at a similar company, or we can just do a 15-minute call — whichever suits you.

Let’s do the call, but next week

conversion: agreed to a call

The opener hooks into context

The script sees which chat the person came from and what they wrote there. Not “hello, we offer services”, but a reply to their own words.
The reply is read, not just counted

“Alright, suppose so” is interest, not agreement. The assistant knows the difference and takes the clarifying branch instead of rolling out the offer.
Marks are set as it goes

Qualified, interested, declined — dialogue states the script later branches on and the stats are counted from.
“Not now” is a branch, not a dead end

A postponement leads into its own part of the script: remind later, rather than repeat the same offer in different words.
The target action lands in the report

You define the conversions yourself. They add up to the campaign funnel, so you can see at which step the conversations fall apart.

The script is assembled from blocks

A dialogue is a diagram on a canvas: nodes and the links between them. Ten block types that add up to any conversation logic you need.

StartThe opener — where the conversation begins

Reply receivedThe person wrote back — the script continues from here

SilenceThe person did not reply — the script carries the conversation on itself

MessageExact text, or a task the AI writes from

ConditionA fork based on profile data, variables and the course of the dialogue

VariableRemember something from the conversation and use it later

ConversionMark the target action so it lands in the report

A/B testSplit people across variants and compare which one worked

NotificationA ping to the Telegram bot when something important happens in the dialogue

FinishThe conversation is over — the assistant writes no more

Written for one specific person

A template with a name dropped into it is still a template, and it shows. The script has thirty fields about the person, and the AI leans on them when wording a message.

What the profile lacks is filled in by an AI variable: you describe the trait in words, the assistant infers it from the person’s messages, and the script branches on it. Repeating chunks of text are spun into variants inside the message itself, so a hundred conversations do not open with the same line.

Name and username Profile bio Hours since last seen What they wrote in the source chat Personal channel title Channel description Pinned post Recent channel posts Premium and verification Conversation history Values of your own variables

What the assistant will not do

The rules of conduct are baked into the platform and outrank your task: if the task pushes towards breaking a rule, the assistant will not break it. This protects your brand — the conversation stays within bounds even when the task is worded too pushily.

Will not invent factsNumbers, links, names, dates and promises — only the ones in your script. Asked something it has no answer to, it says honestly that it does not know.

Will not argue backIf the person declines, asks not to be written to, or is annoyed, the assistant agrees and ends the conversation. There is no second lap.

Will not run aheadUntil the person has answered the question asked, the assistant does not move on to its own agenda — however tempting closing the deal may be.

Will not give itself away with officialeseThis is a messenger conversation, not a business letter: no “thank you for reaching out”, no repeated greetings, no memo structure.

When needed, a human takes the conversation over

Full automation is great right up until the conversation becomes expensive. After that a manager steps in.

Manual mode, one switch

Turn it on in the dialogue and the assistant goes quiet for good. A real person continues from the same dashboard, on the same account.
The notification lands in the bot

The notification block pings you in Telegram at exactly the step where you want to step in — not after the lead has gone cold.
You can see where a conversation stalled

Dialogues in the dashboard are tagged with marks and conversions, so the ones stuck at a given step are found with a filter in seconds.
The context does not vanish

The manager sees the whole thread and picks up from the same line.

This is not a chatbot

The difference is not in politeness of phrasing but in what is possible at all. A Telegram bot cannot message first: until the person presses “Start” themselves, they are out of the bot’s reach.

A regular Telegram account writes. The message arrives with a notification, in the main chat list — like one from a person, not into the bots folder.

The conversation runs on words, not buttons. People answer the way they normally do, in their own wording.

You make the first move. No waiting for the audience to find you and start a bot on their own.

That is exactly why the platform runs on accounts rather than on the Bot API. A line-by-line comparison is in the table on the home page.

Your first script in 15 minutes

You do not have to map the whole conversation tree up front. The working minimum is an opener, a couple of branches and a target action.

Describe the task

Who you are writing to, what you offer and what counts as success. That text becomes the AI’s task inside the messages.

Build the branches

Interest, doubt, “not now”, refusal. One branch per reaction — the rest you will add from real conversations.

Run it in the sandbox

Play the other side yourself and see where the assistant stumbles. Not a single live contact is spent on it.

Attach it to a campaign

Pick the audience and the accounts — the script starts working on everyone you collected.

Build your first script See pricing

Common questions about the AI assistant

Both — it is a choice made per message. The opener is usually fixed as text to keep control over how the conversation starts, while replies are handed to the AI: it sees exactly what the person wrote and words its answer for them. Both modes coexist happily in one script.

No. In the block you describe the task in plain words: what to find out and where to steer. The rules of conduct — how to run the dialogue, what not to do — are already baked into the platform and are not your job.

It will not pretend to understand and will not invent an answer — its rules say so in plain words. It admits it does not know. For cases like that you put a notification block into the script: the ping arrives in the Telegram bot and a person picks the conversation up.

Yes, in the sandbox: you play the other side and watch how the assistant answers and which branches it takes. No live contacts are spent — only the model’s own work, charged exactly as in a real dialogue.

With the A/B test block: it splits people across variants, and the conversions show which one reaches the goal more often. You can test the opener this way as well as any branch inside the conversation.

Yes, no need to stop the campaign — new dialogues follow the updated diagram. The platform checks the script for integrity itself: if an edit breaks it, the campaign halts and tells you why. The script keeps a version history, so a bad edit can be rolled back.