The help assistant's question limits and why they exist, what happens when you hit one, what you must never write to the assistant, what does and does not reach the model, how answers are validated on the server, how long transcripts are kept, and the cases where the assistant is switched off.
Last updated: September 30, 2026
The help assistant is not unlimited, and it should not be. Every answer is an AI call; an uncapped chat makes the cost unpredictable and leaves the door open to abuse. By the same token, every sentence you write is stored somewhere and sent to a model — so what goes, what does not go, and how long it stays should all be stated plainly.
This guide covers the assistant's caps, what happens when you reach one, what you must never write to it, how answers are validated, and how long chat records live.
You are probably here for one of these:
The values below are the defaults that ship with the installation. Your administrator can change all of them, so your numbers may differ.
| Limit | Default | What it means |
|---|---|---|
| Length of one message | 2,000 characters | Paste the line that matters, not a whole log dump. |
| Messages per minute | 6 | A flood brake. Normal use never meets it. |
| Questions in one conversation | 20 | A subject that runs this long should be with a person anyway. |
| Messages per person per day | 40 | Your daily allowance, counted over the last 24 hours rather than reset at midnight. |
| Answers per month, platform-wide | 20,000 | A cap on the whole installation, not on you. |
Answers are also kept short: at most six sentences, with numbered steps where an order matters. The assistant is meant to take you to the right place in the guide rather than rewrite the article; the detail stays in the source.
If a question cannot be answered within 45 seconds the request is cut off and you are told the assistant is busy. That is not a fault — it is a decision to keep waiting short on a live screen.
The real reason for the caps is cost, and that cost sits with the platform, not with you: the model behind the assistant takes nothing from the AI credit your account spends on automations. Because the platform pays, the cap lives on the platform side too.
💡 Tip: If you want to protect your daily allowance, open the source article under the first answer instead of rephrasing the same question over and over. The detail you are after is usually already there.
The assistant does not disappear. The button stays where it is, the panel opens, it tells you plainly what happened, and then it shows you the route to a support ticket. A button that quietly vanishes makes people think the product is broken; an honest sentence plus a working alternative is better.
| What you see | What it means | What to do |
|---|---|---|
| You are writing too fast | The per-minute brake is on. | Wait a few seconds and send again. |
| This conversation is out of questions | The per-conversation cap. | Close it and start a new one, or open a ticket. |
| You have used up today's questions | Your personal daily cap. | Leave it to tomorrow, or open a ticket. |
| We cannot give AI answers right now | The platform's monthly cap is full. | Open a ticket; a person will answer. |
| The assistant is switched off | No AI key is configured in the installation. | Open a ticket; this is not a passing rush. |
In none of these is your question left unanswered: the support ticket route is open in every case, and the transcript is attached when you hand over. The detail is in From the assistant to the support team.
Never type any of these:
The reason is simple: the assistant has no use for any of it. It cannot look at your account, change a setting or perform any operation on your behalf, so knowing a password gains it nothing. What it does do is sit in the chat record — and move into a support ticket if you hand over.
⚠️ Note: The MERKUVA support team will never, under any circumstances, ask for your password, a verification code or your card details. If a message reaches you asking for these, do not reply to it — report it with a support ticket.
If you typed something like that by accident: change the password or key immediately, then open a ticket and ask for the record to be cleaned up. You cannot delete a chat message yourself.
The model behind an answer does not see your question in a vacuum; it is given a context. That context consists of exactly this:
| Reaches the model | Does not reach the model |
|---|---|
| The text of the message you wrote | Your name |
| The last few turns of the conversation, trimmed | Your e-mail address and phone number |
| The language you are writing in | Your portfolio, account and team names |
| The title-and-summary index of the guide | Your access tokens and keys |
| The relevant sections of the chosen articles | Your IP address and device details |
| — | Your invoices, payments and quotas |
| — | The messages and contacts in your inbox |
So when the assistant says it cannot look at your account, that is architecture rather than politeness: the data it would look at is never sent. Even the conversation history travels labelled only as "user" and "assistant"; your name does not appear there either.
Chat records are not used to train the model. What you write does not become training data.
Every answer the assistant writes is checked on the server before you see it. There are three gates:
The rule is therefore clean: you never see an answer without a source. If an answer is on your screen, there is at least one article link under it, and that link goes to a real article.
If you send exactly the same text twice in a row the model is not called again; the previous answer is repeated. That avoids both the wasted cost and the "I asked again and got a different answer" problem.
Unsourced questions and "I couldn't find this" answers are recorded separately. That list is what the guide is missing, and it is where new articles come from.
Your transcript is stored in the platform's database, tied to your account. Where it lives and how long it stays depends on whether the conversation became a support ticket.
On who can see it, the rule is single and identical to that for support tickets: a conversation is visible to its owner and to the support team. Your colleagues, the other members of your portfolio and the portfolio owner cannot see it. Support exchanges are personal.
⚠️ Note: You cannot delete a conversation or an individual message yourself. If you wrote something that needs to come out of the record, open a support ticket and say so.
If you cannot see the button in the bottom right corner, or the panel draws an information card instead of the question box, it is one of these:
What they all have in common: the support route does not close just because the assistant is unavailable. How to open a support ticket describes the second route.
No. The assistant's model runs on the platform's own account and never touches the credit your automations spend. For credit problems, see My AI credits have run out.
It is counted over your messages in the last 24 hours; there is no fixed reset time. As your oldest message passes the 24-hour mark, allowance frees up by itself.
There is no download button for the conversation itself. When you hand over, the transcript moves to the support ticket and becomes permanently readable on the ticket's page.
Most likely not. The platform's monthly cap may be full, or an administrator may have changed a setting. The ticket route is open — open one and a person will answer.
Not yourself. If you typed something sensitive, change that password or key first, then open a support ticket and ask for the record to be cleaned up.
So that no promise is made that cannot be kept. After a handover you are told that someone from the team is joining, without a figure in minutes; the actual response times are in Support limits, scope and privacy.
Nothing happens. Your text is given to the assistant as data, not as instruction. When such a request arrives, the assistant says it cannot comply and hands the subject to a person.
Was this article helpful?