The AI agent answers customer questions on live chat, email and WhatsApp using only your published help center articles. Turn it on per inbox under AI assistant in your workspace settings. When it cannot help, or the customer wants a person, it hands the conversation to your team with a short summary.
Where the answers come from
The AI agent only uses:
- your published help center articles (drafts are never used),
- your workspace name and business hours.
For each question it searches your help center first. If no article is relevant enough, it does not guess: it hands the conversation to your team instead. Every answer links to the article it came from, so the customer can read the full text.
This means the quality of AI answers depends on your articles. See Write and publish help center articles.
When it hands off to a person
The AI agent passes the conversation to your team when:
- the customer asks for a human, for example "can I talk to a person?", or selects Talk to a person in the live chat,
- the topic is sensitive: refunds, chargebacks, complaints, legal questions, account security, or closing an account,
- no published article matches the question well enough,
- the customer says an answer did not help, twice (live chat shows "Was this answer helpful?" under each answer),
- the conversation reaches 8 AI replies,
- your plan's monthly AI conversations are used up, or the AI service is unavailable.
How answers reach the customer
- Live chat: the answer appears in the chat window with a link to the article, under an AI assistant label.
- Email: the answer is sent as a reply from "Your workspace via Kora Reply", signed "AI assistant" and marked as an AI-generated reply, with the article links under the text. Automatic replies (for example out-of-office messages) and emails that fail sender checks never get an AI answer.
- WhatsApp: the answer is sent to the customer's WhatsApp with the article link.
What your team sees
A handed-off conversation lands in the inbox as an open, unassigned ticket. A summary sits above the first message with:
- the customer's problem,
- what the AI agent already tried,
- the articles it referenced,
- what we know about the customer,
- a suggested priority.
The customer is told that a person will reply, with an estimate based on your response targets and business hours.
What it will not do
- It does not promise refunds, discounts, compensation or deadlines.
- It does not change orders, accounts or anything else: it only answers and hands off.
- It tells customers they are talking to an AI and that they can ask for a person at any time. Every AI message carries an "AI assistant" label in the chat widget and in emails, on every plan; this label cannot be switched off.
- Text inside your articles is treated as information, never as instructions, so an article cannot change how the agent behaves.
Limits and cost control
Each plan includes a number of AI conversations per month (50 on Free, 300 on Starter, 1,500 on Team). A conversation counts once, however many messages it has. Repeated questions are answered from a short-term cache while your articles have not changed, and the cache is cleared as soon as you publish, edit or unpublish an article.
On the Free plan, AI answers can also pause until the next month when Free workspaces together use up the share of the AI budget set aside for Free. New conversations then go straight to your team, exactly as when your own monthly AI conversations run out.
If the AI service is down, customers get an honest message, up to three matching help center links, and the conversation goes straight to your team. They are never left waiting on a spinner.