Everything between a question and a good answer
Follow one customer message, stage by stage: where it lands, what the answer is built from, what stops that answer going out wrong, and who takes over.
No credit card · a channel connected in minutes
“the March invoice still hasn’t come through”
Telegram · 09:12“says delivered but nothing came”
Web chat · 03:47“can I move my booking to Friday?”
WhatsApp · 18:30“Re: invoice for March”
Email · 10:11“how do I add a second operator?”
Help center · 11:05
Wherever your customer already is — including the group chat
Telegram, WhatsApp, Viber, web chat, email and your own help center land in the same inbox, with one history per person and the same assistant on every one of them. The hard one is the shared Telegram group, where a billing question, a bug and a feature request arrive braided together — so that is the one this page shows.
The group does not change. Your customers keep typing into one room the way they always have. What changed is that three of your operators can own three of those conversations without tripping over each other — and when a message lands in the wrong thread, you move it into the right one and it stays there.
Answered from your knowledge, and from your own systems
Every reply is assembled out of things you own: your articles, your FAQ, and live data pulled from your own API in the middle of the conversation.
It reads on its own. Before it changes a record, it asks — when you switch that on.
Any tool that changes something can be marked confirm-first. The assistant then stops, states what it is about to do in the customer’s own words, and waits for a clear yes before it runs. Confirm-first is a switch per tool, so a read-only lookup never nags and a refund never fires unasked.
Nothing the assistant cannot back up reaches your customer
After the assistant writes a reply and before anyone sees it, a second pass reads it back against the tool results from this turn, what the customer actually said, and your own content. Anything specific that is in none of those does not go out.
“Amounts, order numbers, transaction IDs, card numbers, recipient names, dates and statuses must come from a tool result, from the customer’s own messages, or from your knowledge base. If you cannot find the value, do not write it.”
Rule 2 of the platform instructions, appended after every business’s own prompt. No setting removes that rule.
Never invent an order, an amount or a name
The instruction sits after your own prompt and cannot be removed. The assistant can ask the customer for something it is missing. It cannot fill the gap itself.
Every link has to come from somewhere
Any web address in a reply has to already appear in a tool result, in something the customer wrote, or in your own content. A plausible-looking domain that nobody gave the assistant is blocked before the message is sent, with no model involved in that decision.
That domain does not exist. The assistant produced it because it looked like the sort of thing that ought to exist — which is exactly how a support bot sends your customer somewhere real, run by somebody else.
It does not only answer conversations. It starts them.
The same assistant, the same knowledge base and the same inbox also reach people before they ask: a popup on the page they are stuck on, a greeting that opens a chat on its own, a campaign from your own sending domain — and the numbers that tell you which of those actually worked.
- Popups, banners and lead capture
- Greetings that open a chat on their own
- Email campaigns from your own domain
- Live visitors, site analytics and goals
Email campaigns on Advanced and up. Popups, proactive greetings and analytics on Pro and Ultimate.
To a person, with everything already in front of them
You decide what a handoff is — an angry message, a dispute, a plain ask for a human, a keyword you name. The platform does the paging, the routing and the audit line. The assistant is a passenger in that decision.
You don't have to write your knowledge base
Give Chatonio your website. It reads the pages worth reading, writes the articles, harvests the questions you already answer in public, and drafts a system prompt in your own voice. Then you edit whatever it got wrong.
from 6 pages·14 articles·9 FAQ entries
Try it on any site, yours or not
Paste any URL and talk to an assistant built from it — no account, nothing to install. It takes about a minute.
The conversation is not the end of it
Once a conversation resolves, it turns into three things: work your team can pick up next time, an article the assistant answers from, and a reason to open the next conversation first.
Swipe — 3 groups
Around the conversation
On every plan. More seats from Advanced. The API on Pro and up.
- Roles and permissions
- Shifts, time zones, auto-routing
- Many operators, clean transfers
- Internal notes, never sent
- Saved replies and tags
- Web push and native apps
- Captcha and rate limiting
- REST API and signed webhooks
Back into the knowledge base
On every plan.
- Conversation into an article
- Rewrite with an instruction
- Translate, screenshots intact
- Re-index for the assistant
- AI on, off or automatic per chat
Out to the next customer
Email campaigns on Advanced and up, with one-click unsubscribe built in. The rest on Pro and Ultimate.
- See who is on your site now
- Site analytics and goals
- Greet visitors before they ask
- Popups, banners, lead capture
- Email campaigns from your domain
- Telegram broadcasts
The questions people ask first
What does the free plan actually include?
Can it do things, or only look them up?
How does it decide to escalate?
Will it answer before I’ve finished typing?
What happens when it doesn’t know?
Stop taking our word for it
Point Chatonio at your own website and talk to the assistant it builds. About a minute, no account, nothing to install — and if you like it, that assistant becomes your first project.
Plans, limits and what AI usage costs — see pricing