AI customer support that never invents an answer

Your customers ask the same forty questions. Matter Chat answers them in seconds, from your own pages, with a link to prove it — and hands over the moment a question needs a person.

Start free See how it works

No credit card. Live in an afternoon.

Do you ship to Ireland?

Yes — we ship worldwide from our roastery in Rotterdam, at a flat €14.95. Orders usually arrive in 7–14 business days.

Shipping & delivery

Answers in seconds: At any hour, in the language the question arrived in.
Cites every source: A link to the page it used, on every single reply.
Refuses rather than guesses: Outside your content it says so and offers a human.
Live in an afternoon: A crawl and two lines of script. No project.

Two colleagues in a small independent business looking at a laptop together, one explaining something to the other.

The queue is full of questions you have already answered

Most support volume is not hard. It is repetitive, it arrives outside office hours, and the answer already exists on a page nobody found. The cost is not the difficulty of the questions but the time between a customer asking and anyone replying.

The same shipping, pricing, and account questions arrive every single day
Replies stall overnight and on weekends, so Monday starts with a backlog
Answers drift between agents because everyone remembers the policy differently
Customers ask in chat what your help centre already explains in detail

What it actually does for a support team

Not a general chatbot pointed at your website. A retrieval system that reads your published content, answers from it, shows its working, and knows where its knowledge ends.

Answers from your content

Retrieves the relevant passage from your own pages and answers using only that — never from a model's general memory.

A citation on every reply

The exact page it drew from, linked, so a customer can verify it and read further without asking again.

An honest refusal

A question your material does not cover gets a plain 'I don't know' and a route to a person, not a confident invention.

Escalation with context

The full transcript travels with the handover, so your agent opens with what has already been ruled out.

The hours you do not staff

Evenings, weekends and every timezone you sell into but have not hired for.

A ranked list of what to write

Every refusal is logged; on Growth and above they are grouped by demand into your help-centre backlog.

Grounded in your content, not in a language model's memory

Matter Chat reads your website, help centre, and documents, and answers only from what it found. When a question falls outside that material it says so rather than guessing — which is the difference between a bot that saves your team time and one that creates tickets.

Every answer cites its page: The link is part of the reply, not a footnote nobody scrolls to.
Refusals instead of invention: No answer is better than a confident wrong one. An invented refund policy costs more than a slow reply.
Scheduled re-crawls: Change a policy page and the bot changes with it. Only what actually changed gets re-indexed.
One source of truth: Because every reply comes from your published content, the bot cannot contradict what an agent said last week.

A knowledge base after a crawl, reading 4 of 4 pages indexed above a table of source pages, each marked ready with a chunk count and a last-updated time. — Everything the assistant can say comes from this list. Nothing else.

Escalation that arrives with context

The handover is where most support bots fail: the customer repeats themselves to a human who has no idea what was already tried. Matter Chat passes the full conversation with the escalation, so the first thing your agent reads is what has already been ruled out.

Hand off to a human: Route to live chat, email, or a ticket, with the transcript attached.
Slack and Telegram: On Growth and above, escalations land in the channel your team already watches instead of another inbox.
Availability-aware: Offer live handover during working hours and capture an email outside them, automatically.

Delivery channel settings showing an email address for instant notification and a webhook endpoint for a Slack incoming webhook, each with a button to send a test. — Escalations and captures route to email, a webhook, or both — whichever your team actually watches.

You can see what it is doing

Answer rate, conversation volume, negative ratings, and the ranked content-gap list that Growth and Agency plans generate. That last one is the most useful number in the product: it tells you exactly which help article to write next.

Answer rate over time: The share of conversations the bot answered rather than refused, tracked as your content improves.
Content gaps, ranked: Every refusal is logged and grouped, so the most-asked missing answer is at the top.

Six analytics cards showing conversations, answered rate, unanswered questions, negative ratings, handoffs to a human, and leads captured. — Answered rate and unanswered sit side by side on purpose — either one alone is misleading.

The honest part

A confident wrong answer costs more than a slow one

Every assistant meets questions its content does not cover. What it does next is the whole decision. Ours says so plainly and offers a person — because an invented refund policy is not a support failure you can apologise your way out of.

It answers only from what you indexed
Every reply links the page it came from
Outside that, it declines and hands over
Each refusal is logged as a content gap

A support specialist wearing a headset at their desk, mid-conversation with a customer.

The chat widget answering a shipping question with a specific reply and a numbered citation marker.

From nothing to answering, in an afternoon

There is no integration project, no data export, and nothing to migrate. The longest step is waiting for the first crawl to finish.

1
Point it at your site
Add your URL as a crawl source. It follows links, reads your published pages, and indexes what it finds. Documents can be uploaded for anything not on the web.
2
Check what it learned
Read the source list and ask it five questions you already know the answer to. Check the citation on each, not just the answer.
3
Set the limits
Lock the widget to your domains and set a daily cap. Both take a minute, and neither can be added retrospectively to a bill.
4
Paste two lines
Before the closing body tag, on any stack. The script is async, so it never blocks your page from rendering.

The Before you go live checklist reading 3 of 4 complete, with knowledge ingested, domains locked and bot switched on ticked, and snippet pasted on your site outstanding. — The dashboard tracks all four, so nothing goes live half-configured.

Questions people ask before they start

Short answers. If yours is not here, the assistant on this page will try it — and tell you honestly if it cannot.

How long does it take to go live?

Most sites are answering questions the same afternoon. You point Matter Chat at your URL, it crawls and indexes your pages, and you paste two lines of script before the closing body tag. There is no engineering project behind it.

What happens when the bot does not know the answer?

It says so plainly and offers a route to a human — live chat, a ticket, or an email capture depending on how you configure it and whether your team is online. That refusal is also recorded as a content gap so you can fix the underlying material.

Will it contradict my published policies?

It answers only from the content you gave it, and links the page it used on every reply. If a policy page changes, the next scheduled re-crawl picks it up — and you can re-crawl the page by hand the moment you publish it. The bot cannot invent a policy because it has nothing to invent from.

Can I control how much it costs?

Yes. Domain locks stop the widget answering on sites you do not own, per-visitor rate limits stop a single person burning your allowance, and a hard daily reply cap puts a ceiling on the day. You cannot be surprised by a bill.

All solutions