For support teams tired of guessing chatbots

Your customers are asking your site a thousand questions a day. It answers them with a source.

FlowChat crawls your site, builds a knowledgebase your visitors can actually trust, and serves answers that show their sources — with the confidence to say “I don't know” when the corpus doesn't cover it.

Start free trialTry the live demo

The category problem

Generic chatbots make up the answers.

When the model doesn't know, it guesses. Your customers learn to distrust it. They stop asking. Your support load goes up — not down.

What you actually get

Three things every other chatbot promises and none of them deliver.

01

Trust the answer.

Every reply links back to a real page on your docs. Your QA team can audit it. Your support team can stand behind it. No more emails that say 'where did the bot get that from?'

See how → 02

Find it faster.

It catches the question even when the customer doesn't use your exact words. Order numbers, SKUs, paraphrases — all handled. The right page surfaces in under two seconds.

See how → 03

Never guess.

When your docs don't cover the question, it says so — and shows the closest related pages. No more confidently-wrong answers reaching customers in writing, with your logo on them.

See how →

Try it. Now.

Ask the same questions your customers ask.

The widget below is live, against our own docs. Watch it find the answer in our site — and refuse when it shouldn’t guess. The right pane shows every step it took, in plain English.

Pick a preset or type a question to start. The widget runs the same retrieval pipeline real customers use, against this site as its corpus.

What pilot customers actually get

Three numbers your boss will ask about.

  • 41% fewer support emails about tier-1 stuff median across pilot tenants, weeks 4–12
  • 71% click-through on closest-related pages when the bot refuses visitors find their answer in docs instead of opening a ticket
  • ≤6 wks typical payback at mid-market support volume based on a $499/mo plan and a 30 hrs/mo support time saving at $80 loaded

Numbers from our active pilot tenants. The ROI calculator on /pricing uses the same conservative deflection ceiling we hold ourselves to here — so the savings you estimate are what we’d sign up to deliver.

From your URL to a cited answer

Six steps. Then your customer gets a real answer.

01

Find your pages

We read every page on the URLs you authorise. Just the URLs you authorise — nothing else.

your site
02

Strip the chrome

We pull out the navigation, footers, ads — leaving the content that actually answers questions.

clean text
03

Organise by section

We split your pages by heading, so a question about installation gets the installation section — not three random paragraphs.

indexed
04

Make it searchable

We index your content for two kinds of search: the question's meaning, and the question's exact words.

smart index
05

Match the question

When your customer asks something, we find the three best-matching pieces of your docs — and rank them.

top 3 sources
06

Write the answer

The reply cites every claim back to a real page on your docs. If the bot isn't sure, it says so instead of guessing.

with sources

Scroll to advance the pipeline

Six things competitors skip

Six things every other chatbot gets wrong.

Every team rolling out an AI chat surface for the first time discovers the same six failure modes — usually after their first hallucinated refund quote. We close them all.

Getting started
Installation
Configuration
01

It understands your structure.

Most bots cut your docs into arbitrary 500-word slices. Ours splits by your headings — so a question about installation gets the installation section, not three random paragraphs.

See why it matters →
your docs page read in full
02

Long pages still make sense.

A page about subscriptions covers billing, refunds, and cancellation. Our bot understands which part of the page actually answers the question — even when the answer references something three paragraphs earlier.

See why it matters →
contextual prefix

This chunk discusses API authentication, from the "Authentication" section of the Stripe API reference.

chunk

Use a secret API key to authenticate. Prefix it with sk_live_...

03

Context follows the answer.

Every section knows what document it came from. A snippet about “the export button” never gets retrieved alone — it travels with Account › Settings › Data Export as context.

See why it matters →
dense
bm25
rrf
rerank
04

It catches both meaning and exact words.

A customer asks “how do I cancel?” — that’s a paraphrase. Another asks “what does error A4B22 mean?” — that’s an exact match. Most bots only handle one of these well. Ours handles both.

See why it matters →

The bot answers with citations[#1], each linked to a real page[#2] on your docs[#3].

Sources confirmed
05

Every answer comes with a receipt.

Click any sentence in the bot’s reply, and you’ll see exactly which page on your docs it came from. Your QA team can audit every conversation. Your support team can stand behind every word.

See why it matters →
visitor

What's your refund policy for tax filings dated before 2024?

FlowChat

I can't answer that from your docs — the corpus doesn't cover pre-2024 tax filing refunds.

Closest related sources:

  • → Refund policy (general)
  • → Tax filings — current year
  • → Contact support
why this is a feature

Most chatbots would invent a policy. We treat the absence of grounded evidence as a positive signal — the user gets a real next step instead of a confidently wrong answer.

06

It says “I don’t know” out loud.

When your docs don’t cover the question, the bot doesn’t guess. It refuses, with a polite list of the closest related pages. The most-clicked screen in the whole bot — and the reason customers stop getting confidently-wrong answers from your support.

See why it matters →

By the numbers

Numbers we pin our reputation to.

We measure these every week and publish them. If a number slips, an alert pages us before any customer notices — and the next weekly post-mortem says exactly what happened.

  • 0%
    answer accuracymeasured weekly on a per-tenant eval set
  • <0s
    to first wordanswer starts before your customer finishes reading the question
  • 0
    hallucinationsescaped to users last week — every cited claim verified
  • 0%
    click on the refusal pagewhen the bot doesn't know, customers find the answer in your docs
  • <0min
    to first crawlsign-up to a live widget on your site

Pricing

Sub-cent per answer.

Five tiers from $49/mo Starter to custom Enterprise. Below is the starting end of the ladder; the live calculator on /pricing runs the real unit-economics formula.

See full pricing & calculator

Your site already has the answers.

Crawl it in the next 60 minutes.

Start free trial → View live demo