Skip to main content
Founder pricing · The first 100 customers get Zupport.chat for $79 (normally $99) · Save $20 forever Claim a spot →
Founder pricing · First 100 customers · $79 instead of $99

Smart answers,
layered into
any website.

A knowledge assistant for your customers built on two layers of intelligence — curated scenarios for instant replies, AI for everything else, all running privately in the visitor's browser. The chat bubble on this page is one — trained on our own docs, so go ahead and ask it anything.

Pay once, embed forever
Zero data leaves the browser
Works on iOS & Android
Layer 01 · Scenarios
"How do I return an item?"
Match found in 47ms — exact answer returned from your curated scenarios.
★ instant · iOS ✓
Layer 02 · AI
"Can you ship this to Iceland?"
No scenario match — AI reads your shipping policy and answers from context.
⚡ desktop · 0.6B params
Layer 03 · Fallback
"Where can I email you?"
Configurable fallback message with your contact details.
95% match → ✓
no match → ai
02 · The cost of "smart" support

Your AI chat tool charges you every month, forever.

Tidio is $29/mo. Crisp $25. Intercom Fin $0.99 per resolution. They scale up as you grow, eat your margins as you succeed, and there's no escape because they own your conversation history.

Zupport.chat is $99 once. Or $79 if you're one of the first hundred. You own everything, the visitors' data never leaves their browser, and the only thing that scales with growth is your scenario library — which is yours to write, edit, and keep.

Tidiomonthly
Communicator
$29.00
$348/year · forever
Intercom Finper resolution
AI agent
$0.99 ×
scales with traffic
Zupport.chatonce
Founder pricing
$79.00
first 100 only · then $99
03 · How it works

Three steps. Roughly ten minutes.

Every Zupport.chat assistant goes from empty to embedded in three short steps. No coding. No infrastructure. No subscriptions.

01

Bring your knowledge

Upload PDFs, Word docs, spreadsheets, Markdown — or just point Zupport.chat at your existing website and pick the pages it should learn from. Crawl results are paginated and selectable, so the assistant only learns what you want it to.

2 minutes
02

Curate scenarios — AI drafts both sides

Type the questions your customers ask, or hit Suggest question and let the AI propose them from your sources. It then drafts the answer too. You skim, tweak, save. These reply instantly on every device — including iOS.

5–10 minutes
03

Embed one snippet

Copy a two-line script tag into your site. Works on Shopify, WordPress, Webflow, plain HTML — anywhere a script tag lives. That same snippet is what powers the chat bubble on this page.

30 seconds

Don't take our word for it — the widget in the bottom-right corner of this page was built with exactly these three steps. Open it and ask it anything about Zupport.chat — pricing, setup, supported browsers — and watch it pull the answer from our docs in real time.

04 · Two layers of intelligence

Common questions deserve perfect answers. Unique ones deserve a thoughtful one.

Every visitor question is checked against your curated scenarios first. If a match is found, the answer is returned instantly. If not, the on-device AI takes over — reading your documents and replying in context.

Visitor asks

"Do you ship to Iceland?"

The query enters the chat widget on your site, embedded in the visitor's browser.

∼ enters embed sdk
50ms
★ Layer 01 — Scenarios

Embed query, match against your curated triggers

Tiny embedding model (120 MB, cached forever) compares the question to every scenario you wrote. If similarity ≥ 82%, that scenario's answer fires immediately.

match → done · no match → continue
∼1.5s
⚡ Layer 02 — AI fallback

RAG retrieval + on-device LLM

Top 4 chunks from your documents are pulled by similarity, fed to Qwen3 (a small 600M-parameter language model) running in WebGPU. Answer streams token by token. English-only at launch.

desktop only · iOS gets fallback message
Device
Scenarios
AI fallback
Desktop & laptop (Chrome, Edge, Firefox, Safari 17+)
Yes — instant
Yes — full local AI
Android phone & tablet
Yes — instant
Yes — with consent on cellular
iPhone & iPad
Yes — instant
Configurable fallback message
05 · What you get

Everything a chat widget should have. Nothing it shouldn't.

Two-tier intelligence

Scenarios for instant exact answers. AI fallback for the long tail. The right tool for each kind of question.

Privacy by architecture

The visitor's question, your documents, the AI inference — all stay in their browser. No backend, no logs, no leaks.

Pay once, embed forever

One assistant, one payment, no usage caps, no per-message charges. The bill never grows with your traffic.

Learns from documents and your site

Upload PDFs, Word docs, Markdown, CSVs, Excel — or crawl your existing website and hand-pick the pages it should learn from. The assistant pulls the right paragraph for every answer.

One-line embed

Drop a single script tag pinned to a version. Works on Shopify, WordPress, Webflow, custom HTML — anywhere a script tag goes.

Update in minutes

Re-upload, re-publish, done. Returning visitors get the latest version within 24 hours via smart cache. New visitors immediately.

Visitor's
browser
Where everything happens
No data sent
Some
AI server
(Not in this picture)
06 · Privacy

Nothing leaves their browser.

Every other AI chat tool sends each visitor question to a third-party cloud. The provider sees the conversations. So does the model vendor. The visitor has no way to opt out.

Zupport.chat loads the AI model into the browser once and caches it forever. Every question, every retrieval, every response happens locally — there is no server to leak. GDPR, CCPA, HIPAA-adjacent: handled by architecture, not by promises.

07 · How we compare

Pricing as a commitment, not a subscription.

A side-by-side look at what you'll pay over a typical 24-month e-commerce lifecycle.

Zupport.chatTidioCrispIntercom Fin
Cost (24 months)$79 — $99 once$696$600varies, $1,000+
Pay-as-you-grow surcharge None Per seat Per seat Per resolution
Visitor data leaves browser Never Yes Yes Yes
AI quality control Curated scenarios + AI fallback AI only AI only AI only
Works on iOS Scenarios layer Yes (cloud) Yes (cloud) Yes (cloud)
Visitor data leaves the browser Never Yes (cloud) Yes (cloud) Yes (cloud)
08 · Pricing

One assistant. One payment.

Each price covers one fully-configured assistant — your scenarios, your documents, your branding. Add an assistant for another site or product line anytime.

Agencies & Enterprise — bespoke deal

High volume or many assistants? Let's talk.

Custompriced for the volume

Running an agency, deploying across many brands, or expecting more than 1M widget loads / month on a single assistant? Send us a note with what you're building — we'll come back with a price that makes sense.

  • Higher-volume traffic beyond the 1M / month cap
  • Bulk pricing on multiple assistants
  • Per-client setup for agencies
  • Internal portals across teams or brands
  • White-glove onboarding if you need it
  • Direct line to the founder, not a ticketing queue
[email protected]
Typical reply in one business day · no robots

Payments handled by Polar.sh as Merchant of Record — VAT, GST, and sales tax automatically collected and remitted in every country. 14-day refund if it doesn't fit.

09 · Honest answers

Things you're probably wondering.

Can I try it before I buy?

You already are. The chat bubble in the bottom-right corner of this page is a real Zupport.chat assistant, trained on our own documentation. Open it and ask anything — pricing, how the embed works, what browsers we support, what's in the box. The response you see is the response your visitors would see, with your knowledge instead of ours.

Why are scenarios necessary if you have AI?

Three reasons. First, iOS Safari can't run a 400 MB AI model — scenarios are the only way to give iPhone visitors a useful answer. Second, even on desktop, scenarios reply in 50ms vs ~1.5s for the AI — much better UX for common questions. Third, scenarios are deterministic — you control the wording. The AI is your safety net for the long tail.

What does "running in the browser" actually mean?

The AI model is downloaded once to the visitor's device (cached forever) and inference happens locally using WebGPU. After the first visit, no model download, no network calls during chat. Compare that to Intercom or Tidio, where every message is sent to their servers.

How long is the first-visit download?

About 30–90 seconds on broadband for the AI model (~520 MB). Cached forever after that — return visits ready in 1–2 seconds. On cellular, we ask permission first. On iOS, no model download at all — scenarios only.

Can I use it for more than one website?

Each $79 (or $99 once founder pricing ends) buys one assistant tied to one knowledge base. Add a second site or product line by buying another assistant. There's no monthly fee no matter how many you have.

How do I update the knowledge base?

Re-upload files in your dashboard, click Build, click Publish. New visitors get the latest version immediately. Returning visitors with cached data refresh within 24 hours via TTL. Each assistant supports up to 30 source files (up to 5 MB each, 100 MB total).

How big can my knowledge base be?

Up to 30 source files per assistant, 5 MB each, 100 MB total. For best answer quality, keep the knowledge base under roughly 1,000 chunks — about 200–300 pages of content. The embed widget scans every chunk on each query, so a focused, well-curated knowledge base gives sharper answers than a huge one.

How does the AI know what to say?

It only answers from your uploaded documents. The system prompt explicitly tells the model: "Answer only from the context provided. If it isn't there, say you don't know." So if your shipping policy says "we ship to EU only," the AI won't invent shipping to Iceland — it'll say it doesn't know. Garbage in, garbage out — but if your docs are good, the answers are good.

What if the AI gives a wrong answer?

Two layers of defense. First, write scenarios for your high-stakes questions (returns, shipping, payment) — those are exact text you control, no AI involved. Second, the AI is constrained to answer only from your uploaded documents, so its accuracy is a function of your docs' accuracy.

What happens to data when I delete my account?

Your dashboard data is wiped from Firebase within 24 hours. Your vectors.json file is deleted from our VPS within 24 hours. Visitors who have your assistant cached locally will lose access on next vector refresh (≤24h). Nothing of yours persists anywhere on our infrastructure.

Is there a refund policy?

Yes — 14-day money-back, no questions asked. If the assistant doesn't fit your site or you change your mind, email us and we'll refund. Refunds processed by Polar.sh, our payment processor.

What languages does it support?

English only at launch. The on-device model is small enough to fit in a browser, which means it doesn't yet have the strength to handle other languages reliably. If you need multi-language support, hold off — we'd rather tell you up front than ship something half-baked.

Ready to answer
everything?

One assistant. One payment. Lifetime of answers.

Get yours for $79