Support that answers itself — everywhere customers write

Feed it your docs in an afternoon. It replies around the clock in any language and only escalates what genuinely needs a person — no per-agent surcharge, because it ships inside every plan.

Day one looks different

10+
places it answers — widget, messengers, helpdesks — one brain across all of them
7+
ways to feed it: site crawl, uploads, Notion, Drive, your old help center, and more
Any
language in, same language out — detection is automatic
24/7
on duty — people join only when a case truly needs them

It quotes your docs — or it stays quiet

A bot that improvises burns trust in one message. This one answers strictly from what you gave it and routes the rest.

  • Retrieval first: the reply is built from your content, not thin air.
  • Low confidence means escalation — the case goes to the right team, picked by context and history.
  • Core product facts get extracted from your site and held constant.
  • Screenshots and photos in customer messages get read, not ignored.
Point it at your docs in the playground and watch how it answers before anything goes live.
Clip: a reply with three cited sources and its confidence score in the test playground

Your answers exist. Connect them.

Point it at the website, drop in files, sync Notion or Drive, or import the help center you already maintain — everything becomes answer material without rewriting a line.

Site crawl

Clip: pages from a site crawl turning into live knowledge sources on a refresh schedule

File uploads

Drop PDF, DOCX, or XLSX straight in.

Notion sync

Databases and pages, kept current.

Drive sync

Sheets and docs, kept current.

Old help desk

Pull articles out of Zendesk, Intercom, or Freshdesk.

Auto-found facts

Product basics harvested from your site and woven into replies.

Built-in articles

Your Sufox help center feeds the agent automatically.

Open the Knowledge Base

Nothing changes unless you change it

The agent never quietly rewrites itself. Every adjustment is deliberate, yours, and testable in private first.

Gaps, collected

Unanswered questions pile into a list. Write the answer once and it is pinned forever.

Clip: turning an unanswered question into a pinned reply with one click

Behavior rules

Direct its style, when it probes for details, and when it calls a person — always-on or keyword-scoped, per channel.

Pinned answers

For questions with one right answer, lock in yours. It always wins.

Voice settings

Length, warmth, escalation instinct — dials, not prompt engineering.

Guided flows

Multi-step troubleshooting that gathers facts before a human joins.

Hire once, staff everywhere

Live chat, Telegram, WhatsApp, Instagram, Messenger, Discord, Slack, your mobile app — plus Zendesk, Intercom, and Freshdesk if that's home. One personality, one knowledge set, tuned per channel.

Web widgetAn answer before the page even scrolls
Sufox Support
Typically replies in a few minutes
Do you offer refunds?
Yes — full refunds within 30 days, no questions asked.
Great, I bought it 12 days ago
You're within the window — refund started, money back in 3–5 business days.
Type your message…
WhatsAppFaster than the customer can type the follow-up
Sufox Supportonline
Can I switch my plan to yearly?13:42
Done — you're on annual now. That saves you 17%, next invoice Aug 12.13:42
Will I get an invoice for accounting?13:43
Yes — sent to your billing email just now.13:43
Type a message
TelegramBot or Mini App — the agent's the same
Sufox Supportbot
Where's my order #4127?09:18
It shipped yesterday — DHL tracking 992-4417. Arriving Thursday.09:18
Can I change the delivery address?09:19
Sure — reply with the new address and I'll update it with DHL.09:19
Message

The old bots weren't AI. They were phone trees.

One follows a script it was given. The other has read everything you publish.

Sufox AI agentScripted chatbot
TrainingReads your site and docs on its ownEvery reply hand-written by you
Dead endsEscalates to your team with the full storySays "please email support" and quits
LanguagesAnswers in whatever language the customer usedSpeaks only what you scripted
ImprovementSharpens from live conversationsFrozen until you edit it
ReachWidget, WhatsApp, Telegram, and 10+ moreThe widget, and that's it

The questions that come up first

Where the knowledge comes from, when humans enter, and how you keep every answer honest.

From sources you connect: an automatic crawl of your site, uploaded files, Notion, Google Drive, an existing help center in Zendesk, Intercom, or Freshdesk, or articles written in Sufox itself. Key product facts are also lifted from your site automatically. An agent only ever uses the sources tied to it.
Every reply carries a confidence score, and below your threshold the guessing stops: it can offer alternatives, pass the thread to your team, or both — your call. A run of shaky answers in a row can also trigger automatic handoff. Escalations arrive in the inbox with the full conversation context.
Replies are assembled from retrieved passages of your own content, behind anti-hallucination instructions and confidence gates. Questions with one right answer can get a pinned response that always outranks generation, and guidance rules fence off topics it must avoid or escalate.
The chat widget, WhatsApp, Telegram — groups and Mini Apps included — Slack, Discord, Facebook Messenger, Instagram, Facebook page comments, and email, plus Zendesk (tickets and Messaging), Intercom, and Freshdesk. One agent, one personality, per-channel behavior dials.
Yes — it recognizes each customer's language and replies in it, searching your knowledge across languages, so English docs can serve a question asked in Portuguese. System notices like handoff messages localize too. Locking it to a single language is also an option.
Talk to the agent yourself in the playground: every answer shows its sources and its confidence score. Vet changes to knowledge, personality, and rules before any customer sees them.
Never. Your sources and conversations serve only your own customers — nothing crosses accounts, and nothing trains anyone else's models. Language models are called purely for inference, and each agent reads only the sources linked to it.

Let the answers find your customers

One afternoon of setup, fourteen free days, zero card details.