Wherever they ask, one hiy answers.
Your page on hiy.ai, live chat on your website, your own product, your inbox — four places, one engine. The Expert Twin leads with its public page; the Support Agent ships behind your own sign-in and your own token; the Team Agent answers colleagues who sign in. All of them end in your inbox, and each channel below says who answers there.
A page that answers, at hiy.ai/you.
Every hiy starts as a hosted page — a link that fits a bio, a talk, an email signature. Paste it anywhere people already ask you things, and the page answers them there.
See a live one
- Unlisted when you want it
- Reachable by anyone holding the link, indexed by nobody — that is the whole difference between unlisted and public.
- Links that unfurl
- Every twin page ships a social card, and a question link arrives with the question already asked.
- A launch kit included
- Your link, a QR code, and three ready-to-post announcements, drafted from your own material.
Who answers here
Expert Twin — its home — public, or unlisted and shared by link.
Team Agent — answers here too: invited colleagues, from their first sign-in.
Support Agent — deliberately not here. It answers only through your site (02) and your product (03), behind your token.
Live chat on your site, inline or a bubble.
The same hiy, as live chat on your own site. Inline, the embed sits in the page like any other element; as a bubble, it waits in a corner. One snippet either way — and it loads nothing until someone scrolls to it or opens it.
Embedding, end to end<script src="https://hiy.ai/embed.js" data-twin="demo-sam" async></script>- On every plan
- The inline embed ships on Free. The floating bubble is part of Founding.
- Nothing loads early
- The inline iframe mounts when a visitor scrolls near; the bubble's iframe is not created until the first open.
- The visitor's theme, and the label
- The embed follows each visitor's own device preference, and the AI disclosure stays on every plan.
Who answers here
Expert Twin — the tag above, exactly as shown.
Support Agent — the same tag plus data-token — a token your server mints per visitor. Without one, the widget answers nobody.
Behind your sign-in, on your pages.
Your backend mints a short-lived signed token for a visitor it already knows, and your hiy answers them and nobody else. This is the Support Agent's home channel — the whole of how it ships inside a product — and because it is plain HTTP, anything that can call an API can be the caller, an AI agent included.
The API, tokens, and limits// your server, per page view — never in the browser
const token = mintTwinToken(secret, twinId, userId) // HMAC-SHA256, 1h max
POST /api/chat
Authorization: Bearer v1.eyJ0d2luIjoi…
{ "slug": "your-agent", "message": "Does the export include attachments?" }
→ the answer, streamed, citations intact- Tokens your server signs
- HMAC over a shared secret, an hour at most — your users never need hiy accounts, and we never see your user list.
- Streamed, cited answers
- The same frames the hosted page renders, citations intact, over one POST.
- Every limit named
- Rates, caps and error shapes are all in the docs — including the files an AI agent can fetch to integrate this without a person.
Who answers here
Support Agent — its home channel. No token, no answer — nobody reaches it except through you.
Expert Twin — callable too: a public twin answers by its address alone, no token needed.
Every conversation ends in your inbox.
When your hiy can't answer, the thread doesn't die in a widget. A support agent hands it to your email with the transcript attached, and the visitor gets a receipt. One direction, stated outright: conversations end in your inbox — nobody writes to your hiy by email.
How handoff works- The transcript, not a ticket number
- The handoff email carries the conversation itself, so you answer the person rather than chase a reference.
- Alerts as they land
- A new lead or a new unanswered question can email you the moment it happens.
- A summary on your schedule
- Daily, weekly, or silence — who asked, what they asked, who opened your booking link. Off by default.
Who answers here
Support & Team Agents — the transcript handoff, and the visitor's receipt.
Every product — the alerts and the summary email.
Three products, the same rules everywhere.
An Expert Twin for you, a Support Agent for your product, a Team Agent for your colleagues. The channels above are where they differ; underneath is one engine, and the rules never change with the channel.
The three products, comparedTeach it once and it knows the answer on your site the moment it knows it anywhere; an answer cites its source on your own pages exactly as it does at hiy.ai/you. The honesty travels too: what it won't answer in one place it won't answer anywhere, and it is labeled AI on every surface — no plan removes that. Change a rule once, and it changes everywhere it answers.
How the loop behind this worksWhat's next, honestly.
From the roadmap's exploring tier — what we intend to build, not a commitment or a schedule.
Where everything standsSlack & WhatsApp. The same twin and the same rules, answering where your people already are.
Helpdesk & CRM. Handoff into Intercom or Zendesk when a person takes over, and CRM or Zapier out.
Voice. An audio twin.
What people ask before they ship it.
Can it answer in Slack or WhatsApp today?
No. You can paste your twin's link there and it unfurls with a card that carries the question, but the twin does not answer natively inside either app. Multi-channel is something we are exploring, not something shipped — the roadmap says exactly that, and so does this page.
Which channels does a support agent use?
Your website and your own product — the same script tag plus a token your server mints, and the API behind your sign-in. It deliberately has no public page: nobody reaches it except through you. When it can't answer, the handoff email carries the transcript to your inbox and the visitor gets a receipt.
Will the embed slow my site down?
It loads nothing until it's needed. The snippet is one script tag; the inline embed's iframe mounts when a visitor scrolls near it, and the bubble's iframe is not created until the first open. A visitor who never engages pays nothing.
Can visitors tell they're talking to an AI?
Yes — everywhere, always. Every surface is labeled AI, and no plan removes that. The only branding a plan can remove is the “Powered by hiy.ai” line.
What happens when it can't answer?
It says so in so many words instead of improvising — and a support agent offers to hand the thread to a person. Either way the unanswered question lands in your teach queue, so the gap closes for the next visitor.
Do I need a developer?
For your hiy.ai page, no — it's live when you publish. For your website, whoever can paste one script tag. For your own product, yes: your backend mints the visitor tokens, and the docs walk through it.
Which plans include which channels?
Your hosted page, the inline embed, the API and your inbox work on every plan. The floating bubble is part of Founding. Pricing has the full comparison.
They're already asking, everywhere.
Set it up once, and every channel above answers with it.