---
title: "Getting started"
description: "What a hiy is, and how to get one working in about a minute."
source: https://hiy.ai/docs/getting-started
---

# Getting started

## What a hiy actually is

Your twin learns from material you've genuinely made — writing, talks, posts, documents — and answers questions using only that. Three things make it different from a general chatbot:

- **It answers in your voice** — Not in the flat register of an assistant.

- **It shows its sources** — Grounded answers carry a citation you can open — the passage itself, plus a link to the original page where your source had one. Only your own words are ever quoted to a visitor.

- **It admits gaps** — When your material doesn't cover something, it says so and queues the question for you, rather than inventing an answer.

That last one is the point of the product. A twin that guesses is worse than no twin.

## Claim your handle

Your twin lives at an address you choose: **hiy.ai/your-name**. Type it on the
home page and, if it's free, it's yours as soon as you make an account —
typing a name doesn't hold it for you, so a name is only really taken once
someone has claimed it. A few names aren't available to claim: hiy's own pages
live at addresses of the same shape, and some words are set aside so a twin
can't sit at one. If yours is one of them the checker says so, and you can
pick another.

## Make one

Make an account and you're in a **guided create flow** — six screens, one
typed field — and an agent of your own in about a minute. It answers once you
train it, which is one press on its Overview and shows you the word count
first. Choose [what you're
creating](/docs/setup), then paste one address — a site, a docs URL, a
YouTube channel — and the next screen shows you everything found under it, so
you pick what your agent reads before a single page is fetched. The import
then runs while you answer the rest. **Nothing trains itself**: reading is not
indexing, and your agent starts answering when you press Train — or overnight,
if you switch **Train overnight** on. Material that isn't on one address
(files, an assistant export, a list) gets its own chip on the address screen,
and is noted against your twin for later. The last screen leaves you on your
new agent's own [Overview](/docs/agent-overview), where that import reports on
itself.

Nothing here blocks you: the address can be left blank, and every screen
after it can be skipped outright. Nothing you skip is lost — add material
from Knowledge any time, and see [How hiy is set up](/docs/setup) for what
stays editable afterwards and what doesn't.

Your twin is **private until you publish it**. Before then only you can see the page and talk to it in the Sandbox, so you can get it sounding right before anyone else can reach it. An account is required to build one at all: a twin speaks under a name, and we will not create one for someone who hasn't said the name is theirs.

## The six steps to a working twin

The **Set it loose** panel on [that agent's Overview](/docs/agent-overview) walks the whole journey and checks itself off as things actually happen — with a [setup bar](/docs/agent-overview) above it showing the same six stages as one reading. It stays visible until your twin is genuinely out in the world — and steps complete in any order, because it tracks what you've done, not what a wizard thinks you should do next.

1. **Choose what you're creating** — A twin — public to anyone, or private to the colleagues you invite — or a Support Agent for a product. The first screen of the [guided create flow](/docs/setup).

2. **Paste one address** — One address — your site, your docs, or your channel. Nothing is fetched by continuing. The next screen lists every page found under it, and what you leave ticked is what gets read.

3. **Ask it a question in the Sandbox** — Before you publish, only you can see your page, and you can talk to your twin freely in the [Sandbox](/docs/testing).

4. **Teach it something it missed** — Add and remove sources any time in **Knowledge** — a few things you actually wrote are worth more than a pile of generic filler.

5. **Publish** — Nothing is public until you publish. When it sounds right, publish — and unpublish any time.

6. **Share or embed it** — Share your link or [embed your twin](/docs/embed) on your own site — the panel stays visible until your twin is genuinely out in the world.

## Teach it

The more real material it has, the more it sounds like you. Add and remove sources any time in **Knowledge**. Quality beats volume: a few things you actually wrote are worth more than a pile of generic filler.

## Give it your voice

The **personality interview** in Voice & rules is a short set of questions about how you phrase things. It's the single biggest upgrade to how much your twin sounds like you. Voice is rebuilt with everything else, so the difference lands at the next [train](/docs/training) — straight away with **Train now**, or tonight if you have **Train overnight** switched on — and most people notice it the moment it does.

## Publish when it's ready

Nothing is public until you publish. Before that, only you can see your page, and you can talk to your twin freely in the [Sandbox](/docs/testing). When it sounds right, publish — and unpublish any time.

### Awaiting review

Publishing doesn't always make your page answer immediately. Because [you may
only twin yourself or something you own](/docs/acceptable-use), a twin whose
name we can't match to your account waits for a person to confirm it's really
you. While it waits, your dashboard says **Awaiting review** rather than Live,
and the publish step stays unticked — we'd rather hold the checkmark than have
you hand out a link that answers nobody. Usually within one business day, and
the decision shows on your dashboard — don't wait on your inbox: that same
line turns Live when it clears. You can keep testing and adding content
the whole time.

If a review is declined you'll see **This hiy wasn't approved** with the
reviewer's reason and what to send us. Your content and settings are untouched.

## Signing in again

You can sign in with Google or with an email and password. If you forget the
password, **Forgot password?** on the sign-in screen emails you a link that
lets you set a new one — the link is single-use and short-lived, and asking for
another always works.

For your own safety the screen says the same thing whether or not the address
has an account. That's deliberate: anything else would let a stranger check
which email addresses have a hiy.

If you typed an address on the home page before signing up, the confirmation
link carries it, so you can start on a laptop and confirm on a phone and still
land on the name you chose.