Skip to content
Getting started

Get your agent live

Paste your website address on the setup screen and wait about a minute. Wiredesk reads your site, drafts the agent's persona from what it found, loads those pages as its knowledge, and hands you a chat window to try it in. Nothing is live until you add the script tag to your site, so everything before that step is safe to get wrong.

Updated 2026-10-23 · 5 min read · More in getting started

Build the agent from your website

The fastest route is the one the app opens on after sign-up: a single box asking for your website address. It crawls the pages it can reach, writes a first draft of the agent's name, phone greeting and house rules from what it read, and files the pages themselves as knowledge the agent answers from.

It will not be perfect, and it is not meant to be — it is a starting point you correct rather than a blank form you fill. If you would rather start from nothing, Agents → New agent gives you the empty version.

  1. Paste your address

    Any public page will do; the crawler finds the rest from there. A staging site behind a password will not work.

  2. Wait for the crawl

    Usually under a minute. The site arrives as one knowledge source, read up to your plan's knowledge allowance, and flips from pending to ready once its pages are processed.

  3. Chat with it

    The wizard drops you into a conversation with the agent it just built. Ask it the three questions your customers actually ask.

Correct the two things that are usually wrong

Read the drafted instructions under Instructions before anything else. The draft is written from your website, so anything your website does not say — prices, plan limits, which carriers you ship with — is either absent or guessed. A guessed price that reaches a customer is the single most expensive mistake this product can make, and it always comes from hand-typed instructions rather than from the crawl.

Then check the knowledge list. If the crawler missed a section of the site, or picked up a blog archive you would rather it ignored, add and remove sources there. Knowledge is what the agent quotes; instructions are what it believes. Keep facts in knowledge.

Work through the setup checklist

The agent's overview carries a short checklist, and every line on it is verified against the database rather than ticked by hand — so it is telling you the truth about what is actually connected. Each unfinished line links to the page where it gets finished.

Three lines have to be done: the chat widget on your site, knowledge indexed, and lead capture on. A business that never wants the agent on its phone line is finished without a number, and the checklist says so: optional lines never hold it open. If a line is done in a way we cannot see from here — a snippet installed behind a tag manager, say — you can tick it yourself, or hide the card entirely.

Put it on your website

Installing it is one script tag, under Channels › Website chat › Install. The page gives you the snippet, instructions for Google Tag Manager, WordPress, Shopify, Webflow, Squarespace and React, a check that tells you whether we have actually seen the tag load, and a way to email the whole thing to whoever edits your website.

Once the check goes green, the agent is answering real people. Watch the first day's conversations rather than walking away — the first ten are where you find out what your website does not say.

FAQ

Related questions

No. You can build an agent, load knowledge, and test it in chat and in a browser voice call without adding a card. Tests are not free, though: Preview replies and browser voice minutes draw on the Free plan's monthly allowance exactly as customer conversations do, and a card only matters once you need more than that.

Still stuck? A person will answer.