Help Center

Docs

Set up your AI Customer Support Agent

Publish your help center, configure the agent, and start it answering, end to end

What the agent is

A customer-facing agent that answers from your published articles and hands off when it cannot.

The AI Customer Support Agent answers your customers' questions on your public help center, in a conversation, without them waiting for a person. It answers only from the knowledge base articles you have published to that help center, and it shows the customer which article the answer came from. When it cannot answer confidently, or when the question crosses one of your guardrails, it stops and hands the conversation to your team as a ticket with the whole exchange attached.

Setting it up is three steps, and all three finish on one page: Service, then the AI Agent tab. Publish the help center the agent answers on, set how it answers, then add the agent. That page shows the three as a checklist headed Set up your agent, each row carrying the one action that finishes it, and the checklist disappears once the agent is answering.

The order matters for step 1 and it does not for step 2. The agent has nowhere to answer until your help center is live, so publish first. But you can configure the agent and save it before you buy one, so the voice, the guardrails and the handoff message are already right on the day it starts answering.

An Ultra plan feature
The AI Customer Support Agent is available on the Ultra plan, as a paid add-on bought per agent. Publishing the help center underneath it, and changing how the agent answers, are available on Business and above. So a Business workspace can finish steps 1 and 2, see the value of a public help center, and have the agent fully set up before it moves up to the plan that turns it on.

Before you start

Three things to have ready.

  • A few articles worth answering from. The agent is only as good as your knowledge base. Five or ten solid articles covering the questions you answer every week beat fifty thin ones.
  • Permission to edit your knowledge base, the kb.edit key. Publishing the help center and changing how the agent answers both need it. Without it the first checklist row states the fact instead of painting a control you cannot use: An administrator with knowledge base access can publish your Help Center. Anyone who administers roles in your workspace can grant kb.edit.
  • Someone who can buy, holding the billing.view key. Adding the agent is a billing action, and the price is only readable by a billing.view holder, so a reader without it sees An administrator with billing access can add it. rather than a path. A reader who holds billing.view but does not manage the subscription sees its own line, An administrator who manages your subscription can add it. Both keys are granted by a workspace administrator under Roles & Permissions, and the buyer can be a different person from the one doing steps 1 and 2.

Step 1: Publish your help center

The public surface the agent answers on, and the article set it answers from.

Your help center is the branded, customer-facing view of your knowledge base, and it is the page your agent answers on. It is off by default, so nothing is public until you publish it, and you publish it without leaving the AI Agent page. If your help center is already published, step 1 is ticked: skim the article gates below and move on to step 2.

1

Press Publish on the first checklist row

Go to Service, then AI Agent. The first row of the Set up your agent checklist is Publish your help center, and it carries a Publish action that opens a small dialog on the same page. Nothing sends you off to another screen. If your plan is below Business the row says so instead, and if you cannot edit the knowledge base it names who can.

2

Choose your public address

The dialog has one field, Public address, sitting under the /help-center/ prefix. It is how the public portal resolves your organization, so choose something your customers will recognize. Use 3 to 40 lowercase letters, numbers, and single hyphens; a line under the field shows the finished address as you type, and a short list of names is reserved. Only a workspace admin can set it: if you are not one, the dialog asks an admin to set the address first and lets you publish from here afterwards. See Your Help Center Address.

3

Publish it

Press Publish. Your public address starts resolving straight away, with whatever articles already clear the three gates below, and the checklist row ticks over to Done. If the address is taken or malformed the dialog stays open and says which, so you can fix it in place.

4

Add your branding when you want it

That dialog publishes and does nothing else. Brand color, Logo URL, Default language and the rest of the public page live in Service, settings gear, Help Center, and so does the switch that takes the help center back off. See Your Public Help Center for that page.

5

Get the right articles onto it

In Service → Knowledge, make sure the articles you want the agent to use are published, set to visibility: public, and verified. See Your Public Help Center and Authoring & Verifying KB Articles for the full walkthrough of each gate.

The three gates decide what the agent can say
An article is in the agent's source universe only when it is published, visibility: public, AND verified. That is deliberate: the agent can only tell a customer something the customer could already have read for themselves. A draft, an internal-only article, or an article whose verification has gone stale is invisible to it, so if the agent will not answer a question you know you have documented, check those three flags first.

What the page tells you about your articles

An advisory line under the status, never a gate.

Once the agent is on, the AI Agent page prints one plain line under its status line saying how many verified articles the agent can answer from. If your help center has nothing in it yet, the page tells you in one sentence: Your help center has no published articles yet. Every question goes to your team until you publish one.

The readiness line is advice, not a gate
It is a signal about your content, and it changes nothing about setup. The three steps, and whether the agent reads as on, are decided by the help center being published, the configuration being saved, and an agent being added. An empty help center leaves all three exactly as they were, so you can add the agent first and publish your first article after: that is a legitimate order to work in. What the line is for is the case the three steps cannot see on their own, where every step is green and every conversation still goes to a person.

Step 2: Configure the agent

Instructions, tone, guardrails, handoff, and which articles it may use.

Everything about how the agent behaves is set on the Configuration tab of that same AI Agent page, and the checklist's Configure action takes you straight to it. You can set it and save it before you buy an agent. Changing how the agent answers needs the Business plan or higher; below that the form is readable but not editable and the page says why above it. Every saved change applies to the next conversation, so you can keep adjusting it while it is live. The defaults are deliberately cautious, so an agent you configure lightly errs toward handing off rather than guessing.

What you can set

  • Agent behavior. Its Handling choice is how far the agent goes on its own: Send everything to my team, Answer first, then hand off, or Let the agent handle it. The last one is never a dead end: when the agent cannot answer, the customer still gets a ticket.
  • Sources. Which of your public, published, verified articles the agent may answer from. Every eligible article is in by default; turn individual ones off here when an article is accurate but not something you want quoted back conversationally.
  • Identity and tone. The voice it writes in, how long its answers run, and the language it replies in.
  • Instructions. Free text describing your product, your support hours, and anything you want the agent to mention or avoid.
  • Guardrails. Switches for Never quote prices and Never promise refunds, plus your own phrases to withhold. An Always on set sits underneath them and cannot be turned off.
  • Handoff. The message the customer reads when the conversation moves to your team.

Nothing is stored until you press Save changes at the foot of the form. If you try to leave the page with an unsaved draft, it asks before letting the change go.

Instructions steer. The output check is what stops a reply.
Treat written instructions as guidance, not a guarantee: they shape how the model writes, but they cannot by themselves make a topic impossible to raise. What actually stops a reply is the output check that runs on every answer before it reaches a customer. It reads the drafted answer and hands the conversation to a person instead of sending it when the answer would quote a price, promise a refund, or accept liability, and whenever the subject falls inside the safety floor. That floor always hands off and cannot be turned off or lowered by an instruction.
Write the instructions your team would give a new hire
The instructions that work are the ones you would give a person on their first day: what you sell, who your customers are, when you are open, and the two or three things you never want said. Vague instructions produce vague answers. Reading a week of transcripts, then tightening the instructions, is the single highest-value thing you can do after launch.

Step 3: Start the trial or add an agent

A 14-day trial for a deposit, or buy an agent outright.

With the help center published and the agent configured, the last step is to add the agent. The third checklist row carries an Add agent action that opens a dialog on the page, with the price, what is included, and your trial if you still have one. There is no separate go live switch: the agent starts answering on your help center as soon as it is added. Without billing access the row names who in your workspace can do it instead, and on a plan below Ultra it says the agent is part of the Ultra plan.

1

Try it for 14 days

While your organization still has a trial, the dialog offers it first. Start it for a one-time $1.99 deposit, charged the day you start. The agent runs for 14 days, and the deposit is credited to your first invoice, so the trial costs you nothing extra if you keep it. The trial is limited to one per organization. This deposit is separate from the deposit on a Laureo plan trial, and paying one does not pay the other.

2

Or add an agent outright

Add an agent for $59.99 per month, each. An annual price is available, and the dialog quotes whichever your organization is billed on. You can run more than one agent, each billed separately. See AI Customer Support Agent Add-on for the full price list.

3

With no card on file, you finish on the billing page

If your organization has no payment method yet, the dialog says what happens before it moves you: you will add a payment method next, then land back on billing. It takes a second, deliberate press on Continue to payment to leave, and that hand off is the one part of setup that does not finish on the page.

4

Know what is included

Every agent includes 500 conversations a month. The allowance resets at the start of each month and unused conversations do not carry over. A conversation is counted the first time the agent answers, so a customer browsing your help center or a question the agent hands straight off never counts.

5

Add conversation credits if you need more

If a month runs long, buy an extra conversation credit pack. Purchased credits do not reset with the month and are drawn only after the month's included allowance is used. If you run out entirely, the agent stops answering and your help center falls back to article search, so customers can still find answers and still open a ticket.

Conversation credits are their own balance
Conversation credits are separate from Extra AI Credits and from your organization's monthly AI budget. Buying one never adds to the other, and agent conversations never draw down your AI budget. Each balance is shown on its own.

Check that it is live

Two minutes to confirm the whole chain works.

1

Read the line at the top of the AI Agent page

One sentence there says whether the agent is answering, and if it is not, what is missing. When everything is done it reads Your agent is answering on your help center. with a View your help center link beside it that opens your public page in a new tab. If you chose Send everything to my team, it reads Every question goes to your team. The agent is not answering. instead, which is that setting working, not a fault. If you can view the knowledge base, a second line underneath counts the verified articles the agent can answer from. Zero is worth fixing before you invite customers in, because the agent hands every question to your team until you publish one.

2

Open your public address as a customer would

Visit your help center address in a private browser window, signed out. You should see your branding, your public articles, and the agent ready to take a question.

3

Ask something you know an article answers

A good first test is a question whose answer sits in one specific published article. You are checking two things: that the agent answers, and that it cites the article you expected.

4

Ask something it should refuse

Then ask about pricing, or for a refund, with those guardrails on. The agent should stop and offer to pass you to a person, using the handoff message you wrote. A handoff here is the system working, not failing.

Where transcripts live
Every conversation is stored with its full transcript: what the customer asked, what the agent answered, which articles it cited, and whether it handed off. Read them on the Transcripts tab of the same AI Agent page. Each row leads with the customer's opening message, so a page of them scans at a glance, and the Outcome filter narrows the list to Sent to your team. Those are the most useful reading you have, because each one is a question your knowledge base does not answer yet. Turning those gaps into articles is what makes the agent better month over month. Reading transcripts needs permission to view tickets.

Plan availability

Ultra for the agent, Business and above for the help center it sits on.

  • AI Customer Support Agent: Ultra plan, as a paid add-on bought per agent. Ultra is what turns the agent on.
  • Public help center and help widget: Business and above.
  • Changing how the agent answers: Business and above, so steps 1 and 2 can be finished before the plan that runs step 3.
  • Knowledge base authoring and verification: included with the ticketing suite.

Related Articles

Your Public Help CenterAuthoring & Verifying KB ArticlesAI Customer Support Agent Add-onAI Autonomy ControlsTriaging Tickets in the Inbox
Previous
Conversation Quality
Next
AI Customer Support Agent Add-on