An AI customer support agent that answers from your content and knows when to hand off

CenterChat puts an AI agent on your website that resolves the questions your team answers every day: policies, shipping, opening hours, pricing, order status and bookings. It shows where each answer came from, and when it cannot confirm something it says so and brings in a person.

What is an AI customer support agent?

An AI customer support agent is software that reads a customer's question, finds the answer in the business's approved content, and replies in plain language, or completes a task such as an order lookup or a booking. Unlike a scripted chatbot it copes with reworded and multi-part questions. A good one cites its sources, refuses to guess, and hands the conversation to a person when it should.

Updated by the CenterChat team

What CenterChat does for your support queue

Answers from approved knowledge
Import your help pages and policies (up to 25 pages per website import), add text, Markdown, CSV or JSON files, and write exact Q&A pairs for facts that must be precise.
Grounded, cited replies
Business facts come from your knowledge or a live tool result, and each answer shows its source. Unsupported questions get an honest "I can't confirm that" and a next step.
Live inbox and takeover
Every conversation lands in an inbox where your team can read it, take over and reply. A handoff request also sends a signed handoff.requested webhook.
Order status and bookings
Connect Shopify or your own order API for verified order status, and switch on the built-in scheduler for appointments.
A limit that fails safe
If you reach your monthly AI reply allowance, the agent stops generating replies, asks for the visitor's email and saves a lead. Nothing is charged automatically.
An improvement loop
The test studio, conversation analytics and improvement suggestions show which questions went unanswered, so you add the missing knowledge instead of guessing.

How to set it up

Most owners finish setup in one sitting. Agencies repeat the same three steps for each client.

  1. Step 1

    Add your knowledge

    Point it at your help pages and policies, then add Q&A pairs for exact facts such as refund windows and delivery times.

  2. Step 2

    Test the hard questions

    In the test studio, ask reworded, unsupported and adversarial questions and fix the gaps before you publish.

  3. Step 3

    Install and review weekly

    Paste one script tag, approve your domains, then review unanswered questions and handoffs once a week.

Which support questions to automate first

Start with questions that are frequent, have one correct answer in your own content, and cost little if a person has to step in. These resolve well on day one and free your team for the conversations that need judgement.

Leave refunds, complaints, account security and anything that needs a policy exception with your team. The agent can still collect the details and hand the conversation over.

  • Shipping, delivery and returns policy
  • Opening hours, locations and service areas
  • Pricing and plan questions
  • Order status, after the customer verifies their order
  • Appointment availability and booking

How CenterChat keeps answers accurate

When a question arrives, CenterChat searches your approved knowledge for the passages that match and gives only those to the AI model, with instructions to answer from them. This is retrieval-augmented generation. The reply shows which source it used, so you can check it.

If nothing in your knowledge supports an answer, the agent says it can't confirm it and offers your contact details or a person. That behaviour is deliberate: a confident wrong answer about a refund or a delivery date costs more than a handoff.

Measure resolution, not deflection

A conversation that ends without a ticket is not automatically a success; the customer may simply have given up. Track conversations where the customer got a sourced answer or a completed action, alongside handoffs, unanswered questions and repeat contacts.

CenterChat's analytics and improvement suggestions are built around that loop: see what was asked, what went unanswered, and which knowledge to add next.

What it does not do

So you can decide quickly whether CenterChat fits, here is what this part of the product does not cover today.

Questions, answered

More in the full FAQ.

Will an AI support agent replace my team?

No. It handles repetitive, well-documented questions and routine tasks, and passes everything else to your team with the conversation attached. Many teams use it to cover evenings, weekends and traffic peaks.

How does it avoid making things up?

It answers business questions only from your approved knowledge or a live tool result, shows the source, and says when it cannot confirm something. You can test any question in the test studio before publishing.

Can customers reach a person?

Yes. Visitors can ask for a person at any time, and the agent offers one when it cannot confirm an answer. Your team sees the conversation in the live inbox and can take over.

Which languages does it support?

In text chat the agent replies in the visitor's language, even when your knowledge is written in English. The built-in voices for voice calls speak English.

How much does it cost?

Plans start at $18.99 a month, or $15.08 a month paid yearly, for 500 AI replies a month and 5 agents. Every plan includes every feature, and there is no free plan.

Do I need my own OpenAI or Gemini API key?

No. Usage of the supported OpenAI and Google Gemini models is included in your plan's AI reply allowance.

Put an agent on your site this week.

Every feature on every plan. Start yourself, or tell us what you need and we'll set it up with you.