Skip to main content

What Centillion is

An AI agent that answers your customers from knowledge you control, on your website, WhatsApp, email and phone.
Centillion builds AI agents that answer customer questions. You give an agent the material your business already has, its website pages, its policy documents, the answers your support team types out fifteen times a week. The agent answers from that material and nothing else. Then you put it where your customers are. A chat on your website, WhatsApp, Messenger, Instagram, Slack, your support inbox. The same agent, the same knowledge, wherever the question arrives. Enable Helpdesk and switch on escalation for a channel, and what it cannot answer becomes a ticket for a person. Start with the Free plan for website chat. Connected channels require Hobby or above; Helpdesk, campaigns and voice require Standard or above. Each workspace has its own plan, limits and credit balances.

The four things you will do

Create an agent

Name it, tell it who it works for and how to behave.

Add knowledge

Import your website, upload documents, write exact answers for the questions that matter most.

Check training

Sources train automatically. Check progress and test the answers.

Put it somewhere

One snippet for your website, or a sign-in for each messaging channel.
In a hurry? The quickstart runs through all four.

Everything the agent can do

One article per capability. Each one explains what the feature does, shows the dashboard, and walks through setup and daily use.

Reach your customers

AI chat widget for your website

A floating chat bubble, an inline chat or a help page, installed with one snippet.

WhatsApp AI agent

Answers on your WhatsApp Business number, inside the 24-hour window and with templates.

AI email support agent

A generated address, a forwarded company address or a connected Gmail or Outlook inbox.

Phone numbers for your AI agent

Import Twilio or SIP trunk numbers and choose which agent answers their calls.

Text messages (SMS)

A US business number your customers can text, answered by your agent.

Messenger

Conversations from your Facebook Pages, with optional comment tickets.

Instagram

Direct messages answered, comments turned into tickets.

Slack

Mention the agent in a channel and get the answer in the thread.

Shopify AI customer support

Chat in your storefront theme, with product search, carts, orders and returns.

Salesforce

Cases answered by email or as drafts, chats handed to Salesforce live chat.

iOS and Android SDKs

Your agent inside a native app, with signed-in users and app actions.

Zapier

Send leads to your apps and use the agent in automated workflows.

Win customers back on WhatsApp

Outbound template campaigns to your own contacts, with replies handled by the agent or your team.

Build your agent

Build an AI customer support agent

Instructions, guardrails, model choice and side-by-side model comparison.

Train it on your website and documents

Files, websites, text, Q&A, Notion pages and support tickets, with weekly auto-resync.

Custom tools

Call your API, run website code, show interactive cards in chat.

Collect leads and data

Structured details captured during a conversation and delivered by webhook or email.

Test it before customers do

Simulated customer conversations, scored by a judge.

Mastermind

Ask about your agent and change it in plain language, with approval before changes unless you choose auto-approve.

Releases, shared tools and workflows

Publish versions, reuse resources, route to specialists, measure outcomes.

Support

AI helpdesk with human handoff

Tickets, teams, AI drafts and live takeover from one inbox.

Product issues in Notion

Group support reports into one bug and keep its progress visible.

Customer data

Connect HubSpot, Salesforce or Intercom

Sync contacts in, log what the agent learns back to the CRM.

Stripe billing answers in chat

Identified customers see their invoices and subscriptions.

Identity verification

Authenticate signed-in website users in the widget and sync their contacts.

Churn reasons

Why customers leave, from every conversation, grouped by reason.

Answers come from your knowledge

This is the part worth understanding before anything else. An agent does not know about your business until you tell it. When a question arrives, Centillion searches what you have given that agent, picks the handful of most relevant passages, and writes an answer from them. Material you have not added is material it cannot use. That is the trade. The agent will not confidently invent your refund window, and it will also not guess one when you forgot to tell it. Both of those follow from the same design. Knowledge is versioned. Saving a source queues training automatically, and a complete training run makes the updated content searchable in the agent’s current configuration. Removing a source or excluding a page updates that configuration too. A published release keeps the knowledge saved with its version, including shared workspace knowledge. Later source edits or removals do not change that saved knowledge. After training finishes, capture and publish a new version to use the changes in new conversations. Existing conversations stay on the release where they started.

When the agent should not answer

Some questions want a person, and the agent is built to hand those over rather than improvise. A customer asking for a human pauses the conversation. So does your own team replying in WhatsApp or Facebook, so two answers never go out at once. Public comments on social posts are never answered automatically at all. Handing over is something you switch on rather than something you get. Helpdesk has to be enabled, the escalation action has to be on, and it has to cover the channel the customer is writing from. Miss one of those and the agent has nowhere to send them, so it keeps trying to answer. The action ships covering website chat, the help page and the API, so WhatsApp, email and the social channels need adding by hand. With that configured, what needs a person lands in Helpdesk, with the conversation attached. Outbound WhatsApp campaigns can send replies straight to your team too. Choose Human takeover when creating the campaign with Helpdesk enabled. Each recipient’s first reply opens a ticket and pauses AI replies until your team closes the ticket or allows the agent to answer again. Teams can also start website and mobile SDK conversations with a person using Human-first conversations, and choose whether ticket closure resumes AI in Human handoff settings. Email support can also go to Zendesk. A Create ticket action pauses the original email conversation while your team answers in Zendesk, then resumes AI replies when the linked ticket is Solved or Closed.

What it is not

  • It is not a search box. It answers in sentences, and it follows the conversation rather than treating every message as a fresh question.
  • It is not a bulk messaging tool. Agents reply to people who wrote first. The exception is an Outbound campaign: an approved WhatsApp template to customers you already have, who agreed to hear from you.
  • It does not learn from conversations. It learns from the knowledge you add and update. A conversation that went badly does not silently change future answers, which means you can fix the cause instead of hoping.

Where to go next

Start with the quickstart if you have an agent to build today. Start with Adding knowledge if you already have one and want it to answer better. Start with My agent is not answering if something is wrong right now.
Last modified on October 20, 2026