AI customer service chatbot with a human inbox behind it
For builders and agencies: an AI customer support agent under your own name. It answers from the knowledge base; your inbox takes the rest.
What the AI agent for customer service does
AI agents for customer support used to be decision trees; this one reads the conversation. You write who it is and what it may promise, once for the bot and again wherever a channel should differ, and it answers from the record you keep and calls for a person at the moment you described.
Answers from what it knows
The knowledge base is read on every message: the business profile, the opening hours, the FAQ in the order you put it in, the catalogue, and your notes on how you work.
Hands over to a person
A rule in your words — the person asks for a human, a manager or a call — the hand-off message it writes, and the teammates it assigns. Up to twenty rules on a source, none by default.
Answers when you decide
Around the clock, or only when nobody on the team is at work. The hours are set once, in the knowledge base, and every source reads them.
Follows up once
A nudge when the person goes quiet, written the way you told it to write one.
Remembers what it learned
What the AI asks for and where it stores the answer — an order number, a plan, a reason — into the contact’s own attributes. Up to forty captures on a source.
Quotes a price it has
Product photos from the catalogue the first time an item comes up, or only when the customer asks. It will not quote a price you did not give it.
The inbox takes the hand-off
A conversation the agent is running shows as Automated, in the same list as everything else. When it hands over, the thread says so — “The AI handed this chat to an operator” — and the contact is marked as having asked for a human until somebody opens it. Take over, and the bot stops answering; you do.
Five channels, one list
WhatsApp, Instagram, Facebook, TikTok and the website widget arrive in one inbox, and the agent answers on all of them — a 24/7 customer support chatbot, or one that only answers outside working hours — across seventeen entry points, each with rules of its own. WhatsApp’s 24-hour window is checked before you type, and an approved template is the way back in.
What it could not answer becomes the next FAQ
Gaps sweeps your own inbox for the conversations the agent handed to a person, groups the questions by word overlap and ranks them by how often they came up — counts of real conversations, never a score. One button writes the FAQ row with the question already in it.
Under your brand, in your code
Your name and logo go in at install. The design system, the API client and the proxy are copied into the project as source under MIT, so the app owns every line it runs. Add the accounts module and people sign up to you: every account gets a bot of its own inside your workspace, on your plan.
From the command to the first handed-over conversation
Four moves. The second is where the time goes, and the third is the one a support deployment actually argues about.
- 01
Run the command, pick the four modules
npx @chatfuel/wizard --modules automations,livechat,knowledge-base,contacts. Anything a pick depends on comes along and is named on screen, and the modules you did not pick are removed from the app before it is written.
- 02
Write what it knows
The business profile and the opening hours, the FAQ in the order the agent reads it, the products with prices and photos, and the notes on how you work. Import a CSV, TSV, text or Markdown file or paste the text, and every row is shown before it is written. Or add Copilot, hand it the PDF, the Office file or the URL, and it drafts the rows and writes them only after you approve.
- 03
Say when it hands over
Per source, in words: what the person has to ask for, what the agent says as it hands over, and which teammates get the conversation. The same page decides whether the AI answers always or only outside working hours, and every setting you do not touch keeps following Default.
- 04
Connect the channels, open the inbox
WhatsApp, Instagram, Facebook, TikTok and the widget connect through Chatfuel, so you ship an app rather than a submission to Meta. Conversations land in your team’s inbox as they arrive. The workspace needs a plan before the AI answers, and the wizard offers the trial on the way.
Who does what
The AI customer service agent and the inbox share one conversation. What each of them can do to it, read off the settings the agent has and the actions the thread offers.
| Name | On a conversation | AI agent | Inbox |
|---|---|---|---|
| Answer from the knowledge base | The record is read on every message the agent answers. | ||
| Hand over, take over | The agent hands over on a rule you wrote. An operator takes over with one button, and the bot stops. | ||
| Assign a teammate | The hand-off rule names who gets it. The inbox reassigns to a person, to Fuely AI or to nobody. | ||
| Send a WhatsApp template | The way back into a closed 24-hour window, filled in from a form. | ||
| Send a file | Catalogue photos from the agent. Images, audio, video or documents from the composer, whichever the channel carries. | Catalogue photos | |
| Follow up when the person goes quiet | One nudge, written the way you told it to. | ||
| Write what it learned to the contact | The agent stores a capture in an attribute. An operator keeps a note on the contact card. | An attribute | A note |
| Mark won or lost | Sets the contact’s sales stage and leaves the conversation as it was. | ||
| Close to a flow | The only close there is: the flow that takes the contact next. |
What a support conversation costs
Two meters, and a workspace plan they draw from. The inbox, the contacts and the knowledge base are part of the plan, whichever plan; the SDK and the app are free. The AI does not answer until the workspace has a plan.
| Name | What it is | Price |
|---|---|---|
| The meters | What comes out of the balance. | |
| AI input | Everything the agent reads on a message: the knowledge base record and the conversation so far. | $2.75 per 1M tokens |
| AI cached input | Input the model has already read once. | $0.275 per 1M tokens |
| AI output | What it writes back. | $16.50 per 1M tokens |
| Conversion event reported to Meta | Only when an event set in the ads module fires — a hand-off can be one. Nothing else on this page sends one. | $0.01 per event |
| Part of the plan | ||
| Inbox, contacts, knowledge base | Every module and every channel, on any plan. | Free |
| The SDK and the app it generates | MIT, copied into your project as source. | Free |
| The workspace | Where the bot lives and where the balance comes from. | |
| Business | One business, one bot, your own app. $20 of credits a month for the meters to draw from. | $20 a month |
| Agency S | Client bots, all on one balance. $100 of credits a month, shared by every bot in it. | $100 a month |
FAQs
Both, and the second is the point. The app carries your name and logo from the install, runs on your domain, and with the accounts module your users sign up to you. This page is written for the person who ships one, not for the person choosing among AI chatbots for customer service to rent, or buying a seat in an AI-powered customer service platform.
One command, and the support agent is in your app
The agent, the inbox and the knowledge base, under your name, in a codebase you can open.
npx @chatfuel/wizard