AI Chatbot

Overview

The AI Chatbot is BOOKBEY's flagship product: a ChatGPT-powered assistant that answers your customers on WhatsApp, in their own words, using only the material you train it on. It runs on a simple prepaid wallet — there is no monthly plan to buy and it is independent of your WhatsApp API plan.

It is a different product from the Website Chatbot, which is a rule-based menu widget for your website and never uses AI. The AI Chatbot understands free-text questions, paraphrases naturally (including Hinglish), captures leads, runs broadcast campaigns, and hands off to a human when it should.

How it answers (and what it costs)

Every incoming message runs through four layers, in order. This is what keeps quality high and token cost low:

If nothing is relevant, the quota is empty, or the API is down, the bot gives an honest handoff and logs the question as unanswered so you can train it. Most everyday FAQ traffic is served by layers 1–3 — free of tokens.

Getting started

Four steps:

1. Buy the AI Chatbot (one-time setup) from your client area.
2. Connect your WhatsApp number on the Channels page so the bot can receive and reply.
3. Train it — add your common questions and answers under the Training tab.
4. Top up credit so it can send, then go live. Test it from the Training tab first.

Credits & messages

Your wallet holds two balances. The first is messages (credits):

• One message is debited for every message — both inbound (the customer’s question) and outbound (the bot’s reply). A typical exchange is 2 messages.
• Messages never expire. They simply sit in your wallet until used.
• You’re alerted when the balance runs low (at 500 left) and again at zero.
• When messages hit zero the bot stops sending — top up to resume.

Even a free (layer 1–3) answer still uses one message — the message count is about WhatsApp traffic, separate from AI token usage below.

AI tokens explained

The second balance is AI tokens. Tokens are only spent when the model actually writes a reply (layer 4) or does a meaning-lookup (layer 3). Trained-answer hits and guard handoffs cost zero tokens.

• Tokens are topped up with each credit pack and restored to the pack’s allowance on every purchase.
• If tokens run out, the bot automatically falls back to trained answers only (still answering FAQ traffic for free) and hands off anything it can’t match, so it never goes silent.

The more you train, the more traffic is served by the free layers and the fewer tokens you spend.

Top-ups & pricing

Credit is bought in a top-up pack. Each pack adds messages to your wallet and restores your AI-token allowance. The current default pack grants 20,000 messages + 10,000 AI tokens for a fixed price (shown at checkout); your price per message is simply the pack price divided by the messages it grants.

Every movement — purchase, grant, adjustment, and usage — is written to your credit history (the ledger) under the Credit tab, so you can audit exactly where credit went. Pack contents and price are set by the platform and may change; the checkout always shows the current figures.

Sending limits & anti-ban

Three things govern how fast and when the bot sends:

• Wallet gate — sending requires messages left in your wallet. At zero, sending stops.
• Throttle — a minimum spacing between messages (your throttle setting), so replies and campaigns never burst. Campaigns release one message per interval per number — never blasted — to protect your number from bans.
• Business hours — if you enforce hours, messages outside them get your configured after-hours reply instead of an AI answer.

Training the bot

Under Training you add question & answer pairs — your FAQ, policies, pricing lines, opening hours, and so on. These are the only source the bot answers from, which is what keeps it accurate and on-brand.

• Add an answer, and the bot immediately uses it for matching questions (free, layer 2).
• Test the bot right there. A test costs the same as a real reply (it uses your wallet), so what you see is exactly what a customer gets.
• Keep answers short and specific — one topic per answer matches better than a wall of text.

Improving answers

The Improve tab is where the bot gets smarter over time:

• Gaps — every question the bot couldn’t answer is logged here. Turn the common ones into trained answers.
• Review — model-written replies can be reviewed; approve good ones (they become trained answers and are then free) or dismiss the rest.
• Freshness — answers are periodically re-indexed so meaning-matching stays sharp.

Campaigns (broadcast)

Under Campaigns you can broadcast a message to a saved audience:

1. Write the campaign and pick a saved lead group as the audience.
2. Send — it’s released through the anti-ban queue: one message per interval per number, never all at once.

The AI Chatbot pays for campaigns from your prepaid wallet (no plan needed) — each queued message debits one credit. Stop is final for a run: any unsent messages leave the queue. Deleting a campaign also removes its unsent messages.

Leads & audiences

Conversations capture leads automatically — name, phone, email and whatever else the chat collects. A repeat phone number updates the existing lead rather than creating a duplicate.

• Organise leads into groups; a group is what you target with a campaign.
• Leads to follow up appear on your Overview so nothing goes cold.

Settings reference

Everything under Settings:

Brand-safety guard

The guard runs before any matching or model call. You list the topics the bot must never engage with (legal advice, competitor pricing, anything off-limits for your brand). A message that hits one is handed to a human with your handoff message — the model never sees it. This, combined with answering only from your trained content, is what keeps the bot from going off-script.

Connecting WhatsApp & handoff

The AI Chatbot works on your WhatsApp number:

1. Connect your number on the Channels page and keep it connected.
2. Inbound WhatsApp messages are delivered to the bot automatically; it replies on the same number.
3. Set an owner WhatsApp number so guard hits, after-hours contacts and anything the bot can’t answer are routed to a human.

No code to embed — this is a WhatsApp agent, not a website widget. (For a website widget, see the Website Chatbot.)

Troubleshooting

• Bot isn’t replying: check the wallet has messages left, and the WhatsApp number shows connected on Channels.
• Replies are generic / wrong: add or sharpen trained answers for those questions; the bot only answers from what you train.
• “Using trained answers only”: AI tokens are exhausted — top up to restore full model replies. FAQ traffic still answers for free meanwhile.
• Nothing sends outside certain hours: business-hours enforcement is on — adjust hours or the after-hours reply in Settings.
• Campaign paused/short: it releases one per interval by design; a low wallet also stops it — top up and resend.

FAQ

Do I need a WhatsApp API plan too? No — the AI Chatbot runs on its own prepaid wallet, separate from API plans. Does every message cost a token? No — only model/meaning replies use tokens; trained-answer hits are free (they still use one message credit). Can I change the model? Yes, the model is a configuration setting. Will it make things up? It’s instructed to answer only from your trained content and to hand off when it can’t — that’s the core safety design.