AI chatbot · Your model, your key

An AI agent on your WhatsApp number
that knows your prices, not the internet’s.

A general model that has never seen your price list is a liability on a customer thread. The AI in TicAte answers from material you give it, replies in the language the customer wrote in, and hands the conversation to a person the moment it is out of its depth — on a model you choose and a key you own.

What is an AI chatbot for WhatsApp?

An AI chatbot for WhatsApp is an agent that reads what the customer actually wrote and answers in its own words, instead of matching keywords to a scripted reply. In TicAte the agent is grounded in your own material — price lists, policies, FAQs — given a system prompt you write, and held to a turn limit, after which it hands over to a human. You connect the model with your own API key: Claude, OpenAI, Gemini, or any OpenAI-compatible endpoint including a model on your own hardware. The key is encrypted at rest, and every AI reply is logged on the thread.

Try it on your own material

What makes it usable on a real customer thread

Six things a general chatbot gets wrong

None of this is about the model being clever. It is about the agent being answerable for what it says on your number.

Grounded in your material

You give the agent its knowledge — price lists, policies, delivery rules, the FAQ your team already keeps — and a system prompt in your own words. It answers from that, not from whatever it read on the web.

Your model, your API key

Claude, OpenAI, Gemini, or any OpenAI-compatible endpoint — Groq, Mistral, Together, OpenRouter or a private model you host. Your key, encrypted with AES-256-GCM, your usage, your choice to switch.

Answers in the customer’s language

A customer writing in Tamil, Hindi or Malayalam gets an answer in the same language, without you maintaining four scripted versions of every reply.

It knows when to stop

A handover marker and a turn limit: the moment the agent is unsure, or the customer asks for a person, the bot stops and the conversation is assigned — with everything it learned already on the record.

Draft instead of send, if you prefer

Keep a person in the loop: the agent writes the reply, the agent-in-the-chair edits or approves it. Useful for the first few weeks, and for regulated conversations forever.

Every word is logged

Each AI reply is stored on the thread like any other message. You can read exactly what was said to a customer six weeks ago, because it is on the conversation, not in a vendor's black box.

Setting it up

An afternoon, not a project

Connect a model

Pick Claude, OpenAI, Gemini or a custom endpoint, paste your API key, choose the model. The key is encrypted before it reaches the database and is never shown again.

Give it your knowledge

Paste in what the agent may answer from: prices, delivery and return policy, opening hours, the questions your team is tired of retyping. Write the system prompt in plain language.

Decide where it may speak

Put the AI step inside a chatbot flow, or wire it into an automation to classify, summarise or extract a date. Set the turn limit and what a handover looks like.

Watch it, then loosen the reins

Start in draft mode with a person approving replies. When the answers are consistently right, let it send — and keep reading the logged threads.

AI agent: Sales assistant

ModelYour key · Claude, OpenAI, Gemini or custom

Encrypted

KnowledgePrice list, delivery rules, FAQ

Yours

LanguagesReplies in the customer’s language

Auto

HandoverOn doubt, or on request

Max 6 turns

AuditEvery reply stored on the thread

Good to know

AI chatbot questions

Whose AI is it, and who pays for it?

Yours. You connect your own provider account with your own API key — Claude, OpenAI, Gemini, or an OpenAI-compatible endpoint including a model you host yourself — and you pay that provider for usage. TicAte charges for the platform, not for tokens, and nothing in TicAte stops working if you turn the AI off.

Will it make things up about my prices?

It answers from the material you give it and the instructions you write, which is what keeps it on the rails. Start it in draft mode so a person approves each reply, read the logs, and tighten the prompt before you let it send on its own.

Can it hand over to a human?

Yes, and that is the point. The agent stops on a handover marker, on a turn limit, or when the customer asks for a person — and the conversation is assigned to the right team with everything it collected already on the contact record.

Is my customer data sent to an AI company?

Only the message text the agent needs to answer, and only when you switch it on. You choose the provider, so you choose where that goes; a self-hosted or private model keeps it on your own infrastructure. It is listed as an optional sub-processor in our privacy policy.

Does it work with the scripted chatbot?

They are built to be used together. The scripted flow owns the parts that must be exact — menus, qualification questions, booking — and an AI step handles the open-ended middle, then returns to the script.

What about the 24-hour window?

The AI works inside the same Meta rules as every other reply: freely for 24 hours after the customer's last message, and an approved template after that. Nothing about the AI changes what Meta allows.

Keep reading

Around the AI

The no-code chatbot builder

Triggers, buttons, lists, questions and handover — the structure the AI step sits inside.

WhatsApp catalogue

Show products, take a cart and send order updates without leaving the conversation.

WhatsApp Business API

The official API underneath all of it: how to connect, what Meta charges, how long it takes.

An AI agent on your WhatsApp numberthat knows your prices, not the internet’s.