Messaging
AI chatbot that answers on WhatsApp, Telegram and web chat around the clock
AI Chatbot writes a short, knowledge-base-grounded reply to every incoming customer message, tags the intent and its own confidence, asks for contact details when it sees buying intent, and hands the chat to your team when it isn't sure.
- As a workflow node
agent.chatbot- Default model
- Claude Sonnet 5
- Economy model
- Claude Haiku 4.5
- Tools
- 7
- Report
- Chatbot insights
What the agent does
This AI chatbot works on WhatsApp, Telegram, Slack, Discord, Messenger, Twilio SMS and the web chat you embed on your site. Before replying it reads the conversation history, so it never asks for something the customer already said. Every factual question (products, prices, stock, shipping, returns, opening hours) is looked up in your knowledge base; if the answer isn't there, it says it will check with the team and keeps its confidence low. Web search is only for general facts unrelated to your business.
Replies are chat-sized: 1–3 short sentences in the customer's language, plain text, one question at a time. When it sees purchase or booking intent it asks for a name and an email or phone number, one field at a time and explaining why, and saves the details as a lead on the Leads page. It hands the chat to your team when the customer asks for a person, is angry, raises a legal issue, a refund or a complaint it can't resolve, or when its confidence drops below the threshold; asked directly, it never pretends to be human.
Conversations show up in the Inbox; when you take one over, the bot stays silent in it. The recommended setup is the WhatsApp and Telegram AI chatbot template. If no active message-triggered workflow matches, incoming messages are still answered with the project's active AI Chatbot settings, or the defaults if there are none.
Inputs / Outputs
Inputs
- Role and tone
- Describe in one paragraph who the bot is and how it talks: role, tone, what it won't do. Example: “Friendly order assistant for our café who keeps answers short and never haggles over price.” Your brand voice already comes from the project.
- Knowledge base
- The documents replies are based on: FAQ, price list, returns and shipping policy. Pick a specific knowledge base to limit the search to it; leave it empty to search your project's knowledge bases.
- Hand-off threshold
- The bot rates every reply with a confidence score from 0 to 1 and hands off when it falls below this value (default 0.4). Around 0.3 suits sales bots and 0.5 support bots as a starting point.
- Collect contact details
- When on (default), the bot asks for a name and an email or phone number once it sees purchase or booking intent and creates a lead; when off, it never asks for personal data. Tell it in the task to link your privacy notice when it asks.
- Team working hours
- Used in the hand-off message to tell the customer when to expect an answer. Example: “Weekdays 09:00-18:00, Saturday 10:00-14:00” or
09:00-18:00 Europe/London. Leave it empty and no hours are mentioned.
Outputs
- reply
- The plain-text reply for the customer: 1–3 short sentences in their language, delivered on the channel the message came from.
- intent
- The message's intent: question, purchase, support, complaint, booking, small talk, spam or other. Useful for routing in a workflow.
- confidence
- How sure the bot is that its reply is correct and grounded in the knowledge base (0–1). Values below the threshold trigger a hand-off.
- leadCaptured
- Contact details the customer shared in this message (name, email, phone, interest). When present, a lead is created or updated.
- handoff
- Whether the chat was handed to a human, and why. On hand-off the conversation switches to Human in the Inbox and your team is notified.
Saves a report to the Reports page on every run: Chatbot insights.
Tools
- Knowledge base search
knowledge_search - Conversation history
get_conversation_history - Add a lead
create_lead - Update a lead
update_lead - Hand off to a human
escalate_to_human - Send a message
send_message - Web search
web_search
Example tasks
- “Answer questions only with facts from the knowledge base. If you don't know a price, stock level or delivery date, don't guess; say you'll check with the team and hand off.”
- “When a patient asks for an appointment, get their name, phone number and preferred day and time. Say the team will confirm; don't promise an exact slot.”
- “Reply in the language the customer writes in. Keep each message to 3 sentences; for refunds and complaints, apologise and hand off.”
- “When asked about prices, quote the list price from the knowledge base and point to the campaign page; never negotiate discounts.”
Tips for good results
- Knowledge base quality is answer quality: add FAQs as question-answer pairs and returns, shipping and warranty policies as separate short documents; one big PDF matches worse.
- Start with a hand-off threshold of 0.5 for the first week and review hand-offs in the Inbox; lower it if there are too many, raise it if answers are wrong.
- For support bots, consider removing web search from the tool list so the bot relies only on your knowledge base.
- List sensitive topics in the task: refunds, invoices and complaints should go to you without argument.
- AI Chatbot runs on every message, so it writes no reports; for a weekly summary, run Report Writer from a scheduled workflow.
Model advice
Default model
Claude Sonnet 5
Per 1M tokens: $2 input · $10 output
Economy model
Claude Haiku 4.5
Per 1M tokens: $1 input · $5 output
AI Chatbot runs on every message, so the model choice directly sets your monthly cost. For FAQ-heavy support bots with a well-prepared knowledge base, the economy model is enough. Keep the default balanced model for sales conversations, multi-step booking flows or when fluency across several languages matters. Either way, keep max tool steps and the run budget low.
Templates that use this agent
Frequently asked questions
What do I need for a WhatsApp chatbot?
A number connected through the WhatsApp Business Cloud API and a webhook. Because the bot replies to the customer's message, it sends free-form text inside the 24-hour customer service window; messages you start outside that window need an approved template.
What if the bot gives a customer wrong information?
It answers factual questions only from the knowledge base; when the answer is missing it says it will check with the team and lowers its confidence. Review hand-offs in the Inbox and add the missing answers to the knowledge base.
What happens when a customer asks for a human?
That is a hand-off reason regardless of the confidence threshold: the conversation moves to your team in the Inbox, and once you take over the bot stays silent in it.
Does it work on my website too?
Yes. Web chat needs no separate account; the widget you embed brings conversations into the same Inbox.
How many conversations can it answer per month?
Up to your plan's monthly conversation limit. Once it is reached, new conversations are still stored in the Inbox but the bot doesn't reply.
Get started with OrqLabs
Build your first workflow today
Start on the free plan: connect your accounts, pick a template and stay in control with approval steps. No credit card required.