A WhatsApp chatbot that handles real conversations, not just the happy path.

A WhatsApp AI chatbot, defined.

A WhatsApp AI chatbot is a software agent that connects to the official WhatsApp Business API, listens to inbound messages on your business number, and replies using a language model grounded in your business content. It can answer questions, complete structured tasks (book, quote, qualify, refund), call external systems through APIs, and escalate to a human when it is not confident. A real WhatsApp AI chatbot has four things at the same time: live access to your knowledge (FAQs, catalog, policies), the ability to take action in your systems (Shopify, CRM, calendar, payments), memory across the full conversation, and a clean handoff to your team when the AI hits its limits. Take any of these out and the bot becomes a frustration generator.

Five things happen on every inbound message.

This loop runs on every single message, including the messages that arrive while a structured flow is mid-execution. That is what 'AI over the whole conversation' actually means.

• Intent detection — the AI classifies the message against the flows and intents you've configured, and decides whether to follow a flow or stay in free conversation.
• Knowledge retrieval — the model pulls relevant snippets from your Company Brain (website, PDFs, catalog) and uses them as grounded context, so it never invents pricing or policy.
• Action execution — when the customer needs something done (check order, book slot, raise refund) the AI calls the matching tool with typed arguments.
• Reply generation — the response is composed in the right tone, with WhatsApp-native formatting and CTA buttons where helpful.
• Confidence check — if the AI's confidence drops below threshold, the conversation is routed to a human with the full transcript and an AI summary already loaded.

Flows are guardrails, not the boss of the bot.

Structured WhatsApp flows are excellent for the moments that must be deterministic — collecting consent, capturing an address, running a payment, asking the seven questions every lead needs to answer. They are bad at handling the messy middle of a real conversation: clarifying a typo, answering an out-of-band question, switching language mid-sentence, recovering from a misclick. Waaru runs the AI as the orchestrator. Flows act as guardrails the AI can step into when a structured outcome is needed. The moment a customer steps off the flow, the AI handles the detour and steps the flow back to the right node. The user never sees a mode switch. The flow never dies on an unexpected reply.

The chatbot uses your business as its source of truth.

Every Waaru workspace has a Company Brain. You point it at a URL, upload PDFs, paste FAQs, or sync your Shopify catalog. The chatbot answers from that content only — it does not improvise pricing, policy, or stock. If a question falls outside the indexed content, the bot says so plainly and escalates instead of inventing. The Brain re-indexes on a schedule, so when you update a product page or a refund policy on your website, the chatbot updates with it. You do not run a training job.

• Website URLs (crawled and chunked)
• PDF documents — rate cards, menus, policies, handbooks
• Plain-text FAQs and macros
• Shopify product catalog and order data
• Custom data via REST or webhook

Actions, not just answers.

Answering questions is the floor. A real AI chatbot completes tasks. Waaru ships a tool layer that lets the AI call external systems with typed arguments. The customer asks 'where is my order'; the AI calls the Shopify order lookup, formats the reply, and offers next steps. The customer says 'reschedule my booking to Friday'; the AI calls the calendar tool, books the slot, confirms the new time, and updates your CRM. Every action is logged, every argument is validated against a schema, and every tool call is rate-limited per workspace. You can ship a Shopify, Razorpay, Calendly, Cal.com, Google Calendar, or custom-API tool inside the same conversation surface.

Human handoff with the full picture, not a cold transfer.

When the AI hits a question outside its knowledge, hits a policy boundary you've set, or hits a low-confidence reply, the conversation is routed into your team inbox. The agent sees the full message history labelled by speaker, an AI-generated summary of the situation, the contact's CRM record, and any internal notes from a previous session. They reply, the customer continues, and the AI watches — once the human resolves the conversation, the AI picks up the next round without losing context. On the Scale plan, the resolution itself becomes context. When the human answers a question the AI couldn't, that answer is indexed and the next time the same intent appears, the AI handles it alone.

Built on Meta's official Business Cloud API, not unofficial workarounds.

Waaru is built directly on Meta's WhatsApp Business Cloud API. Every message is delivered through Meta's infrastructure, every template is approved by Meta, every business number carries a verified WhatsApp Business Account. There is no QR code shortcut, no third-party gateway, no risk of your number being banned for unofficial automation. This matters because the unofficial route is what gets numbers killed. Tools that promise 'no API needed' are routing through unofficial WhatsApp Web automation that Meta actively detects and shuts down. Waaru is on the official rails from day one.