Build WhatsApp AI agents with the official API

Use Kapso for the WhatsApp infrastructure around your agent: official API access, message triggers, Workflows, webhooks, MCP, and human handoff.

Start building Read Build with AI
Official Meta Business Partner

What is a WhatsApp AI agent?

A WhatsApp AI agent reads a customer message, uses business context and tools, then replies, waits, or hands off. In Kapso, it can run inside Workflows or connect from your own app over APIs and webhooks.

It speaks through WhatsApp

Kapso receives messages and sends replies through the official WhatsApp API, without rebuilding Meta plumbing.

It acts through tools

Run Workflows, functions, webhooks, MCP tools, and external services from the conversation.

It hands off when needed

Move complaints, sensitive cases, low-confidence replies, and account-specific work to a human with context.

How the agent loop works

From inbound message to reply or takeover.

  1. 1

    WhatsApp message

    A customer sends a message to your official WhatsApp number.

  2. 2

    Kapso webhook

    Kapso normalizes the event and delivers it to your app, workflow, or agent.

  3. 3

    Agent decision

    Your AI agent reads context, calls tools, and decides the next action.

  4. 4

    Human handoff

    When judgement is needed, route the conversation to a person with context.

The infrastructure around the agent

Models are easy to call. Production WhatsApp needs events, logs, tools, handoff, and humans around the model.

Inbound replies and statuses

Receive customer messages, delivery updates, failures, and read status events through webhooks.

Official API surface

Send text, media, templates, reactions, locations, buttons, lists, and WhatsApp Flows.

Kapso WhatsApp MCP and CLI

Use Kapso WhatsApp MCP for agent tools and CLI for terminal-native automation.

Inbox for takeover

Keep ownership, history, and context available when a human takes over.

Build agents with Kapso Workflows

Create agents with WhatsApp triggers, Agent nodes, functions, waits, branches, MCP tools, and handoff.

Operational visibility

Follow messages, templates, workflow runs, logs, webhooks, and handoff from one WhatsApp operating layer.

Recommended starting points

Pick the surface that matches your architecture. Keep the first agent narrow and observable.

Build inside Workflows

Use WhatsApp triggers, Agent nodes, functions, conditions, wait steps, and handoff when Kapso should own orchestration.

Workflow triggers

Connect external agents

Use the WhatsApp API and webhooks when your agent runtime already lives in your app.

API reference

Use MCP and CLI while developing

Use Kapso WhatsApp MCP for live WhatsApp operations and CLI when shell access is the right interface.

Kapso WhatsApp MCP docs
Start building Read Build with AI

Human handoff is part of the product

Agents should not trap customers. Kapso keeps the thread, owner, and operational context available when automation pauses.

  • Customer asks for a human
  • The agent is uncertain
  • Approval or sensitive data is involved
  • A complaint or emotional message arrives
  • A VIP or high-value lead needs ownership

Safe by default

Limit autonomy, ground answers, and make every operation reviewable.

WhatsApp policy built in

Model opt-in, templates, and the 24-hour service window before your agent starts sending.

Approved knowledge

Ground replies in docs, support articles, CRM records, and product data.

Reviewable operations

Track messages, webhooks, workflow runs, handoff, and retries.

Questions builders ask

The decisions teams make before putting an AI agent on WhatsApp.

Is a WhatsApp AI agent the same as a chatbot?

No. A chatbot usually follows a fixed tree. An agent can use models, tools, memory, and workflow rules while staying connected to WhatsApp policy, templates, and human handoff.

Can the agent send outbound WhatsApp messages?

Yes. Use approved templates when required, respect the 24-hour service window, and route sensitive cases to a human.

Can Kapso support human handoff?

Yes. Kapso provides WhatsApp conversations, webhooks, inbox tooling, workflow hooks, and logs so your app can pause automation and assign an owner.

Can I build a WhatsApp AI agent with Kapso Workflows?

Yes. Kapso Workflows can start from WhatsApp messages, use AI agent nodes, call tools, run functions, branch with conditions, wait for replies, and hand off to humans. Developers can also connect external agents through APIs, webhooks, CLI, and MCP.

Can coding agents use Kapso directly?

Yes. Agents can use Kapso WhatsApp MCP for live WhatsApp operations and the Kapso CLI when they have shell access.

Rules agents must respect

Good WhatsApp agents need product judgement, not only model output.

The 24-hour service window

Design agent replies around free-form messages, templates, and customer-initiated windows.

BSUIDs and usernames

Prepare agents and CRMs for identity fields that may no longer include a phone number.

Human handoff

Route uncertain, sensitive, or high-value conversations to humans with context.

Start with a safe agent loop

Connect WhatsApp, receive replies, call tools, and hand off when a person should take over.

Start free Kapso WhatsApp MCP docs