作者开源 Wappy：一分钟在 WhatsApp 上搭建自己的 AI 智能体

Every company seems to want to build a personal AI agent. Muse, Instinct, Dots. The demos are impressive. But I keep asking a different question: why should I need someone else's agent to answer a question about my calendar?

I wanted to text “What’s on my calendar today?” on WhatsApp and get an answer from my own setup. No new app. No hosted chatbot service holding the memory of my conversations.

That idea stayed on my list because building it meant doing everything around the agent first: WhatsApp webhooks, message handling, memory, model calls, tools. A bot that says hello is easy. One that can look something up, remember what you meant, and reply usefully is another project entirely.

So I built Wappy and made it free and open source. It is not another finished personal assistant. It is a TypeScript starting point for building the one you want. Bring your number, choose your model, keep the agent and its default memory on your machine, and connect your own tools. There is no Wappy account or Wappy-hosted backend.

This gets interesting beyond my calendar. Imagine a restaurant number that checks its real reservations before saying a table is free. Or a store that answers “Where's my order?” by checking the actual order system, not repeating a canned message. Those are possibilities you can build, not integrations already in the box.

Why now? Models can finally use tools and carry a conversation. WhatsApp is already where people talk. The opportunity is not just to make a smarter chatbot. It is to let a useful agent meet people there, with access to the systems that know the answer.

I included one small example: an optional, read-only Gmail and Calendar assistant. Connect your own Google account, then ask in plain language. Or skip it and build something else.

To start, you need Node.js 22 or newer:

npm create @wappy_ai/agent

The generator asks a few questions and scaffolds the agent in about a minute. A live WhatsApp connection takes longer. The generated guide walks you through setting up your own Meta Cloud API app and webhook.

One important boundary: WhatsApp messages still pass through Meta. If you choose a hosted model, it receives your prompts. Your agent can run locally, but WhatsApp itself is not offline.

This is early. If you try it, tell me what was hard. And if you could connect your own data to an agent on WhatsApp, what would you build?

GitHub: github.com/csr1010/wappy-kit