I put an AI agent behind my WhatsApp that roasts me back
This is the project that got me hooked on agent infra: OpenWA, an open-source WhatsApp API gateway that let me put an AI agent behind my actual WhatsApp.
The whole idea started as a joke - picking what to post was harder than building the thing. So I built the thing.
Reverse-engineered WhatsApp Web
The project begins by reverse-engineering WhatsApp Web and turning it into a set of clean APIs. You get a Swagger-style spec, so every action that WhatsApp Web can do becomes a callable endpoint - send a message, media, create and leave groups, the works.
Webhooks for any event
The fun part is the webhook layer. You attach event handlers for pretty much anything: messages sent, messages received, photos, status changes. That event stream is the door to an agent.
The 3 things I built
- An AI agent behind my WhatsApp - it reads incoming messages and composes replies through a system prompt.
- Every message to Postgres - nothing is deleted. Every conversation is logged so I can learn from my own chat style.
- Fine-tune an LLM to sound like me - because you have the transcript, you can steer a model toward your voice.
The result single was a system prompt that when the punchline hits, replies with a roast. "Hey love", type, send - and back comes the roast.
The whole setup runs through a simple flow: a new session on openwa.littlebitpositive.com, scan the QR, register a webhook, and your agent is live.
Where to find it
- App: openwa.littlebitpositive.com
- GitHub: github.com/rmyndharis/OpenWA
If you want it set up on your own WhatsApp, comment on the video or DM me.
Reels like this, every few days.
Follow for short bites on AI, tools, and everything I am shipping.
Follow on Instagram