Kaleo — WhatsApp AI Assistant

What it is

I message it on WhatsApp exactly like texting a friend, in plain language or a voice note, and it acts: it reads my email, calendar, and reminders to brief me and chase what matters, runs skills, hits tools and APIs, executes multi-step agentic jobs, then replies in the same thread. It is multi-channel, answering on WhatsApp and Discord off the same brain, with persistent memory so it carries context between sessions. Voice notes are transcribed and routed to the model, with a daily spend cap as a guard. Everything runs on my own infrastructure, not a cloud SaaS, so I control the data, the numbers, and the model. A hard token gate and a file-based kill switch mean only authenticated commands ever execute and I can stop it sending within seconds.

Highlights

• ◆Reads my email, calendar, and reminders to proactively brief me, surface what needs a reply, and keep me on schedule.
• ◆Bridge architecture: a Go whatsmeow/Baileys bridge holds the live WhatsApp Web session and exposes a local REST API; a Python daemon polls it, transcribes voice, routes to the model, and sends the reply.
• ◆Fully self-hosted on my own boxes (Hostinger VPS + Mac/Linux server over Tailscale), with no third-party assistant cloud in the path.
• ◆Multi-number, multi-channel: a Singapore number wired to the Hostinger instance and a US WhatsApp Business number to the Mac, plus a Discord bot off the same stack.
• ◆Survives reboots: runs as a systemd user service with linger enabled, and the WhatsApp session migrates between hosts without re-scanning the QR.
• ◆Safety rails: a fail-closed token gate, a touch-file kill switch that halts sending within seconds, and a daily transcription budget cap.