AsistAI: an AI personal assistant inside WhatsApp
An AI agent that lives in WhatsApp and manages calendar, email, tasks, Shopify, Meta Ads and social media from a single chat — by text or voice notes.
What it solves
A CEO/COO spends the day jumping from Gmail to Calendar, from Meta Ads Manager to Shopify, from Shopify to WhatsApp. AsistAI unifies all of it into a single chat: the user talks to the agent — by text or by dictating voice notes from their phone — and from there manages calendar, email, tasks, e-commerce, advertising and social media. No learning each panel separately, no depending on a human assistant for repetitive work, and with memory: the agent remembers each contact’s past conversations.
How it’s built
A five-stage architecture:
- Intake — a webhook receives the message and a switch separates text from audio.
- Audio → text — voice notes are transcribed with Whisper before reaching the agent.
- AI brain — an agent (o3-mini) with eight tools: Tasks, Calendar, Gmail, Shopify, Meta Ads, Content, WhatsApp and Calculator. Postgres memory with a 50-message window per contact.
- Classifier — a second model (GPT-4.1-mini) decides whether the reply works better as text or as audio.
- Delivery — text reply via WaAPI, or audio generated with ElevenLabs and sent back as a voice note.
Estimated fixed cost: ~17 USD/month (WaAPI + ElevenLabs); everything else runs on free tiers or is already included in the VPS.
Why it’s interesting
The relevant part isn’t connecting APIs — that’s the easy bit — but designing the full loop: multimodal input, tools with persistent context, and output adapted to the channel. It’s the “personal assistant” pattern that trendy startups sell, built with open-source and low-cost tools, running inside the app the user already has open all day.
The workflow, in the open
The full workflow is published on GitHub: you can download it, import it into your own n8n and see exactly how it’s built, node by node → Agente_Asistente_Personal_para_CEOs.json