My army of silent AI agents — When automation orchestrates my daily life

It always starts with an invisible friction, that kind of small routine that drains energy without creating value.

Opening the weather app to mentally compare different days of the week. Combing through Notion every morning to sort out what's urgent, what's overdue, and what can wait. Pulling out your phone while cooking, hands covered in flour, just to add butter to the shopping list.

The data was there, my tools (Notion, Todoist, Home Assistant) worked well, but they all waited for me to make the effort to query them. The real problem wasn't a lack of tools, it was a lack of orchestration. I needed a personal assistant, but one that adapts to my workflows, not the other way around.

To build this system, I chose n8n.

Why this choice? Because it's a node-based automation solution that can be self-hosted (perfect for my homelab), and natively handles advanced AI concepts like memory, Tools, and LangChain agents.

Instead of coding isolated Python scripts running in obscure CRON jobs, n8n gives me a visual canvas. I can create specialized agents, give them access to specific APIs, and have them communicate with me via the platform I use most: Telegram.

Here is how I structured my three main agents.

I'm based in the Paris region, and I try to fit in a late-morning bike ride once a week. But the weather is unpredictable.

So I created a workflow that triggers automatically every Monday morning at 8:45 AM.

The agent retrieves a 5-day weather forecast via the OpenWeatherMap API.

I gave it a strict instruction: only compare Tuesday and Wednesday of the current week, specifically between 10:30 AM and 12:00 PM.

The model (GPT-4o-mini) analyzes the data and writes a concise report telling me which day is best suited for my ride, before sending it to me on Telegram and updating my calendar.

No more analyzing rain radars: by Monday morning, I already know when I'm riding.

I manage my to-do list on Notion. But opening a Kanban board first thing in the morning isn't the best way to get motivated.

I set up a "Daily Brief" that triggers Monday through Friday at 8:50 AM.

The system queries the Notion database to retrieve all unfinished tasks that are either overdue or scheduled for today.

This raw data is aggregated and sent to the AI.

The agent's job is to write a simple morning briefing, grouping overdue tasks on one side and today's tasks on the other.

As a final touch, I asked it to add a personalized piece of advice to motivate me and boost my productivity.

I receive a clear, motivating Telegram message. Zero hallucinations, just what I need to do, presented in the best possible way.

This is where the architecture gets really interesting. I created a "SuperAgent" that acts as a global orchestrator for my home.

I can interact with it via Telegram, using either text or voice messages.

If it receives an audio message, the workflow uses the OpenAI (Whisper) API to transcribe it into text.

The SuperAgent never acts directly: it analyzes my request and delegates it to the appropriate sub-agent.

For example, it uses a "TodoistAgent" specialized in interacting with the Todoist API.

This sub-agent manages my shopping list (adding, deleting, reading items).

It's actually the perfect extension to my Anti-Gaspi project. API integration via a new sub-agent is already planned. If I say "Add milk to the shopping list" or "Remind me to eat the chicken tomorrow", the SuperAgent understands the intent, handles the dates, and routes the request to the right sub-agent.