Notifications with a Purpose
May 28, 2025 — 6 min readDigital Minimalism
Nowadays our lives are filled with distractions from our modern technologies. It has gotten out of hand and we need to take back control. For me it's not even that extreme. I don't even have an Instagram account and I never had any social media apps on my phone. Still, I find myself easily scrolling mindlessly through LinkedIn on my laptop, just out of habit. Or regularly checking news or messaging apps.
While I was searching for dumb phones, someone recommended the book Digital Minimalism. Even though I'm not an extreme case, I still found some interesting thoughts in this book.
Digital minimalism is not necessarily about reducing the use of technology to a minimum. Instead, it is about thinking carefully about the role that each of these technologies should have in our lives. It is about making sure that the technologies that we choose, and the way that we use them, are the best possible way to achieve a certain goal.
The technological developments with smartphones and apps happened so quickly that nobody stopped to ask why we need to be notified immediately when a new podcast is available. Or if it's really a good idea to order by QR in a restaurant... I think we are too often only thinking about the short-term benefit and don't consider the long-term downsides enough.
I can also resonate with the princple "Intentionality is satisfying". By being intentional about the use of technologies, it becomes more satisfying. Instead of reacting to a technology that tries to grab your attention, you make the concious decision to engage with it. I can really feel the difference when I sit down and choose that this is the moment to check my WhatsApp messages from the past couple of hours.
The network effect
One of the biggest challenges in pursuing digital minimalism is the widespread use of technology by everyone around you. If you're the odd one who's not on Instagram or WhatsApp, you might be lucky if they call you. But it's more likely that you're gonna miss out on things.
But actually WhatsApp is a tool that makes it easier than ever before to meet up with people in real life. Real human interaction is certainly more valuable than missing out. For this reason, many digital minimalists decide that this technology deserves a place in their lives. However this is kind of a trade-off, because you might have to check regularly in order not to miss out. And this is where I suddenly got this great idea.
Meet the virtual secretary
Since the idea of digital minimalism is not necessarily to reduce technology, but to make smarter and more concious use of it, I conceived the idea of making notifications smarter. In general, I hate to be disturbed by notifications. But there are two exceptions where I feel that my quality life actually improves by receiving instant notifications:
- When arranging to meet eachother in-person that same day.
- When there is an issue with a server that might affect my clients.
Any other messages can be treated like e-mail, and will be checked 2 or 3 times per day (with intention).
By now, AI has gotten good enough to read incoming messages for me and recognize what's important. It's like a virtual secretary, who makes sure that the boss doesn't get disturbed all the time. It's actually a very simple idea to use AI for this, but I find it very powerful.
How I did it
Creating the virtual secretary was relatively straight-forward. I extract the messages from WhatsApp, forward them to an AI Agent and then send out notifications with a custom notification app.
Whatsapp-web.js is an open-source tool that connects to your WhatsApp account. I just modified the default script to forward all messages via a REST API. The only disadvantage of this tool is that it might break at any time, if WhatsApp decides to push an update. Alternatively the WhatsApp for Business API could be used.
For running the AI Agent, I used n8n. It's a powerful tool to create custom workflows and easily integrates with a ton of services already. To receive the WhatsApp messages, I used the custom webhook block. I do some basic filtering to only parse direct messages.
The configuration of the AI Agent is really flexible as well. At the moment I'm still using the ChatGPT API, but this can easily be swapped out for a self-hosted Ollama with DeepSeek or Llama for example. In the system prompt I specify clear instructions and requirements for a message to be notify-worthy. I also ask the AI to generate a confidence score and an explanation, which can be helpful for adjustments to the prompt if necessary.
For the actual notifications, I set up a self-hosted instance of ntfy.sh. The n8n flow can publish the notifications to ntfy.sh using a simple REST API call. And since I now have this custom solution for notifications, I disabled notifications for WhatsApp completely.
Is this really useful?
Now that everything is set up, the actual experiment starts. So far I already had some successful notifications, allowing me to establish in-person meetups quickly without unnecessary distractions. But there will be some false positives and false negatives. Only time will tell if I can turn this into a reliable solution.
I'm also curious to see how my friends adapt to this virtual secretary. Because it's quite easy now to grab my attention by just saying that you would like to meet me today. If people start abusing this, I might have to disable the service for certain people.
Potential improvements
At the moment it is very much MVP. I just put $5 of credit into ChatGPT, which should keep this running for a while. However I don't really like to send my personal messages to OpenAI. They promise not to use my data, but I'd rather not send my data there in the first place. I am currently collecting the results of ChatGPT in a database, which I want to use later to improve the system prompt and to try out different AI models. DeepSeek or Llama can run on my own server, but since the server has no GPU it requires about 2 minutes to parse each message. That's actually fine, but the feedback while tuning the system prompt is just too slow now.
It's also a really tempting idea to have the AI actually reply to my friends. However, I really hate interacting with bots myself and I definitly don't want to push this onto my friends. Maybe only some very specific opt-in features, for example to get my location with a specific command. But that doesn't really require AI in the first place.