8 months building MONO — 5 decisions that changed everything

1. Don't build an app

The first instinct is to build an app. It has nice screens, push notifications, a place for your logo. And almost nobody opens it after day three.

The question that changed the product: where do people already live?

The answer was WhatsApp. Not because it's elegant, but because friction is zero. No download, no signup, no new UI. You just write.

Going where people already are tripled conversion vs the web prototype. People don't want another app. They want things to happen.

2. Memory first, intelligence later

Every assistant on the market competes on "how smart is the answer". But people don't remember how clever the response was. They remember whether next time it already knew their context.

An assistant that forgets each conversation is a stranger every time. An assistant that remembers your kid is named Diego, that you hate coffee, that your main client is Ana — that one becomes part of your life.

I spent months on the memory system before polishing reasoning. It was the decision that added the most value and shows the least from outside.

3. One server per person

The entire industry stacks millions of users on shared infrastructure. Cheaper, scales better, compromises privacy.

We took the opposite path: every user gets their own dedicated server. More expensive, more work, completely private.

It's not dogma. It's math: if your data doesn't share a disk with anyone else's, no single failure can expose them all. This type of architecture pays for itself the day a competitor has its first breach.

Also: when your assistant is yours, it's really yours. You can delete everything with one button and nothing remains.

4. Proactive > reactive

An assistant that waits for questions is infinitely cheaper to build. And infinitely less useful.

The assistant that messages you first — "you spent 60% more on Uber this week", "Ana hasn't replied, follow-up?", "9am meeting tomorrow and it'll rain" — that one feels like having someone looking out for you.

Building proactivity takes more work: rules for when to warn, filters to not spam, smart timing. But it's the difference between a tool and a companion.

5. Execute, don't answer

The chatbot temptation is eternal: "here's the email template to request your appointment". Wrong.

People don't want explanations of how to do things. They want things done. Every time the assistant hands back a template instead of executing, it recreates the problem it's supposed to solve.

The effort of integrating with 20+ services, handling phone calls, managing OAuth, generating invoices, charging via Stripe — all of that looks "extra". It isn't. It's the product. Without it, it's just ChatGPT with branding.

What's next

Still lots missing. Real-time voice. Shared memory between family members. Intent-based integrations, not OAuth-based. An assistant that understands local slang.

But the foundation is clear: not an app, not a chatbot, not ChatGPT with memory. Something new: an operational companion that lives where you live and does what you ask.