The original idea for Spillbox.app was supposed to be completely different. Actually, to tell the full story, we have to talk about writing code.
Like a lot of developers in recent years, when we first started using AI coding assistants—tools like Copilot, Cursor, and Windsurf—it felt like magic. The way these tools could anticipate what you wanted to type and auto-complete entire blocks of logic was genuinely mind-blowing.
At the same time, we had a very personal, very human problem: keeping track of our own thoughts.
For a long time, the fastest way to capture a thought was simply texting it to a private Telegram chat. It was quick, but it lacked context. You'd leave a raw, three-word note, and a week later, stare at it with no idea what it meant. There was no way to connect related thoughts or find past ideas.
So, naturally, we had a "brilliant" tech-bro idea: What if we built an AI-assisted note-taking app that acts like Copilot, but for your personal thoughts?
The Autocomplete Experiment
In June 2025, we set out to build exactly that. The goal was an intelligent text box that would look at what you were writing, scan your past notes for context, and auto-complete your sentences to give those raw thoughts more depth.
We ran into two major roadblocks.
The first was technical. Running accurate AI models on-device (to maintain non-bargainable privacy aspects) is heavy.
But the second roadblock was much deeper and far more fatal: Note-taking is fundamentally different from coding.
Code is built on logic, patterns, and predictable syntax. Personal thoughts are messy, nuanced, and deeply human. We realized that even when the AI had enough context about our past notes, its suggestions often missed the mark. Because, frankly, who I am today is not inherently who I was yesterday.
Instead of helping, the AI autocomplete became a massive source of friction. It disrupted the flow of thought. If you think about it, even basic grammar and spell-check suggestions can sometimes pull you out of the zone when you just want to dump a raw idea. Having an AI actively trying to finish your very personal sentences was maddening, to say the least.
The Privacy Paradox
Beyond the friction, we hit a philosophical wall: we realized that AI and private notes are fundamentally incompatible.
A core requirement of capturing personal thoughts is absolute, uncompromised trust. Your notes are often messy, vulnerable, and unfiltered. When you introduce heavy AI into that environment, you inherently introduce doubt. Are my thoughts being used as training data? Who is processing my 2 AM brain dumps?
To give an AI enough context to actually be useful, you have to surrender your most private data to a server. Philosophically, that is a privacy trade-off we simply weren't willing to make. Your notes should belong to you, and stay on your device, full stop.
Stripping It Back to the Studs
We tried pivoting to a lighter, on-device AI feature: semantic tag suggestions. But even then, after using the app for a few months, we realized we kept it disabled.
The core value we were getting out of our own app wasn't the AI at all. It was the speed. It was the ease of capturing raw thoughts in a chronological stream without overthinking. The AI features were just shiny distractions we stubbornly thought we needed.
So, we made a drastic decision: we dumped the AI.
Fixing the Real Problem
With the AI gone, we rewrote the app from scratch. Instead of chasing the latest tech trends, we looked closely at our original hack: texting ourselves on Telegram and WhatsApp.
Why do so many of us do that? Because the chronological stream interface is simply the fastest, most frictionless way to dump a raw thought.
We realized our job wasn't to augment your notes with AI; it was to build a tool that actually supported that raw, stream-of-consciousness workflow. We took the speed of a messaging app and added the missing primitives you need for genuine note-taking:
- Follow-ups: We repurposed the "reply" mechanic from messaging apps to act as follow-ups. This lets you build a thread showing how a thought evolved over time, without ever having to edit and destroy the original raw capture.
- Hashtags and Filters: To fix the chaotic, unsearchable nature of a private chat log, we built in simple inline hashtagging that can be used to add context without much friction. You organically tag a note on the fly, and can instantly filter the stream later to resurface related concepts.
Instead of wrestling with heavy ML models, we focused entirely on making these core mechanics as fast and robust as possible.
Will We Ever Lean Into AI Again?
We might. There are absolutely valid use cases for ambient AI—tools that help you find insights or resurface forgotten connections after the fact. But that would be a completely different product, designed for a different use case. Spillbox.app will always be the frictionless, private, chronological stream of your raw thoughts.
When you open Spillbox.app today, it’s just you and a blinking cursor. No backseat driver trying to guess your next thought. Just a quiet, frictionless place to capture the messy reality of being human.