Boards
ATLAS
A personal knowledge engine and memory protocol.
Give an AI Every Memory and It Does Worse
I assumed more memory meant better answers. It only got slower and more expensive. load_memory turned out to be about deciding what NOT to return — and, in the end, about an AI knowing itself.
Designing a Knowledge Structure an AI Can Read
ATLAS's data model was never designed up front — it grew by breaking. This is how a flat pile of identical nodes became the Nodes, Tags, and Relations an AI can actually navigate.
Instead of a Note App, I Built a Knowledge System
Once I stopped trying to beat Claude, one question was left — if not the model, what actually stays mine? The answer was knowledge. I looked hard at Obsidian, and ended up building my own.
I Studied LangGraph, Then Decided to Use Claude Better
Studying how to connect multiple AI models led me back to a more fundamental question — what should I actually be building? The answer changed everything about what ATLAS became.
Instead of an English App, I Added a Layer on Top of AI
I wanted English learning to follow naturally from the AI experience I was already using every day — not a separate app to open, but a layer woven into the workflow I already had.
I Started Building a Personal AI App on the Cloud
To move my working environment out of my MacBook, I began putting ATLAS on the cloud. With several projects scattered across separate Cloud Run services and min instances, I started building a unified VM environment where I could keep making and discarding small experiments.
I Wanted to Build My Own OS
ATLAS did not begin as a large knowledge engine. It started from a simple desire to move my working environment out of my MacBook and build a personal OS for using AI in my own way.