I wanted it, it didn’t exist, so I started building it myself
Second Brain Series — EP6: The App
I talk to AI every day. Planning, coding, bouncing ideas. But one day I caught myself thinking: “Six months of this — what do I actually have to show for it?”
While building my second brain, things started accumulating bit by bit — but running the wiki surfaced the next problem. There was no tool that let me use it more comfortably, more effectively. So I went looking.
I searched the market
Notion, Obsidian, Mem.ai, Granola. Used them all or looked them over. Each one does something well, but none of them did “automatically organize AI conversations into a personal wiki.” Even the ones with AI bolted on did generic summarization with no sense of my context.
There’s a reason this doesn’t exist. The hard part isn’t summarization. LLMs handle summarization just fine. The hard part is: “when the same topic comes up again in a new conversation, do you update the existing doc or create a new one?” And tracking decision reversals — “we decided on A yesterday, today we flipped to B.”
A detour: OAuth
My first idea was “just connect it to my Claude subscription account via API.” Then I read the terms. Anthropic updated them in February 2026 to explicitly prohibit using Claude subscription OAuth tokens in third-party apps. They were closing the price arbitrage of running API-level workloads on a $20/month Pro plan.
One wall later, the direction changed. BYOK — bring your own API key, users enter it themselves. That pivot ended up shaping the app’s architecture down the line.
Why a three-layer structure emerged
After living with the problem long enough, I had four pain points: the friction that made me give up on organizing things every time, conversations fragmenting all over the place, context from other projects bleeding in and contaminating things, and having to re-explain all the background from scratch at the start of every new chat.
Trying to solve all four naturally converged on three layers: raw conversations (sessions), organized output (documents), and isolated containers to hold them (spaces). One core principle: “automated processing only within the current space.” Pull in everything and you get contamination.
I’m the first user
A solo developer solving their own problem has one extra weapon. I’m the first user, and I know exactly what’s broken. No market research needed — I feel in my gut whether something is right.
I started with a second brain, wanted to use it better, so I built an app. Then I used the second brain to build the app. Using the app made the second brain better. That loop is what slowloop is.
← Prev: EP5 — Usage
Background: EP1·EP2 — Build · EP3 — Integration · EP4 — Discord