I Almost Quit. Here's Why I Didn't.
A few weeks ago I spent some real time thinking about whether I should keep building Brains. Not the motivational kind of thinking. The uncomfortable kind.
By Brains Team
A few weeks ago I spent some real time thinking about whether I should keep building Brains.
Not the motivational kind of thinking. The uncomfortable kind, where you actually list out every reason this might be a bad idea and sit with it.
I've been building Brains, a tool that gives your AI a persistent memory for your projects. The idea is simple: every time you open a new chat with Claude or ChatGPT, it starts with zero knowledge of you, your work, your decisions, your context. Brains fixes that. Your project lives in a structured knowledge base that any AI can connect to. You stop re-explaining yourself. Your AI actually knows what's going on.
That's the pitch. Here's what I had to reckon with.
The case against
Google has NotebookLM, and as of early 2026 it's baked directly into the Gemini app. Every Google Workspace user is one click away from it. OpenAI already supports full MCP read and write in ChatGPT Business. They're already doing the cross-tool knowledge layer, they just don't call it a brain. Anthropic's Claude Projects auto-saves your workflow habits and architecture decisions now.
These are not startups. These are trillion-dollar companies building in the exact same direction.
I have zero paying customers. Zero public launch. The product exists because I needed it.
When you lay it out like that, it looks pretty bad.
The thing I kept coming back to
None of them are building the layer that lives outside the AI.
OpenAI's MCP is great inside ChatGPT. Claude Projects are great inside Claude. But what happens when you want to switch? What happens when a better model drops and you want to use it? What happens when you've spent months building up context inside one platform and they change the terms, change the pricing, or just change?
You lose it.
Brains is different in one specific way: it doesn't belong to any AI. Your knowledge base is plain markdown files. You can export it any time. It connects to Claude, ChatGPT, or anything else that speaks MCP. The AI is the tool. Brains is the memory. Those are two different things, and right now almost nobody is treating them that way.
I'm not saying that because it sounds good. I'm saying it because it's how the architecture works. Your data isn't locked in. That's a feature, not a promise.
What actually kept me going
I run a YouTube channel called OIO Racing. We build and race old Japanese cars and other questionable machinery... autocross, rallycross, anything we can get into trouble with. Right now I've got a 1985 MR2 in the middle of a motor rebuild targeting an event in late May, a '72 Celica out for body work and an 18R-G engine build, a '77 Corolla that's been in a six-year resurrection, a '62 Renault Dauphine that won't start... and the list goes on and on and on.
Every one of those is an ongoing project with its own history, its own parts list, its own decisions made and forgotten, its own rabbit holes. And I use AI constantly across all of them. Specs, rebuild sequences, sourcing questions, diagnostic logic, wiring diagrams... it touches everything.
The problem is that every single chat starts from scratch. I'm re-explaining which car I'm talking about. Re-establishing what we already decided on the cam timing. Re-pasting the compression numbers we took three weeks ago.
Brains fixed that. Claude knows which cars are in the fleet, what state each build is in, what parts are on order, what we tried that didn't work. I open a new chat and we're continuing the conversation, not starting it over.
That's real. It's not a demo or a use case I invented for a pitch. It's how I actually work. And if it's this useful for someone juggling a fleet of project cars and a racing schedule, I have a hard time believing I'm the only one.
The honest condition
I'm not just deciding to keep going and calling it a day. There's a kill condition.
If I can't get one real external user... not a friend, a stranger who actually cares about the problem... within 60 days of a public launch, the market has answered the question.
I think that's fair.
Until then: I'm shipping the free tier in the next couple of weeks. 100 pages, no credit card, connect it to Claude or ChatGPT and try it. If it solves your problem the way it solved mine, I'd genuinely like to know.
If it doesn't, I'd like to know that too.
Brains is at usebrains.app. Free tier is coming. If you're the kind of person who hates re-explaining your projects to an AI that forgot everything since last Tuesday... this is for you.