Your smart home knows when you changed the oil last time. Your AI tools don't. That's the paradox at the heart of modern artificial intelligence: tools that can beat us at chess and diagnose diseases, but forget everything the moment a conversation ends.
Nate B Jones has a solution. It's called Open Brain — a personal database that gives every AI model a working memory. And in this piece, he reveals something most AI enthusiasts have missed: the real breakthrough isn't getting agents to store data. It's building a visual layer that humans can actually see.
The Memory Problem
AI agents are exploding in popularity. Over 1.5 million autonomous agents launched in recent weeks alone. But every single one faces the same fundamental limitation: they can't remember you. Each new session starts from zero. Every tool switch wipes the slate clean.
Jones calls this the "keyhole problem." When you chat with Claude, ChatGPT, or OpenAI, you're conversing through a tiny keyhole. The data returns, but it's purely text-based. You can't build a visual app from it — at least not without significant development work.
The solution Jones proposed in an earlier video was Open Brain: a personal database you own, connected to any AI through an MCP server. It can access frontier models and pull data from multiple sources.
But the question that poured in from thousands of people was practical: what do you actually do with a memory system once you have it?
The Human Door
Jones realized that building a database wasn't enough. You needed something he calls "the human door" — a visual interface that lets humans read and write data the same way agents do.
The key insight is this: don't add middleware, sync layers, or external services between you and your data. Instead, start with a structured table in Open Brain. Add a lightweight visual layer on top. Your agent reaches it through MCP. You reach it through a simple web page, mobile bookmark, or eventually an app.
When the agent writes to the table during conversation, the entry appears immediately when you pull up the view on your phone. When you update something on your phone, the change is there the next time your agent reasons about your week. One source of truth. Two interfaces. Each built for what each side does best.
This eliminates sync layers that might break, lag, or lose data. The consistency is architectural — built into the table itself.
Use Case: Household Knowledge
One practical application Jones builds on this system is a household knowledge base. Every family runs on information that lives nowhere: paint colors, kids' shoe sizes, the plumber you used two years ago, Wi-Fi passwords for guest networks.
The answer always exists somewhere — in a text thread, in a receipt drawer, in someone's memory. But you can never find it when you need it.
Jones suggests feeding this information into Open Brain as you encounter it. You're talking to Claude about something else and mention the paint color: "Save this. The living room paint is Benjamin Moore Hail Navy. We got it at Sherman Williams back in October." Two seconds. Done. You don't think about it again.
The agent captures these facts in structured form from ordinary conversation. Over a few weeks, the table fills with institutional knowledge — where the keys live, when you last changed the tires, all those details.
Then build a search interface over the top. A search bar at the top of the page lets you find anything instantly. Categories pop up naturally: living room, car, kids' school. You can see where you've recorded household knowledge, identify gaps, and get that mental labor out of your head into something scalable.
Use Case: Professional Relationships
Another powerful application tracks professional relationships. You might have dozens or hundreds of connections that matter, with very limited ability to maintain them all. Context is scattered across emails and messages, and your memory biases toward the most recent interaction.
You can ask Claude: "Anyone I've been neglecting?" It scans Open Brain. If you've been logging interactions, it might say: "You haven't reached out to James in a few months. Last time you talked, he was worried about his team." Then it surfaces what matters — not just when you last spoke, but what you discussed.
Implementation
The agent side is straightforward if you've already built Open Brain. You use the same Superbase dashboard: create a table, add columns for what you're tracking, and your agent reaches it immediately through MCP.
The visual side takes more effort. You can start with Superbase's built-in table editor — it's not beautiful but it's functional. You see data, you edit it, and your agent sees every change immediately.
For something polished, talk to your favorite AI. Say: "I want a mobile-friendly view of my maintenance table. It should show every appliance in the house, warranty dates, last service date. Highlight anything expiring in the next 30 days." The AI generates a small web application. Then use Versel — a free hosting service with paid plans that won't hit limits for this type of project. Upload what your AI built, and you get a live URL you can open on any phone or tablet.
Bookmark it on your home screen. It behaves just like an app without an app store or subscription.
Counterpoints
A reasonable concern is complexity. Building these visual interfaces requires some technical comfort — creating tables, connecting to hosting services, generating code through AI assistants. For non-developers, that afternoon investment might feel steep, especially when the payoff is organizational convenience rather than core functionality.
Also worth noting: Superbase's built-in table editor isn't beautiful. It's functional but bare-bones. Jones acknowledges this and suggests moving toward a polished version eventually.
Pull Quote
The table remains the source of truth. Both you and your agent have direct access to it, just through different doorways.
This is the philosophy that makes Open Brain different from every app you've tried — no sync layer, no middleware sitting between you and your data.
Bottom Line
Jones's strongest argument is architectural: one source of truth, two interfaces, each built for what each side does best. The visual layer isn't just a nice-to-have — it's necessary because humans are visual creatures who scan, sort, and browse differently than agents do.
The vulnerability is practical rather than theoretical. Building these custom views takes time and some technical skill. But if you're already running Open Brain and have an afternoon to spare, the use cases Jones outlines here — household knowledge bases, professional relationship tracking, maintenance dashboards — suggest this system might finally give AI tools something they've always lacked: memory that actually works for both sides.