Open Brain
Every new conversation with an AI started without memory. This is the system that changed that.
Every session started from scratch.
Every new conversation with an AI was stateless. No memory of what was active, what had been decided, what was turning over in the background. Notes lived in two places — Obsidian and Apple Notes — useful in isolation, invisible to each other. The raw material of daily thinking existed across multiple systems. It just wasn't connected to anything.
Treat personal knowledge as infrastructure.
The design principle: anything worth building for a product team's data pipeline is worth building for yourself. Not a monolith — a system with clear component responsibilities and MCP as the connective tissue. Daily notes flow in. Durable thoughts get extracted and stored. Any agent, any session, context flows back.
One bank. Many assets.
The bank stores more than distilled thoughts. The design system lives alongside it. So does the voice library, the glyph set, the session definitions. Every surface draws from the same source. This site is one output. The interview prep PDF is another. The job tracker is a third. Different audiences, different chrome, same bank.
The system, at a glance
Here's the system. Let me show you four ways into it.
Open Brain is the bank. Four tiers around it. Capture flows in. The bank holds the distilled memory. Definitions — design system, voice library, glyph set — live alongside. Surfaces and artifacts get drawn out as needed. Every layer is reused by the next, which means an improvement to any one piece ripples to every output the same week.
Point: introduce the whole, before zooming.
Before outputs: definitions, stored
Before any output exists, the patterns are defined and stored where every session can find them.
The design system existed before any portfolio page did. The voice library was sketched before any voice-passed paragraph. The patterns get defined once and stored alongside the bank — where every future session, every new surface, can find them. That's why a v1.1 design-system promotion in May rippled through every output the same week: not because each surface was rebuilt, but because all four reach into the same definitions.
Point: the bank holds capital, not just content. Definitions persist across sessions.
Capture, every day
Raw content gets captured continuously and refined over time.
A bank only works if something keeps feeding it. Daily notes flow in. Captures land in the drop zone. Sessions get distilled into durable thoughts. Refinements re-shape what's already there. Every day, raw content moves in — and over time, it gets sharper. The portfolio build itself was one of those refinement passes. The case study you're reading was another.
Point: the bank isn't static. It grows. Capture is part of the system, not a one-time setup.
One bank. Many assets.
One bank. Many assets. The portfolio is one. Tomorrow's interview prep is another.
This site is one output. The job tracker is another — different audience, different chrome, same source. Interview prep is a printable PDF. The JD × experience comparison is a per-role analysis. The company summaries are one-off artifacts drawn out before a single conversation. Some recur, some don't. All of them draw from the same bank, all of them inherit the same definitions. One bank, four surfaces, three artifacts — more on the way.
Point: mostly surfaces, some artifacts, one source. The compounding payoff.
The return arc.
The briefing skill runs every morning. It queries Open Brain and returns a structured summary in under 30 seconds: what was captured yesterday, what's active, what needs attention. New sources, skills, and query patterns plug into the same MCP layer without touching the core.
The return arc — context flowing back from memory into every conversation — is the part most AI systems skip. It's also the part that required the most deliberate design. If I wouldn't build the feedback loop for myself, I have no business building it for anyone else.