The Idea
The Build
The Lesson
The Proof
The Reach
Case Study · Personal Project · HS-CS-001

Open Brain

When you're not holding it all in your head, there's more room to create.

Domain
Agentic Systems
Year
2025 — 2026
Status
Published
Type
Personal Infrastructure
FIG. 01 — The Idea

More than I could hold in my head.

I always have more going at once than I can hold in my head — several projects, each with its own live thread: what was decided, what's half-built, what I meant to circle back to. Keeping all of it loaded is its own kind of work, and it's the kind that crowds out the part I actually care about — making the next thing.

The cost showed up every time I came back to something. Reload the context, rebuild the picture, find my place — then, finally, five minutes of real work. The ideas were never the problem. Holding onto them was.

FIG. 02 — The Build · Ingest pipeline. Memory layer. Return arc.

I built it like infrastructure, not a feature.

The fix was to stop holding it in my head and give the remembering a home outside it. Not memory bolted onto a session — that version forgets the moment the session ends. Memory is an asset, and like any asset it has to be maintained, not just accumulated. So the system runs three loops, not two — capture, retrieval, and maintenance. The third is the one most memory systems skip, and the one I didn't realize how much I'd lean on.

The principle followed from the reframe: 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. Built on the shoulders of Nathan Jones' Open Brain.

trigger · skill autonomous tasks tool calls capture · query write summaries read notes context retrieval CW-01 Cowork interface · input OC-01 OpenClaw agent identity · autonomous AG-01 Claude Agent orchestration · skills MC-01 MCP Layer connectors · protocol OB-01 Open Brain memory · thought capture HV-01 had9000 vault daily notes · apple notes · summaries FIG. 02 · System Architecture · HS-CS-001
Active data flow
Context retrieval (return arc)
Data packet in transit
Secondary / async packet
Stack · Tools & Platforms
FIG. 03 — The Structure · Three loops, one bank.

Three loops, one bank.

Capture flows in on the outer ring — every day, ambient. Retrieval flows back on the middle ring — every conversation. Maintenance runs on the innermost — slower, deliberate, the loop that keeps the bank true. It's the part of the system you don't see running, and the part you stop noticing you rely on.

FIG. 03 — Three loops around OB-B6 organic cluster kernel Three concentric elliptical rings around the OB-B6 organic cluster kernel at 55° axial tilt. Saturn-style depth layering — front halves of rings cross in front of the kernel; back halves pass behind. Inner maintenance ring is hatched as the load-bearing element. Particles orbit at different cadences per loop with back-leg motion slowed an additional 15%. FIG. 03 three loops · OB-B6 kernel OPEN BRAIN OB-B6 · the bank CAP-01 CAPTURE daily cadence RET-01 RETRIEVAL every conversation MNT-01 MAINTENANCE innermost · deliberate, periodic DAILY NOTES Obsidian · *.md DROP ZONE vault/+/ unsorted SESSION DISTILL /remember · key thoughts REFINEMENTS edit · re-distill COWORK interface · input OPENCLAW autonomous agent BRIEFING SKILL morning context · <30s AUTO SAVE catches what I'd lose HEALTH CHECK prunes · merges · flags FIG. 03 · Three Loops · HS-CS-001 v1.0.0

The diagram puts the memory store at the center on purpose — everything else orbits it. The inner ring is drawn heaviest because it's the one that matters most: the quiet maintenance work that keeps the memory accurate, even though you never see it run.

FIG. 04 — The Lesson

I made it prove itself, every session.

For a stretch of an important build, the system looked like it was working — and was saving nothing. By the time I caught it, hours of good work were gone, and the frustration was real. A piece of it had quietly stopped firing, and because nothing errored, nothing told me.

So I built a health check. Every session now tests the whole pipeline end to end — capture, storage, retrieval — and if any part is broken, it says so loudly before I do a minute of work. The fix wasn't more storage. It was making the system prove itself, every time — because a memory you can't trust is worse than no memory at all.

FIG. 05 — The Proof

Seven projects in six weeks.

With the remembering handled, I could keep more in motion than I could hold in my head. In six weeks the record shows real work across seven projects — a geology knowledge graph, a branding persona tool, the memory system itself, this site, a design system, a family photo archive, and an active job search.

7
Projects
Real work advanced across all seven in six weeks.
41/45
Days active
Showed up nine of every ten days.
26
Multi-project days
Moved two or more projects forward in a single day.
424
Notes kept
Context saved and returned, not lost between sessions.
ProjectWhat it isActive days
Coherence Engine / In the RoomBranding persona tool29
OpenBrainThis memory system16
Job huntSearch + applications15
Exploration GraphGeology knowledge graph12
Portfolio siteThis site7
Family historyFamily photo archive4
Design systemBrand + glyphs3

The count isn't the point. With the remembering handled and my ideas out of my head, I could drop straight into the work and stay there — no rebuilding where I left off. That's the real payoff: flow.

FIG. 06 — The Reach

The work follows me.

The same brain is there from my laptop, my desktop, and my phone. An idea I catch at night is waiting on whatever machine I sit down at in the morning.

Exploration Graph runs on exactly that. I build it at my desk and query it from an iPad in the field, off the same memory — and new projects plug into the same place. The work doesn't care which device I'm on.

Context that comes back to you is what turns a pile of ideas into a body of work.