00 / OKF v0.2 · NOTHING TO INSTALL, RUN, OR MAINTAIN
Mindbase is your own personal knowledge graph. Capture what you learn, at work or anywhere else, and let the links between concepts surface things you'd never spot by hand. It's plain markdown conforming to an open standard — OKF v0.2, not a proprietary format — and that's exactly what lets it scale: new folders emerge organically as your notes outgrow today's structure, and a health-check skill keeps the graph tidy instead of you maintaining a schema by hand.
Clone it to use with any coding agent, or, on Claude Code, install it once as a plugin instead — no cloning needed.
01 / The problem
It's buried in a Slack thread, a Teams message, an email you'll never search for again, one of a dozen OneNote or Notepad files, a wiki page that's gone stale — and a dozen other places besides.
When something does get written down, it's usually one flat note, disconnected from everything else it relates to — so when you actually need it, you're piecing it together from memory instead of just looking it up.
When was the last time you lost ten minutes just trying to remember where you wrote something down? That's not a small tax — it's a context switch that pulls you clean out of flow state, every time.
We've drawn knowledge as branching trees since the 3rd century AD (the Porphyrian Tree) — Mindbase just makes that instinct practical.
Chats, emails, tabs, and note apps all hold a fragment each — nothing connects them.
The information usually still exists somewhere. Finding it means re-reading old threads, guessing search terms, or asking whoever remembers better than you do.
Even the notes you do write are flat — a doc or a page sitting alone, with no way to see what it's connected to or what's changed since.
02 / The payoff
If you're a developer already living inside an agentic coding tool — Claude Code, Cursor, or
anything similar — Mindbase doesn't ask you to open a second app or learn a new workflow. The same agent
that's already reading and writing your code can capture a decision, a debugging insight, or a gotcha the
moment it happens, and hand it back to you months later when you'd otherwise be re-deriving it from a stale
Slack thread or a git blame. The value isn't the note-taking — it's the compounding: every
capture makes every future query a little sharper.
That's also why Mindbase doesn't let an agent silently auto-capture everything it sees, or sync in from a dozen connected tools that each add their own layer of noise until you stop reading any of them. You stay in the loop — you decide what's actually worth keeping, and the agent writes down exactly that. The graph only holds what you decided mattered, which is exactly why it stays useful instead of becoming more noise to dig through.
It's the same instinct behind kakeibo, the Japanese practice of writing expenses down by hand instead of letting an app auto-track them — the small effort of choosing what to record is what makes you actually notice it.
Capture happens inside the same agent session you're already coding in — no separate app to tab into, no separate login to maintain.
A decision captured once is a decision you never have to re-explain to yourself — or to your agent — six months later.
No vector database to provision, no embedding pipeline to babysit, no service to keep patched and running — a handful of markdown files is something you can actually trust to still work in a year.
03 / Compose your own vaults
Mindbase is a template, not a hosted product — clone it as many times as you want, one per vault, and compose them however suits how you actually think. Installed the Claude Code plugin instead? Same idea — a "vault" is just whichever project folder you have open, so switching vaults means opening a different folder, no cloning required.
Once what matters is captured, an agent can reason across a vault and surface connections a flat search never would — a spending pattern buried across months of notes, a symptom recurring across appointments, the same stock turning up across a dozen scattered research notes. The boundary between vaults is yours to draw, and the cleaner it is, the better this scales: split off a domain once it's earned its own space, so each graph stays sharp instead of one pile straining under everything you've ever captured.
04 / Skills
Every skill lives in skills/ as a plain markdown prompt — no vendor-specific
syntax. Claude Code gets native /brain-* slash commands for free; any other agent can run the
same skill by reading the file.
index.md stubs, and Obsidian graph colors.
/brain-setup I'm a software engineer — set up Mindbase for me
/brain-capture Sprint planning today — team decided to move the notification service off cron polling to an SQS-based event queue after missing three alerts last week
/brain-query path outage-incident deploy-process
/brain-maintain health
05 / Architecture
Any coding agent reaches the canonical skills, which read and write a plain markdown bundle — Obsidian is one optional way to view it as a graph, not a required part of the chain.
Notice what's missing from that diagram: no server, no background
process, no database, no port to open. skills/ is a folder of markdown files an agent reads
directly — nothing to deploy, nothing to keep running, nothing that can crash.
06 / Not memory. Not a platform.
Fair question, and worth being precise about all three. Some agent platforms now ship memory as a built-in feature, letting an agent maintain its own context file across sessions so it doesn't lose the plot on a long task. A growing ecosystem of open-source memory servers does something similar from the outside — different tools, different storage underneath, but almost all scoped to one codebase or one agent's sessions, and built to make that agent's own recall cheaper and faster, not to give you something to read. At the far end, some knowledge tools ship as infrastructure — a service to run, a database to maintain, a reasoning engine to configure. Mindbase sits outside all three, on purpose.
Most agent memory — built-in or bolted on — is scoped to one tool's sessions or one codebase. A Mindbase vault is yours across domains — work, finance, health, anything — plain markdown any agent, or you, can open.
Even memory tools with a graph or index underneath build it to make the agent's own recall faster and cheaper. Mindbase's graph exists so you can browse, query, and learn from it directly — the agent is a contributor, not the only reader.
Agent memory isn't meant to be browsed — you'd need to ask the agent for it back. A Mindbase vault is meant to be opened, edited, and curated directly — the agent contributes to it, it doesn't own it.
There's also a heavier version of this idea — full platforms with their own database, background service, and reasoning layer, built for teams and compliance rather than one person's notebook. Mindbase deliberately stays beneath that, too.
Honestly, the right answer depends on what you're trying to do and how complex it is. If the problem is "my agent forgets mid-task," a purpose-built memory layer is the right tool for that job — use it, that's a good outcome, not a threat. If the problem is "I want a knowledge base I can browse, curate, and learn from myself, across whatever I'm working on," that's a different job — the one Mindbase is built for: an open format, portable across whatever agent you use, readable without one in the loop at all, with a human still deciding what's worth recording instead of an agent capturing everything on your behalf.
07 / Get started
git clone https://github.com/Madhusuthanan-B/Mindbase.git/plugin marketplace add Madhusuthanan-B/Mindbase/plugin install mindbase@mindbase-marketplace/brain-setup, whichever way you installed it.)
Set up Mindbase for me — I'm a software engineer.Remember that I just set up Mindbase as my personal knowledge base, following the OKF spec.knowledge/ in Obsidian (optional) — or any markdown viewer/editor
you already use — and watch the graph grow as you capture more.