The Living Vault
The Living Vault
Applied Knowledge — Intellectual Infrastructure. Part of the Wheel of Learning.
Most people store their knowledge in folders — a hierarchy of containers nested inside other containers. This mirrors the way files were organized on the earliest computers, not the way a mind actually works. A mind works by association: one thought calls another, a concept is illuminated by three others from different domains, a question opens onto a network of partial answers. Folders cannot represent what is inherently relational. They atomize what should be woven.
The personal knowledge base is not a productivity hack. It is the externalization of mind — the discipline of making thought visible, structured, and retrievable so that understanding compounds over time rather than dissipating. Within Harmonism, it belongs to the Wheel of Learning, because learning without structured retention is like pouring water into a vessel with no bottom. And within the Wheel of Learning, it sits under Digital Arts — the pillar concerned with working intelligently with technology as an instrument of Dharma, not as a substitute for consciousness.
Two tools currently represent the highest expression of what a personal knowledge base can be: Obsidian as the knowledge environment, and Cowork — Claude’s desktop agent — as the intelligence layer that operates within it.
Obsidian: Sovereignty Over Your Own Mind
Obsidian is a markdown-based knowledge environment where every note is a plain text file stored locally on your machine. This single design choice — local files, open format, no cloud dependency — ripples across every dimension of the Wheel of Harmony.
Your knowledge belongs to you. It is not held by a company’s servers, subscription model, or terms of service. If Obsidian disappeared tomorrow, every word would remain on your hard drive in a format any text editor can open. In an era of increasing platform fragility and digital enclosure, this is not a minor feature — it is a philosophical stance. Sovereignty applied to the life of the mind.
Obsidian’s bidirectional linking lets you connect ideas as the mind actually works — not in rigid hierarchies but in webs of association. A note on a research project links to three annotated papers, to a lecture where you taught the concept, to a collaborator’s thesis, to a strategic question about where the field is heading. Over time, a graph emerges: a living map of your knowledge in which the density of connections reveals what you actually understand, what you are still integrating, and where the blank spaces are.
The plugin ecosystem lets you adapt the environment without leaving it — calendars, databases, spaced repetition, graph navigation, templates. Because the foundation is plain markdown, no plugin is load-bearing. You can add and remove capabilities without risking knowledge integrity. The tool serves you; you do not serve the tool.
A vault is a folder of files. It can be backed up, version-controlled, encrypted, synced across devices, or carried on a USB drive. It works offline, on a plane, in a power outage, on any machine. For anyone building toward long-term resilience — personal, familial, or civilizational — this matters more than any feature list.
No other tool offers this combination. Notion locks data in a proprietary database. Roam Research depends on cloud infrastructure. Apple Notes is a walled garden. Logseq comes closest to Obsidian’s philosophy but lacks the maturity, plugin ecosystem, and publication pipeline that make Obsidian operationally superior.
How a Vault Is Structured
Structure is not given — it is chosen. And the choice of structure is itself an intellectual act: it reveals your model of what matters, how domains relate, and what you think thinking is for.
The Harmonia vault — the knowledge base from which harmonism.io is built — is organized around the Wheel of Harmony: a philosophical architecture with eight pillars and a fractal 7+1 structure at every level. But the vault of a university researcher would be organized around research projects, teaching, and disciplinary concepts. An entrepreneur’s vault would be organized around ventures, strategy, and people. A physician’s around patients, protocols, and evolving clinical knowledge. The architecture is yours to design — what matters is that it reflects the actual structure of your intellectual life, not the default folder tree of your operating system.
Whatever the domain, the magic is not in the folders — it is in the links between them. A literature note on a recent meta-analysis links to the research project it informs, to the lecture where you teach the concept, to the theoretical framework it challenges. When you open any one of these, the backlinks panel shows every other note that references it. Over time, the densest nodes in your graph reveal your actual intellectual centers of gravity — which may surprise you. The graph does not merely store what you know; it shows you the shape of your own understanding.
Anchor Files
Every serious vault has a small set of files that orient the whole system — the gravity centers you return to constantly. In the Harmonia vault, these include the philosophical foundation document, a glossary of terms, a decision log, and a priority-ordered task list. For a researcher, the equivalents might be a research agenda (current questions ranked by priority), a glossary of key terms in your field, a methodological decision log (why you chose this statistical approach, this journal, this theoretical framing), and a TODO ranked by deadlines. For an entrepreneur: a strategic map of all active ventures, a decision log, a contact graph.
The decision log deserves particular attention. When you choose one approach over another — one theoretical framing, one investment, one architectural direction — you record the reasoning. Six months later, when a reviewer asks why, or when you cannot remember what led you to a conclusion, the record is available. Intellectual life runs on the integrity of such choices. The Harmonia vault has logged over three hundred architectural and doctrinal decisions in this manner, each numbered, dated, and reasoned. It is not a changelog — it is case law.
Cowork: Intelligence That Operates on Your Knowledge
Obsidian gives you the environment. Cowork gives you a collaborator that can read, write, and operate within it.
Cowork is Claude’s desktop agent. Its key capability is that it can be given access to a folder on your computer. When you point Cowork at your Obsidian vault, Claude can read every file in real time — not a snapshot or an upload, but the live state of your notes as they currently exist. It can create new files, edit across multiple documents in a single pass, propagate structural changes throughout the knowledge base, and maintain coherence across a body of work too large for any single human session to hold.
This changes the nature of the collaboration entirely. In a standard Claude chat, you bring information to Claude. With Cowork pointed at the vault, Claude comes to your information. It can read a note you wrote three months ago, cross-reference it with three others, identify a contradiction, draft a new article that builds on all of them, and save the result back into the vault as a properly formatted markdown file — all in a single session.
The vault is not a reference Claude consults — it is the workspace Claude inhabits.
The CLAUDE.md: Making Claude Remember You
Claude has no persistent memory between sessions. Each conversation begins fresh. The CLAUDE.md file is the solution: a markdown document, placed in your workspace folder, that Claude reads automatically at the start of every Cowork session. It transforms Claude from a generic assistant into a collaborator who already knows your world.
The CLAUDE.md is not a prompt — it is an orientation document. A prompt instructs Claude what to do in one interaction. An orientation document tells Claude who you are, what you are building, and how to engage with your world on an ongoing basis. The division is clean: the CLAUDE.md holds the permanent; the conversation holds the transient.
What goes into it depends on your life. A researcher’s CLAUDE.md might contain: identity and institutional context, a ranked list of active research projects with their current phase, writing conventions and citation style, student supervision notes, and references to documented workflows for paper reading, grant writing, and lecture preparation. An entrepreneur’s might contain: identity, active ventures ranked by priority, strategic conventions, tone preferences, and pipeline references. A physician’s might contain: practice context, diagnostic frameworks, treatment protocols, and documentation standards. The Harmonia CLAUDE.md runs to several pages — it encodes the complete philosophical architecture, all naming conventions, deployment pipelines, translation workflows, and cross-references to the decision log and task system.
Crucially, the CLAUDE.md evolves. It is a self-improving document. Every productive session teaches you something about how to work with Claude — a new convention that produces better output, a pipeline that should be formalized, a preference you had not yet articulated. You update the file. The next session starts better. Over months, your CLAUDE.md becomes an increasingly precise encoding of your intellectual working style — and Claude’s output quality rises in direct proportion.
Skills and Pipelines
The CLAUDE.md handles what is always true. But some knowledge is domain-specific and too detailed to include in a general orientation — it would bloat the document and bury the essentials. For this, the system supports skills: dedicated markdown files that Claude reads on demand, each containing specialized instructions for a particular type of work.
Skills live in a .claude/skills/ folder inside the workspace. When a relevant task arises, the CLAUDE.md instructs Claude to read the appropriate skill file before beginning. A researcher might build skills for paper-reading (how to extract claims, assess methodology, produce a structured vault note in your frontmatter format), academic-writing (citation style, section structure, journal requirements, common reviewer objections to pre-empt), lecture-prep (learning objectives, key concepts, links to case library), and grant-writing (funder conventions, institutional boilerplate, budget categories). An entrepreneur might build skills for investor-update, content-publishing, event-planning, and board-meeting-prep.
Pipelines are the repeatable workflows that skills encode. The distinction is precise: a skill is the instruction set; a pipeline is the workflow it governs. The Harmonia vault maintains pipelines for website deployment, knowledge extraction from external sources, translation across four languages, and social media engagement — each documented in its own file, each invokable by Claude when the relevant task arises. The same architecture applies to any domain: a paper-drafting pipeline, a patient-intake pipeline, a due-diligence pipeline. What matters is that the workflow is documented once, refined through use, and available to Claude without re-explanation.
Scheduled Tasks
The system extends beyond reactive collaboration into autonomous monitoring. Cowork supports scheduled tasks — automated operations that run on defined intervals, each writing a report to a designated location in the vault.
The Harmonia vault runs eight scheduled tasks: a daily website health check, a daily client profile sync, a twice-weekly knowledge drift detector (catching when vault edits have made the AI search index stale), a weekly vault state report (surfacing incomplete articles, broken links, and recommended writing priorities), a weekly translation staleness report, a weekly task reconciliation, and a biweekly integrity check on the CLAUDE.md itself. These are not alerts — they are the system’s ongoing self-examination, the digital expression of the Monitor principle at the center of the Wheel of Health.
The principle generalizes. A researcher could schedule a weekly literature scan (checking preprint servers for papers matching their research keywords), a monthly grant-deadline tracker, a biweekly student-progress review (surfacing supervision notes that haven’t been updated). An entrepreneur could schedule a daily news scan for their industry, a weekly competitive intelligence digest, a monthly investor-relations reminder. The vault becomes not only a knowledge base but a sensory organ — detecting what needs attention before you think to ask.
The Session Archive Protocol
Every working session produces knowledge — decisions made, problems solved, insights reached, configurations changed. Without a disciplined extraction protocol, this knowledge exists only in the ephemeral context of the session and vanishes when the conversation ends.
The archive protocol ensures that nothing valuable is lost. At the end of a significant session, you extract every decision into the decision log, update the CLAUDE.md with any new conventions or pipeline changes, route any new knowledge to the appropriate vault location, update the task list, and verify that all changes are saved to the persistent workspace. The discipline is simple: if it was worth discovering, it is worth recording. The next session inherits everything the previous one learned.
This is the living vault’s immune system against entropy. Without it, the knowledge base decays — conventions drift, decisions are forgotten, pipelines fall out of date. With it, every session leaves the system stronger than it found it.
The Compound Effect
The true power of this architecture reveals itself over time. A knowledge base is not a static archive but a living system that should grow more coherent, more interconnected, and more useful as it expands. Most tools make this harder at scale: more files mean more organizational overhead, more broken links, more forgotten corners.
Obsidian and Cowork together invert this dynamic. Graph structure means every new node strengthens the network rather than cluttering it — each note you add does not merely add itself to the collection but enriches every note it links to, and is enriched by every note that links back. A vault with ten notes has modest utility. A vault with a hundred notes begins to surprise you with connections you did not consciously intend. A vault with five hundred notes becomes an intellectual partner. The growth curve is not linear; it is closer to exponential, because each new node multiplies the possible connections across the entire graph.
Persistent memory means Claude’s ability to help you maintain coherence increases with corpus size. The CLAUDE.md sharpens with every session. The skills accumulate domain knowledge. The decision log provides precedent for future choices. The scheduled tasks catch problems earlier as the system grows more complex. Each layer reinforces the others.
For anyone engaged in serious intellectual work — building a philosophical system, managing a complex research program, accumulating deep expertise across multiple domains, leading an organization, or simply refusing to let hard-won understanding slip away — this is the current state of the art. It is not perfect and will be surpassed. But today, there is no better foundation for a mind that insists on sovereignty over its own knowledge.
What This Is — and Isn’t
This is a practical system, not a philosophy of note-taking. The philosophy comes from Harmonism — the conviction that every dimension of human life admits structure, that structure should serve Dharma, and that the tools we build should embody the same integrative coherence the Wheel of Harmony describes. The personal knowledge base is the Learning pillar’s expression of that conviction: sovereignty over your own mind, compounding understanding across time, and technology placed in service of consciousness rather than substituting for it.
It is also not a system that requires technical expertise to begin. Obsidian is free. Cowork requires a Claude subscription. The CLAUDE.md is a text file you write in plain language. The skills are markdown documents describing how you work. The entire architecture rests on open formats, local files, and tools you control. The barrier is not technical — it is the decision to take your own intellectual infrastructure seriously.
The first step is the simplest: install Obsidian, create a vault, and begin writing notes about what you are working on — linking ideas as the connections occur to you. The graph will emerge. The CLAUDE.md will grow. The skills will accumulate. And the system will compound, because that is what living systems do when they are built with the right architecture.
See also: Wheel of Learning, Sovereignty, The Living System, Dharma