Build a second brain
your AI can read.

The whole method rests on one choice: your notes are Markdown files in a folder on your own disk, synced between your devices. Not a product’s database, not a chat history, not a “memory” feature inside an AI app. A folder of text files can be read by any tool, searched by any agent, backed up anywhere, and moved to a different app in an afternoon.

Create the vault, turn on sync, and open it on your second device to confirm the files appear. If you run an agent on a server, install the sync client there too, so the agent and you read and write the same files.

Every knowledge base needs exactly one entry point, and every AI tool you connect will be told to read it first. Call it Home Base. It holds your current situation in a paragraph, a map of the top-level folders and what each holds, your daily rhythm, start-here links for each area, your priorities for the week, and the links to the rules in step ten.

Keep it to one screen. Home Base is a map, not a place where content lives. If you find yourself writing paragraphs there, they belong in a note it links to.

Top-level folders get a two-digit number in the order you want to read them: 01 Daily Notes, 02 Personal, 03 Health, and so on, with 99 System last, because the vault’s own rules are infrastructure and should sit out of the way. Folders inside a stage-based workflow get numbers too, because the number is the sequence.

Topic folders inside an area (Ideas, Recipes, Strategy) get no number, so they sort alphabetically. Two digits everywhere, no single-digit prefixes, so there is one rule to remember.

Do not create an inbox folder. Capture-then-sort only works if the sort happens, and in practice it does not. File things where they belong the moment you capture them.

A knowledge base rots when every hub invents its own name. Pick one word per job and retire the rest.

• Home Base: the one root entry point. Exactly one in the vault.
• 00 [Area] Index: a folder’s contents list. It lists and does not explain. One per folder that needs one.
• 00 [System] Home: a system with rules and a workflow. It explains what each sibling file is for. One per system.

Both carry the 00 prefix so a folder’s front door always sorts first; nobody should have to hunt for a folder’s own entry point. Words like Hub, Dashboard, Panel, and Vault are retired as entry-point names. Dashboard keeps its literal meaning for a note about a real built dashboard.

The test for any note name: what is this, and why would I search for it next month? A few rules cover almost everything.

• Date as a suffix for topic notes that happen to be dated: Costco Receipt - 2026-07-24, Agreement Review - 2026-08-01. You will browse these by topic.
• Date as a prefix for logs where the date is the identity: 2026-08-03.md for a daily note, 2026-08-03 Podcast Digest. You will only ever browse these by date.
• Area prefixes for operating notes: [Business] - Tax and Admin Questions.
• Person notes are First Last, or First Last - Company when the company disambiguates.
• A leading underscore for templates, so they sort above real notes: _Daily Note Template.

Every note starts with a small block of metadata (frontmatter). Folders are the primary filing system; metadata adds the cross-cutting context that tools can filter on. The minimum is three fields: tags, the date created, and a status.

Give each key exactly one meaning. A single status field that tries to carry lifecycle, provenance, and reading progress ends up with eight values nobody can filter on. Split them: status for lifecycle (active, reference, archive, log, template), source for provenance (which notes you wrote yourself, which an AI drafted), progress for media only, and review for who needs to see it (daily, periodic, agent-only).

Tags are a small number of stable families in lowercase kebab-case, such as health/diabetes or work/learning. A tag that only repeats the folder path is not earning its place. Keep a short table of retired tags so old names do not creep back. And the rule that matters most: whoever creates the note writes the metadata, whether that is you, a template, or an agent. A cleanup pass later always loses to a script writing one untagged note a day.

This is the step that makes an AI actually know you. Create an About folder with an index and a handful of evergreen notes: your values and operating principles, your background, your family and the people closest to you, lessons you have learned, your favourites, what you follow, the recurring shape of your week.

No dates and no logs here; only settled material. And write them yourself. Every AI you connect will read these notes for context, and if an AI drafted them, every tool downstream learns to imitate an earlier machine rather than you. An hour of your own writing here is worth more than any amount of structure elsewhere.

The daily note is where the vault stays current without effort. One file per day, date-prefixed because the date is its identity, made from a template with a few optional sections: what actually happened in your own words, energy and health, work, people, a line to carry forward.

Every section is optional, and a three-line entry beats a skipped day. Link people by name with double brackets, so the day shows up on their note and an agent asked about someone can see the history.

Now tell your tools about it. Each AI tool you use has somewhere for standing instructions: a global instructions file for a coding agent, custom instructions in a chat app, the identity file of a personal agent. Add the paragraph below, adjusted to your paths. It does three things: names the vault as the source of truth, says where to start reading, and tells the tool to write facts about you back into the vault rather than into its own private memory.

That last part is what keeps one knowledge base instead of five. An AI’s own memory feature is convenient and invisible; the vault is yours, readable, and shared by every tool.

If you run personal agents, give each one a folder it owns inside the vault, with an ownership note. The companion guide on giving an agent a memory and limits covers that.

A knowledge base that only grows stops being trusted. Three habits keep this one current.

• Every note has a job and a lifecycle. Active while it is useful; completed when the action happened but the result still matters; archived to a folder under 99 System when it is only history; deleted when it has no durable value. Setup checklists and one-time prep do not stay active after the thing is done.
• One strong hub per domain. Before creating a second index, backlog, or home note for something, check whether one exists. Duplicated hubs drift apart within weeks, and then neither can be trusted. When you rename or merge a note, rewrite every incoming link to the new name; do not leave a stub behind.
• Live views that read the metadata. In Obsidian, a Base is a table that queries frontmatter: everything by status, recently changed, templates, archive. A wrong status drops a note out of the view, so you notice. If you run an agent, give it a daily sweep that reports missing metadata, broken links, and stale notes, and a short audit note every month or two recording what changed and why.