Found something good?

One evening of setup. A vault that files itself at 7 a.m. A model that opens every session already knowing your work.

A 33-year-old science journal editor in Porto kept his best thinking in 2,400 markdown files for 8 years. He hadn't opened a single one of them in the last year. The graph in Obsidian looked like a star chart with no constellations - beautiful and dead. On Monday night he connected Claude through MCP. By Sunday he'd shipped a piece he started writing in 2019 and abandoned on page 47.

Here are the eight rules that turn the vault into something that works for you, instead of something that works on your guilt.

1. Voice beats keyboard

Notes that require typing don't get made. Notes you can dictate in two seconds at a crosswalk always get made. Any voice recorder with auto-transcription drops a file into inbox/. Claude sorts the rest.

2. One inlet, many outlets

The vault has exactly one folder where you dump raw thought: inbox/. Every other folder - topics, projects, people - Claude builds out of it. If you have 14 places to put a note, you'll spend 14 seconds deciding and end up putting it nowhere.

3. Morning belongs to Claude. Night belongs to you

Split the clock. Between 6:30 and 8:00, a cron job kicks off: Claude reads everything new in inbox/, files it, drops [[backlinks]] into the existing graph, and writes a digest into the daily note. You open the laptop at nine and find yesterday already mapped. No "I need to clean up my notes" on a Saturday afternoon.

4. The raw/ folder is sacred

Nothing inside inbox/raw/ gets edited - not by you, not by Claude. It's your unfiltered voice, with a date and a context. If something needs rewriting, Claude saves the cleaned version into processed/. Raw stays as a geological layer. Three years from now you'll be able to reread what you actually said, not what the system decided you meant.

5. The backlink matters more than the note

A note with no connections is just a file. A note with five [[backlinks]] is a node of thought. Write it into Claude's system prompt: "every new note gets at least three backlinks to existing notes, and one to a note that's 2+ years old." The graph starts thickening in the first week.

6. The Sunday synthesis is the only thing you'll actually reread

Once a week Claude does the one thing you'd never do on your own: it reads everything from the past seven days and writes a single file - weekly-synthesis/2026-W26.md. Recurring themes. Contradictions in your own thinking. Promises you keep half-making to yourself. That's the one file each week you'll come back to.

7. Context into every session

Every new chat with Claude doesn't start with "hi, I'm Ivan" - it starts with an automatic load from the vault: what you're working on right now, which hypotheses are still open, what you decided last Tuesday. Twenty seconds saved at the start of each session turns into 30 minutes of real conversation instead of 30 minutes of catching the model up.

8. The graph is the vault's pulse

If the graph is growing and the link density is climbing, the vault is alive. If files are accumulating and connections aren't, you're running a warehouse, not a second brain. Look at the graph for five minutes once a month. It's the only metric that means anything.

Six weeks in, the editor in Porto noticed his own vault had become smarter than he was. He'd ask Claude, "I had a thought about the epigenetics of memory the autumn before last" - and get back not one note but three linked ones he'd long forgotten about, plus a draft of a conversation with a colleague from November 2023 where, it turned out, the answer had been hiding the whole time.

This isn't a "second brain" in the influencer sense. It's just a working archive. The difference is simple: one you hold together with willpower; the other holds itself together.