Last night I wrote that connecting Obsidian to Claude felt like a turning point. After a day of using it, I can say better why.
The problem was simple: every Claude Code session starts from zero. Every time I had to explain again what Eon is, where HNCC runs, that the TV has a 32-bit userspace. Now that knowledge lives in an Obsidian vault, a folder of Markdown files that I read and Claude reads too.
How it’s organized
- Context: a short note, read first, with the current state of every project in a table and the open tasks. Only the present, no history.
- Protocol: the rules for how Claude should use the vault, shared between Claude Code and the desktop app.
- Projects: one note per project, with status, decisions and a dated history. Next to them, a folder of full technical summaries, one per project, like dated snapshots.
- Topics: who I am and what I prefer, the home infrastructure with every machine, the vault’s conventions.
- Conversations: a log of what we talked about and what got done, day by day.
- Inbox: quick notes, which Claude later files in the right place.
A CLAUDE.md file at the root of the vault is loaded at every start and brings Context and Protocol along. Two custom commands close the loop: /inbox sorts the notes, /chiudi saves everything before a session ends.
Light on purpose
I also looked at a ready-made plugin that turns the vault into a kind of model-generated wiki. It was well made, but it added fifteen skills, and every skill weighs a little on every message. It mostly did the same job as Context and Conversations, so I said no.
It’s the same idea I’m learning to apply everywhere: a small system I fully understand is worth more than a big one I half understand.
What has already changed
The articles I updated on this site today started right here: from the vault’s technical summaries, compared with what the articles said. That’s how I found out that two of my projects still claimed to run on the Mac mini.
Bye.