Claude Cowork on an Obsidian vault: my second brain, finally with a colleague

By Nick Beaugeard · 7 minute read · ← All posts

I have kept an Obsidian vault since 2022. Client notes, meeting transcripts, governance documents, blog drafts, prospect intel, project logs, code snippets, ideas I will never get to. Several thousand interlinked markdown files. A second brain that, like a lot of second brains, was very good at storing things and very bad at doing anything with them.

Claude Cowork changed that. Pointed at the same folder of markdown files, Claude is no longer a chatbot you query. It is a colleague who has read everything you have written and can act on it. After several months of daily use, this is the most material productivity change I have made in years.

The setup

The pieces are unremarkable. The combination is not.

• Obsidian vault. Local markdown files in a folder. Mine sits in OneDrive for sync, with a git repo on top for history and remote backup.
• Claude Code with Cowork. Running over the vault as if it were a codebase. Plugins for productivity, sales and governance toolkits are loaded.
• A CLAUDE.md at the vault root. Brief, declarative. Who I am, what the vault contains, what files are sensitive, what naming conventions I use, what I want Claude to never do without asking.
• A memory/ directory. Curated facts about my clients, projects, people, recurring jargon and shorthand. Updated by me, read by Claude on every session.
• Sensible permissions. Read everything. Write only inside specific folders. Never push to git without a heads-up.

That is it. No vector database. No custom MCP server. No fine-tuning. The vault is the context.

The workflows that paid off

Some of these were obvious. Others took weeks of use before clicking.

Triaging the inbox

Every morning, Claude reads the previous day's email summary file, cross-references it with my active client and project notes, and proposes a triaged task list. Things I have already committed to. Things that look new but are actually duplicates of an existing thread. Things that need a reply by end of day. The triage takes me 90 seconds to review instead of the 20 minutes it used to take to do from scratch.

Meeting prep

Before a client call, "prep me for the meeting with X". Claude pulls every prior note, every previous transcript, every commitment we have made, and gives me a one-page brief: who attends, what we discussed last time, what was promised, what is overdue, what to raise. The first time I tried this, I walked into a meeting feeling embarrassingly well prepared. Now I rely on it.

Meeting follow-up

After the call, paste the transcript into a new note. Claude reads it, updates the relevant client folder, writes the follow-up email in my voice, files the action items into the right project notes, and flags anything that contradicts a previous commitment. The "second-brain hygiene" tax that used to consume an hour per meeting now consumes five minutes.

Drafting in my voice

Because the vault contains years of my writing, Claude has an unfairly good model of how I talk. Blog drafts, client proposals, board memos, LinkedIn posts — the first draft is no longer the slow part. I edit, I do not author from zero. Every post on this blog now starts as a Claude draft against a brief and a reference set of my prior writing.

Governance and admin

The vault holds my company's governance documents: registers, policies, board pack templates. Asking Claude to update the share register after a transaction, draft minutes from notes, or check whether a policy needs a review is faster than opening the right file myself. The work is still mine; the typing is no longer.

Knowing what I have already said

The killer feature, honestly. "Have I written about this before?" "What did I tell client X about pricing in February?" "What is the latest thinking I have logged on this topic?" My vault was a search problem for years. Now it is a conversation.

The guardrails

Letting an AI loose on your second brain requires care. Mine, in order of importance:

1. Git on the vault. Every change is committed. Anything Claude touches can be reverted. This is non-negotiable.
2. Read by default, write by request. Cowork plugins can read freely; write actions need explicit approval. I review every PR-like change before it lands.
3. A list of "never touch" folders. Anything sensitive (client confidential, financial, legal) is read-only and flagged in CLAUDE.md.
4. Memory hygiene. The memory/ directory is curated weekly. Stale facts get pruned. Duplicate entries get merged. This is the difference between Claude being useful and Claude being confidently wrong.
5. Privacy review before sharing. Nothing in the vault gets pasted into a shared workspace or a public document without a check. The vault is mine; what leaves it is a choice.

Where it falls down

Honest caveats from daily use:

• Context window is finite. The vault is bigger than the window. Claude has to choose what to load. Good naming conventions, clear folder hierarchies and an honest CLAUDE.md help. Sloppy structure means Claude misses things.
• Stale facts persist. If you do not curate memory, the system will confidently tell you facts about your business that have been wrong for six months. Treat memory like a database, not an inbox.
• Tool sprawl. The temptation to install every plugin is real. Resist. Three or four plugins, well used, beat thirty plugins you forget you enabled.
• It does not replace thinking. Claude can draft, summarise, cross-reference and remind. It cannot decide. The decisions are still yours; if anything, you make more of them, faster.

What changes when your second brain answers back

The most surprising effect has not been speed. It has been confidence. I commit to things in meetings knowing the follow-up will not fall off the radar. I sit down to write knowing the first draft will exist within an hour. I look at a busy week knowing the triage is solved. The vault used to be a slightly guilty pile of unfinished thoughts. It now feels like a workspace with a colleague in it.

The point of a second brain was never the storage. It was being able to act on what you know. Cowork closes that loop.

If you want to try it

The setup is genuinely a weekend project if you already use Obsidian:

1. Install Claude Code and configure Cowork against your vault folder
2. Write a one-page CLAUDE.md at the root: who you are, what is in the vault, what is off-limits
3. Add git on the vault and commit before you let Claude write anything
4. Curate a memory/ directory with a dozen factual notes about your business and your people
5. Pick one workflow and run it daily for a week. Meeting prep is the easiest win.
6. Add plugins only when a real workflow demands one

If you run a firm or a team and you are trying to work out where AI delivers actual productivity, the answer is rarely "buy another SaaS tool." It is usually "make your existing knowledge usable." Cowork plus Obsidian is the cleanest version of that I have found.

The short version

Claude Cowork over an Obsidian vault turns a notes folder into a working partner. The setup is simple; the discipline around memory, permissions and git is what makes it durable. The shift is not from "doing things slowly" to "doing things faster." It is from "remembering things" to "acting on them." If you have a vault sitting idle, this is the upgrade worth making this quarter.