Claude and Obsidian
It's funny how the things I do that I think nobody cares about get the most interest. One of those topics is how I'm using Claude and Obsidian together. This topic has been blowing up on the internet as well, so I thought I might as well share what I'm doing.
How I Use Obsidian
The first and most important thing is that what I use Obsidian for makes all the difference. Task management, note management, 2nd brain building, brainstorming, authoring are all different things you can do with Obsidian, and affects the rest.
Digital Garden
I use Obsidian to build my digital garden. This means that a large portion of my vault is public facing and published on the web. I use plugins like Templater and Dataview to breathe life into this and make my notes and website more dynamic and manageable.
This also means that a big part of what I do falls under authoring. I have notes set up for ideas on articles to write that Claude will use to suggest ideas, angles, and refinement. I don't use Claude to draft articles, but that's because writing is never the challenging part to me.
Client Work
In my vault I also do my client work. This isn't likely the best way to do it but I also can't justify splitting into another vault yet either. This means that my notes about my clients, meetings, etc are organized and available to Claude to read. While I don't really leverage Claude for my client work, reading what has happened helps with it having context over what I do, don't do, and challenges I deal with.
Daily Journal
I maintain a daily journal. It's nothing fancy. I use the daily note, write what's on my mind, and maintain a running to-do list of junk I need to do that may be related to plans with my family, beekeeping, consulting, or anything else.
Note Taking
I keep notes on all kinds of miscellaneous things like important things in my Minecraft worlds, beekeeping notes, plans for reko.day, and more.
Here Comes Claude
My workflow in combining Claude and Obsidian relies on pointing Claude Code to my vault, setting up a claude.md file and a directory that it can operate in. Mine is called z_AI. The z_ just helps keep the directory at the bottom of every list and out of my way.
If you're not familiar with the importance of a claude.md file, the main point is that this is the file that tells Claude how to behave, guardrails, processes to follow, and everything else. It defines how Claude will work with you.
If you've not made one, ask Claude to help and to interview you along the topics you're interested in help with.
Consulting Partner
The main use case I have is to have Claude act as a consulting partner. It knows how I do my job, what I'm good at, and what I'm not. It reads everything I write so it can provide much more accurate recommendations.
It knows that every day we work together. It reads my daily notes, works in them to record what we discuss and work on together so that every day it can pick up where we are and what we're doing.
The daily notes and broad access to your vault of writing and material is the secret sauce that allows Claude to never forget, never lose track of what's important, and provide relevant output no matter what happens. Normally Claude will lose "Context" and start over. The vault and daily notes are how you fix that "forgetfulness."
An example is that right now I'm building out a workshop based on Measures, Metrics, and Signals and Claude knows this. So every day it helps suggest what promotion to focus on, to build out material, and make sure progress is steady.
Garden Maintenance
Some of the coolest stuff that you get from Obsidian is born out of keeping your notes well-linked and maintained. That's some boring stuff that Claude is more than capable of doing.
Claude is happy to check my notes to make sure they have the right front-matter meta-data and things like that. It is also more than capable of figuring out what topics I might want to revisit.
Patterns and Ideas
One of the subtle things that I enjoy with Claude hooked up to my vault is that I can ask it really vague questions and it can produce very appropriate answers. This is because it has context to everything about my professional life.
One thing we did was leverage my client notes to build an Obsidian CRM so we could better keep track of outreach and keeping relationships going. This was born out of its analysis of my notes, and awareness of how I work with Obsidian.
This experiment quickly helped me realize there is a benefit to a CRM and so we quickly migrated to a real tool that Claude can also integrate with to help me maintain.
An Imperfect Tool
Look, Claude is great and I enjoy working with it, but it isn't perfect and there are plenty of tasks that it is not great at.
For example, I have yet to work with Claude on anything date-related and have it get the date right. Ever. Could it run a date command to know? Sure, but it won't. It'll make an assumption and build on it.
Another thing is that its responses are only based on training data, not thought or reasoning. This means I never look at what Claude provides as well-reasoned. Its answer is based on complex probabilities, and while that means its responses do have merit, it is not an expert that really thinks through anything.
An example from this morning was it told me that my response rate for something I've done with reko.day means I need to consider killing it. I asked it to actually research typical response rates first, then it backtracked.
I use Claude to help me stay accountable, brainstorm, challenge me, and maintain my work. I don't consider it an expert, I don't look at its responses as thoughtful, and I don't misattribute human reasoning to it, because it isn't those things.