I have tried many ways to take notes and jot down my thoughts. I have used popular note-taking apps like Evernote, OneNote, the native OS notes apps, and Google Notes, and I have tried Google Docs and handwriting notes on paper and on my Kindle. My notes have always felt all over the place. Obsidian is the single app that has changed that.
I have been using Obsidian for about nine months now. It has become my second brain. The first time I paired it with Claude Code, it was daunting. I remember watching Teresa Torres’ interview with Peter Yang where she demoed her setup, and finding it super helpful. Once I set it up and used it consistently, my comfort, knowledge, and understanding increased.
Almost a dozen people in the last month told me that this setup was new to them. I am sharing it here if you want to set up yours.
The Setup
This is not a fancy setup. There are no heavy plugins or extensions required. What you need:
An Obsidian Vault
An LLM of your choice.
A willingness to create shared context, rules, and skills.
Building Portable Context
Context matters for LLMs, and for a non-technical PM, a Markdown file is the easiest way to manage context without building a heavy architecture of knowledge graphs, APIs, and backend system connections.
My setup is a set of interconnected Markdown files with an Obsidian vault and any LLM of your choice. My personal vault stores context, rules, skills, workflows, and templates. All my projects live in Obsidian in folders and Markdown files. And I mean everything: my Substack posts, research topics, UX teardown analysis and reports, systems I have built, and daily reflections.
My setup has also improved over time. The first time I asked Claude Code to access folders connected to my Obsidian vault, it was with just a prompt to read what I wrote from there. Eventually, my CLAUDE.md file got an “Output Locations” table with my harness rebuilt. Claude also has instructions to read previously written information in the vault to pull context where necessary.
There is no other setup that is more robust for me than Obsidian connected to an LLM. So naturally, I use it with multiple LLMs, making the context portable.
Shared context between LLMs
The biggest unlock for me has been having one LLM review and give feedback on the other’s work. Connecting both to the same Obsidian vault has made it easy to collaborate with the LLMs. I use Claude Code and ChatGPT models, assigning each the builder or reviewer role interchangeably.
ChatGPT with Obsidian is my recent setup, and it is evolving. For now, I gave ChatGPT access to the vault in its project settings. Its Agents instructions file has four lines:
Read my CLAUDE.md and follow it.
Skip the parts only Claude uses
Read my guardrails before writing anything public, and
Never overwrite my drafts.
With two different models reviewing each other, I often find that each identifies gaps, and they collaborate well. This can be a double-edged sword when a lot of time is spent reviewing and updating the gaps, especially if I am working on a large body of work. For small changes, a couple of turns are usually enough.
I haven’t stuck with one note-taking app for this long. Do you use more than one LLM, and where does your context live?