Claude + Obsidian Second Brain: Why This Combination Just Works

TL;DR

You can build an Obsidian second brain with ChatGPT, Gemini, or any modern LLM. But Claude wins for this specific job, and the reason isn't model intelligence — it's the surrounding tooling. Claude Code reads \CLAUDE.md\ files from your filesystem automatically. Claude.ai Projects let you persist context across sessions without re-pasting. Claude Skills (introduced 2025) let you compose reusable workflows on top of that context. Together they turn a static Obsidian vault into a living context layer that's loaded the moment you open a session. ChatGPT has Custom GPTs and Memory, but neither matches the directness of dropping a markdown file into a folder and having it auto-load. This post covers the why, the setup, and the three pitfalls that bite people switching from ChatGPT.

Why Claude Specifically

Most "second brain with AI" advice is model-agnostic. That's a mistake, because the tooling around the model is what determines whether the workflow actually sticks day to day.

Claude wins for an Obsidian second brain for three specific reasons:

1. Claude Code reads \CLAUDE.md\ natively. Drop a \CLAUDE.md\ file at the root of any project and Claude Code auto-loads it on every session. Add a \CLAUDE.md\ inside a subfolder and it stacks on top when you work in that area. This is exactly the architecture an Obsidian-based second brain wants: a root identity file plus folder-specific context that activates automatically. ChatGPT's Custom GPTs require manual configuration in a web UI; Claude reads from the filesystem.

2. Claude.ai Projects persist context across chats. Upload your three core files (identity, projects, decisions) once to a Project, and every conversation in that Project inherits them without you re-pasting. ChatGPT has Projects too, but Anthropic's implementation runs cleaner — files stay accessible via the file viewer, and the context budget is generous enough that you can include the full vault when needed.

3. Claude Skills (released 2025) compose on top. Skills let you save a workflow — a prompt + behavior + permissions — and invoke it later. Steph Ango (Obsidian's CEO) published an official Obsidian skill set that adds vault-aware capabilities. Combine your personal second brain (identity/projects/decisions) with reusable skills (capture, retrieve, summarize) and the AI starts behaving like a coworker with shared tooling, not a chatbot with amnesia.

None of these are individually transformative. Together they remove enough friction that the workflow becomes habitual instead of effortful.

The Setup, Step by Step

Assuming you've read the pillar second-brain post and the template breakdown, this is the Claude-specific layer.

Step 1 — Pick your loading mode

  • Mostly coding: install Claude Code and keep your three core files at the root of each project as \CLAUDE.md\ + supplementary files. Add per-project \CLAUDE.md\ files inside specific repos that need extra context.
  • Mostly writing/chat: use Claude.ai Projects. Create one Project per major area of your life ("Client work", "Personal writing", "Qobouli"). Upload the core files; add a Project description summarizing what the AI should do.
  • Both: use both. The same markdown files work in both surfaces.

    • Step 2 — Mirror your Obsidian vault into Claude

    If you keep your second brain in an Obsidian vault, the question is how to get those files into Claude's context. Three workable options:

  • Git-sync. Check the vault into a private git repo. Open it in Claude Code. Everything just works.
  • Manual paste. For one-off chat sessions, copy file contents at the top of the prompt. Slow but no setup.
  • Claude.ai Projects upload. Drag the three files into a Project. Re-upload when you make significant changes (you don't need to re-upload for every edit).

    • Most people end up using all three depending on the surface and the task.

    Step 3 — Add Steph Ango's Obsidian Skills

    If you're already in Claude.ai, install the official Obsidian skill set. They add vault-aware actions (capture, link, retrieve) that compose with your personal context. The combination — your personal files + the official skills — is where most of the leverage lives.

    Step 4 — Define your reload trigger

    The discipline that separates a working second brain from a stale one is the reload trigger. Pick one:

  • Friday afternoon weekly review. Open the vault, update active projects, re-upload to Project.
  • End of project milestone. When you ship something, capture the decisions to \decisions.md\.
  • Start of any new client engagement. Add a folder + \CLAUDE.md\ for the new context.

    • If you don't pick a trigger, the vault decays. Stale context is worse than no context.

    What Claude Does Differently in Practice

    Once the setup is in place, here's what changes about how Claude behaves:

  • It stops re-suggesting options you've already ruled out. Because \decisions.md\ lists rejected paths with reasons, Claude doesn't loop back to them.
  • It matches your stack without asking. "I'm building a thing" → Claude assumes Vite + Supabase + Tailwind because \identity.md\ told it that's the default.
  • It uses your tone. If \identity.md\ says "direct, no fluff, push back when warranted," Claude does. ChatGPT will too, but Claude's defaults match this style more naturally.
  • It pulls cross-project parallels. When projects share patterns ("this client wants what the last client wanted"), Claude notices because \projects.md\ is small enough to reason across.
  • It calls out when context is stale. If \projects.md\ says a project is "in progress" but your message implies it shipped, Claude often asks. ChatGPT tends to just go along.

    • These are small effects individually. Compounding across thousands of interactions, they're the difference between productivity AI and theater AI.

    Three Pitfalls When Switching from ChatGPT

    People moving from a ChatGPT-based workflow consistently hit these:

    Pitfall 1: Treating Custom GPTs and Projects as equivalent. They're not. Custom GPTs are GPT-with-a-system-prompt. Claude Projects are workspaces with persistent file context that any chat inside inherits. Don't try to recreate a Custom GPT in a Project — use the file system, not the system prompt.

    Pitfall 2: Over-stuffing the vault. Claude's context window is generous, which tempts people to upload everything. Don't. The same curation rule applies: every file you include competes for attention. A focused 30-file vault beats a 400-file vault every time.

    Pitfall 3: Skipping \CLAUDE.md\ because it feels redundant. If you use Claude Code at all, name your identity file \CLAUDE.md\ and put it at the root. The auto-load is the single biggest workflow upgrade over manual prompts. People who skip this step keep wondering why their setup feels clunkier than it should.

    When NOT to Use Claude for This

    Honest counterpoint: Claude isn't always the right tool.

  • You're already deep in a ChatGPT Teams or Enterprise workspace — switching surfaces is expensive. ChatGPT's Memory feature has improved enough that a ChatGPT-native second brain is workable, even if Claude is cleaner in isolation.
  • You need image generation or browsing baked into the same surface. Claude has both via tools, but ChatGPT's experience is more integrated.
  • You're testing an open-source model locally for privacy. Ollama + a local model is a different setup; the principles in this post still apply, but the tooling is different.

    • For most working professionals who want AI to remember their context and stop being generic, Claude + Obsidian + the three-file template is the cleanest path.

    Where to Go Next

  • Start with the 3-file template if you haven't yet.
  • Read the pillar second-brain post for the philosophy.
  • Layer in Steph Ango's official Obsidian skills once the basics are in place.

    • The compounding effect is real but it requires the first month of friction. Push through it and the workflow becomes invisible.