Daily reflection is about clearing mental clutter and reviewing your choices. By pairing that reflection habit with a local Markdown vault in Obsidian and collaborating directly with an AI agent in the terminal, notetaking transforms into a natural conversation: you focus on thinking and planning, while the AI handles indexing, connecting insights, and turning raw notes into real-world action.

Many people assume building a "second brain" requires mastering complex keyboard shortcuts or installing endless plugins. For me, taking notes has never been about hoarding information. I just need a calm, reliable space for reflection: to review daily decisions, capture quick workplace lessons, and plan next steps.

Putting thoughts into writing lifts the burden of remembering trivial details. It slows down scattered thinking and brings immediate clarity to complex problems.

To keep this habit frictionless, I never organize folders or tag files by hand. Instead, I open my terminal and use Claude Code to build and maintain the entire system for me.

  [ Reflection & Voice Dictation ]
      Willow Voice / Raw Notes
                 │
                 ▼
[ Core Knowledge Vault (Obsidian) ] ◄───► [ Live Databases (Supabase) ]
   • Plain Markdown Notes (Local)           • Language study, finances, tasks
                 ▲
                 │
      [ AI Co-pilot (Terminal) ]
        Claude Code / Gemini
     (Auto-indexes, hooks, and skills)
                 │
                 ▼
   [ Publishing Platform (Notion) ]
     Public blog essays and shared docs

The system operates on three simple layers:

  • Obsidian: Stores every note as plain Markdown files locally on my laptop. The data belongs entirely to me, remains safe offline, and is never locked inside a proprietary cloud format.
  • Claude Code in Terminal: Acts as my operations partner. I let the AI write its own procedural skills, set up automated hooks, and maintain the central index (INDEX.md). Whenever I add a new reflection or lesson, the agent reads it, links it with past notes, and files it in the right place.
  • Supabase and Notion integration: Beyond plain text, the terminal agent connects to Supabase databases to query my Chinese vocabulary progress and personal finance, while publishing polished articles directly to Notion with a single command.

This setup mirrors the lightweight LLM Wiki architecture popularized by Andrej Karpathy: instead of paying for complex vector databases and embedding pipelines, you keep clean Markdown files and let an AI agent follow links and maintain your index.

If you are new to this and do not use the terminal yet, you do not have to struggle with command-line tools. You can use desktop applications like Codex Desktop App or folder-based workflows like Claude CoWork to open your Obsidian vault directly.

You can also watch this quick 5-minute video to see how to set up the foundation:

To get started right away in Claude CoWork or Codex Desktop, simply copy and paste this starter prompt to the AI:

"You are my LLM Wiki knowledge agent for this Obsidian vault. Help me set up my reflection and knowledge system:
1. Create a raw/ folder for unprocessed clippings and notes, and a wiki/ folder for distilled atomic concepts.
2. Initialize a central INDEX.md to track all topics and a log.md to record operation history.
3. Generate a CLAUDE.md (or AGENTS.md) guide defining how you will automatically ingest new notes, link related ideas using [[Note Title]], and maintain one-line summaries in INDEX.md.
Walk me through the setup step-by-step in plain language."

With an AI agent doing the heavy lifting, daily reflection stops feeling like a tedious filing chore and turns into an energizing, playful conversation with your own thoughts.

References & Further Reading

  • Nate Herk. (2025). Andrej Karpathy Just 10x'd Everyone's Claude Code. Watch on YouTube.
  • Andrej Karpathy. (2025). The LLM Wiki Architecture: Compiling Raw Notes into Structured Markdown Knowledge Bases.