A small kit for building a personal research wiki in Obsidian with any AI assistant (Claude, Gemini, ChatGPT…). Copy the template into Obsidian, paste a prompt into your assistant together with the paper, and paste the result back.

<aside> 💡

The rule of thumb: the agent fills in the facts; the questions and opinions stay yours.

</aside>

<aside> 🙏

Credits. This is a lightweight take on Prof. Joon An's LLM-Wiki lecture (CC BY 4.0) and his AGENTS.md template, which build on Andrej Karpathy's LLM Wiki. Those set up a full agent-run wiki with scripts, logs and a synthesis layer. This kit keeps only what fits in a copy-paste prompt, so it runs in any chat assistant without burning through tokens. If you want the full setup, start there.

</aside>

What it grows into

With consistent properties and [[links]], Obsidian's graph view turns the notes into a map of your reading: topics cluster around a few hub papers, and your own work sits where they meet.

My reading graph in Obsidian

My reading graph in Obsidian

Zoomed in, with the note titles blurred: formal semantics and pragmatics at the top, linguistic relativity at the bottom right

Zoomed in, with the note titles blurred: formal semantics and pragmatics at the top, linguistic relativity at the bottom right

How to use

  1. Create a note in Obsidian from the template below (one note per paper).
  2. Open your AI assistant, attach the paper's PDF, and paste the Fill-in prompt.
  3. Paste the answer into the note. Check the bibliographic details and numbers against the PDF.
  4. While you read, add your own questions and opinion. Later, use the Update prompt to add limitations or turn your margin notes into a checklist.
  5. Find more related papers: drop the paper into a citation-map tool such as ResearchRabbit to see later work that cites it and similar papers (the paper's own reference list only looks backwards). Add only the ones you will actually read to To read.

Keep the wiki small