Obsidian and Claude Code: maintain a wiki without losing the sources
Follow the two-stage setup and ingestion approach, then check citations and scheduled maintenance separately.
Video in English
Watch on YouTube ↗Start with a small, explainable archive
Luke uses transcripts from his last 30 videos to build an interlinked wiki in Obsidian with Claude Code. The description attributes the LLM-wiki idea to Andrej Karpathy and distinguishes this organised collection from RAG. Treat that as the approach presented, not a claim that every retrieval system works the same way.
- Install Obsidian from https://obsidian.md and create a vault for the material you want to query. Only include information you are comfortable sharing with your AI service.
- Read the linked idea: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f#llm-wiki. Decide whether separate vaults would keep unrelated work easier to manage.
Set up the rules before ingestion
The video separates infrastructure setup from adding data. Its additional prompt names an AGENT.md schema file, index.md and log.md. Preserve those exact filenames when following that example rather than silently substituting another convention.
You are now my LLM Wiki agent. Implement this exact idea file as my complete second brain. Guide me step-by-step: create the AGENT.md schema file with full rules, set up index.md and log.md, define folder conventions, and show me the first ingest example. From now on, every interaction follows the schema.
- Use the setup prompt retained in the video description to define the schema, folder conventions, index and log. Inspect the resulting files before proceeding.
- Keep original material in raw and derived, interlinked pages in wiki. Ask for a first ingestion example and check how it links back to the source.
Ingest and interrogate the evidence
ScriptiKit is the retained transcript resource: https://scripti.ai/scriptikit. Obsidian Web Clipper is linked for web material: https://obsidian.md/clipper. Their presence does not establish access to every transcript or permission to copy private content.
- Add a small set of permitted sources, run ingestion and inspect the new pages against the originals. Record what was added in the log.
- Ask a question whose answer you can check. Follow the citations to the exact video or note and keep unsupported conclusions visibly uncertain.
Treat maintenance as another test
The demonstration includes scheduled updating at 13:06 and a separate mem0 sponsor segment. The description’s three-month offer with code LUKE is historical; this article does not establish current eligibility.
- Before enabling a routine, inspect its source locations, timing and instructions for missing or changed material.
- Check a completed update for duplicate pages, preserved originals and valid citations. A saved schedule alone does not prove successful ingestion.
The takeaway
Keep the wiki navigable and the originals intact. The useful result is a checked answer with traceable evidence, not merely a larger collection of generated pages.
About this article and its sources
Prepared 1 November 2026 from this exact published video description and chapter sequence. Steps are editorial follow-along guidance, not quotations from a reviewed transcript. The full video was not independently watched or transcribed and the product was not retested. Historical offers and setup details need a fresh check at the original resources.
