Descriptions:
Nate Herk demonstrates how to build a self-organizing knowledge base — an “LLM wiki” — by combining Fable (an agentic layer built on Claude), Andrej Karpathy’s LLM Wiki system prompt, Obsidian as the local storage layer, and Claude Code to automate ingestion. The setup takes roughly five minutes: create an Obsidian vault, open it in VS Code, paste Karpathy’s gist into Claude Code, and the agent configures the schema, folder conventions, index structure, and ingest pipeline automatically. From there, any text source — YouTube transcripts, meeting recordings, research notes — gets pulled in and linked to related concepts without manual tagging.
The demo uses Herk’s own YouTube channel backlog as source material, with the resulting wiki surfacing relationships between tools and concepts like Claude Code, Vercel, N8N, and agentic workflows as a clickable graph. An interactive HTML visualization generated by Fable in a single prompt shows how ideas connect across videos, with backlinks, summaries, and a layered framework visible at a glance. Herk contrasts this with a similar project he built manually over a full day with Claude Opus 4 — same underlying data, but a notably less approachable interface.
A practical note for viewers: Fable was offering a free usage window through July 7, 2026, with the developer indicating plans to return it to standard subscription access. Herk acknowledges that Claude Code alone can handle the wiki scaffolding; Fable’s additional value is in reasoning over the accumulated context once the data is ingested.
📺 Source: Nate Herk | AI Automation · Published July 03, 2026
🏷️ Format: Tutorial Demo
