Definition
Wyndo is the author of the GenAI Unplugged newsletter and a practitioner of the LLM Wiki pattern. Their production experience with the system led to the development of the four guardrails that many implementations now follow. They demonstrated a complete LLM Wiki workflow: 2 sources → 20 cross-linked wiki pages in ~10-12 minutes.
Key Facts
- Authored the GenAI Unplugged article “AI Second Brain: Claude + Obsidian” that documented the LLM Wiki pattern in detail
- Developed the four production guardrails: link cap, twice-referenced rule, forward-only linking, edit ceiling
- Demonstrated the key insight: “The LLM finds correlations between things you saved separately that you would never have drawn on your own”
- Showcased a mobile-to-desktop workflow: voice note → Obsidian mobile → sync → Claude Code processes
- Runs multiple LLM Wikis for different domains: YouTube transcripts, meeting recordings, business data
Dynamic Structure
Wyndo’s production experience revealed that the LLM should decide the folder structure based on content, not the human. His YouTube wiki has nested folders (concepts, entities, tools, techniques) while his meeting wiki started flat — the LLM decided it didn’t need subcategories yet. The principle: “Sometimes keeping it flat is better. The goal is that the AI can easily search through all content.”
Implications
This means the folder structure is an output of the ingest process, not a design decision made upfront. The LLM’s judgment about what structure best serves retrieval should be trusted over human intuition about how to organize.
Routing
Wyndo runs multiple LLM Wikis for different domains: YouTube transcripts, meeting recordings, and business data. The system routes queries to the right wiki based on context. Multiple wikis can coexist, each handling different domains.
Implications
This means the LLM Wiki pattern scales horizontally — you don’t need one monolithic wiki. Different domains can have their own wikis with their own folder structures, and the routing layer directs queries to the appropriate one.
Wyndo’s Ingest Process
Wyndo describes a simpler 5-step ingest process:
- Drop a source (PDF, URL, text) into the raw/ folder
- Tell the LLM to ingest it
- The LLM reads the source and the existing wiki
- The LLM decides what pages to create/update
- The LLM writes the pages and updates the index/log
One source might generate 5-50 wiki pages depending on content richness.
Implications
This means the ingest process is intentionally simple from the human’s perspective — the complexity is entirely on the LLM side. The human’s job is just to drop sources and trigger the operation.
Open Questions
- How does Wyndo handle the routing between multiple wikis? Is it manual or automated?
- What determines when a wiki should be flat vs nested? The LLM decides, but what criteria does it use?
Related Concepts
- llm-wiki — The pattern Wyndo practices and documents
- genai-unplugged — The newsletter where Wyndo publishes LLM Wiki findings