Concept: LLM Wiki
An LLM Wiki (sometimes referred to as Obsidian RAG) is a pattern for building a personal, compounding knowledge base managed primarily by an AI agent like Hermes or Claude Code.
Architecture
Instead of standard RAG (where an LLM retrieves raw document chunks on the fly), the LLM reads sources and incrementally builds and maintains a structured collection of markdown files.
Key Workflows
- Ingest: The LLM reads a raw source, summarizes it, updates relevant concept/entity pages, and logs the action.
- Query: The LLM searches the wiki’s index and synthesized pages to provide context-aware answers.
- Lint: The LLM routinely checks the wiki for contradictions, orphan pages, or stale claims.
Integration with Hermes
By creating a dedicated “LLM Wiki” skill, Hermes can interface with this directory. Users can ask Hermes to manually index articles or set up automated background tasks (cron jobs) to ingest information like daily meeting transcripts into the wiki.
Sources: