Hermes Wiki

Hermes Wiki implements Karpathy’s LLM Wiki pattern as a production-ready plugin for the Hermes Agent. Instead of rediscovering knowledge on every query, agents build and maintain persistent wikis that compound knowledge over time.

Why LLM Wikis?

Traditional RAG uploads documents and retrieves chunks at query time. The LLM never accumulates understanding. Karpathy’s insight: LLMs should curate, not just retrieve.

With Hermes Wiki:

• Agents ingest sources into structured, cross-referenced wiki pages
• Knowledge compounds — each new source enriches the whole wiki
• Pages stay current via automated lint, monitors, and drift detection
• Everything is attributed and git-committed

Quick Links

Getting Started	Install, create your first wiki, ingest a source
CLI Reference	Full command documentation
Agent Tools	Using wikis from Hermes agent conversations
Architecture	Storage, pipeline, adapters, privacy
Hooks Architecture	Per-wiki executable customization design
Dashboard	Web UI setup and views
Karpathy Pattern	How Hermes Wiki implements the LLM Wiki concept
Quality Audit	Audit findings and roadmap for evals, features, and test suites
Media Ingestion Design	Decision record and build plan for multimodal ingestion

The Pattern

Raw Sources (immutable)     →  Ingest Pipeline  →  Wiki Pages (agent-curated)
articles, papers, transcripts   classify/process     entities, concepts, comparisons
                                                     cross-linked, attributed, versioned

Markdown is authoritative. SQLite is a rebuildable projection. Git tracks everything.