Obsidian LLM wiki setup with Codex and Ollama

Callum walks through building an Obsidian wiki as shared memory for humans and an AI agent. The setup separates unchanged raw sources from linked wiki notes. A schema layer defines naming rules, metadata and workflows through agents.md, templates and repository skills.

The tutorial requires Obsidian, Python, Git and a coding assistant. Callum uses Codex to initialize the vault as a repository, create its structure and write maintenance scripts. He installs Obsidian skills and enables its command line interface. Obsidian Web Clipper captures a YouTube transcript, which the agent turns into concept pages with links back to the source. The maintenance pass checks formatting and queries; it also exposes a catalog search gap that the agent then fixes.

The advanced section adds a custom wrapper that checks approved vault names before calling the Obsidian CLI. The demonstration rejects a write to another vault, but does not establish a general security guarantee.

For the local LLM workflow, Ollama generates proposed notes in a review folder. The user approves a draft before applying it and running maintenance. Callum demonstrates double-click launch commands on Mac and says a Windows launcher could also be built. He recommends stabilizing the core workflow first, based on his view that local models make more mistakes. PDF ingestion and richer RAG queries remain proposed extensions rather than completed parts of this walkthrough.