Install & Quick Start

Install

pip install my-llm-wiki

Optional extras:

pip install my-llm-wiki[docling]   # layout-aware PDF/DOCX/PPTX/HTML/EPUB extraction with OCR fallback
pip install my-llm-wiki[all]       # everything: PDF + office + Docling + better clustering

Build a graph

cd your-project
llm-wiki .

Output goes to wiki-out/:

wiki-out/
  graph.html       ← open in browser
  graph.json       ← persistent graph data
  WIKI_REPORT.md   ← analysis report
  wiki/            ← Wikipedia-style articles
  vault/           ← Obsidian vault — index.md catalog + [[wikilinks]] + YAML frontmatter

Open in Obsidian: Point Obsidian at wiki-out/vault/ (Open folder as vault) — graph view, backlinks, tag pane, and Properties all work out of the box. Start from index.md (auto-generated catalog by file type + communities). Community colors are pre-configured via .vault/graph.json.

Query

llm-wiki query gods                      # most connected nodes
llm-wiki query search "authentication"   # keyword search
llm-wiki query node UserService          # details + doc comment
llm-wiki query neighbors UserService     # what connects to it?
llm-wiki query path Auth Database        # shortest path
llm-wiki query community 0              # largest community
llm-wiki query stats                     # summary
llm-wiki query orphans                   # isolated nodes (no connections)
llm-wiki query stale-refs wiki-out/vault # broken [[wikilinks]] in vault

Health check

llm-wiki lint    # orphan nodes, tiny communities, confidence breakdown

Keep the wiki alive

llm-wiki watch .                               # auto-rebuild on changes
llm-wiki add https://interesting-article.com   # ingest URL
llm-wiki .                                     # rebuild (cache skips unchanged)

Capture from LLM sessions

Extract insights from Claude Code session logs:

llm-wiki capture --enable    # opt-in once (required for privacy)
llm-wiki capture             # scan sessions for note candidates

Outputs to wiki-out/captured/pending-notes.md. Review candidates and run llm-wiki note to promote them to the vault. Filters by keywords (rationale, trade-off, decided, etc.), min length 50 chars, and skips secrets.

Claude Code skill

For agent-mode semantic extraction and vault maintenance:

mkdir -p ~/.claude/skills/my-llm-wiki
cp "$(python -c 'import my_llm_wiki; print(my_llm_wiki.__path__[0])')/SKILL.md" ~/.claude/skills/my-llm-wiki/
cp "$(python -c 'import my_llm_wiki; print(my_llm_wiki.__path__[0])')/MAINTAIN_SKILL.md" ~/.claude/skills/my-llm-wiki/

In Claude Code:

/wiki . — full extraction pipeline with deep synthesis (DOCX, scanned PDFs, images)
/wiki maintain — semantic vault audit (contradictions, orphans, broken links)

Ignore files

Create .wikiignore in project root (gitignore syntax):

vendor/
node_modules/
dist/
*.generated.py

Supported languages

Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, Ruby, C#, Kotlin, Scala, PHP, Swift, Lua, Zig, PowerShell, Elixir