Integrations

Obsidian compatibility

Every page in the wiki uses [[wikilinks]], so the whole wiki/ folder already works as an Obsidian vault. If you have an existing vault, just symlink:

ln -sfn /path/to/JanusLM/wiki ~/your-vault/wiki

Recommended settings

  • Graph View — exclude index.md and log.md (-file:index.md -file:log.md) — they connect to everything and clutter the view
  • Dataview — the agent injects YAML frontmatter on every page (type, tags, sources) — Dataview queries work out of the box

Dependencies

PackagePurpose
Claude Code / Codex / Gemini CLI / Cursor / ClineAny agent that reads a config file
NetworkX + Louvain + vis.jsKnowledge graph generation and visualization

No server, no database — everything is plain markdown files.

Automatic dependency installation

Python dependencies (requirements.txt) are installed automatically on first use — no manual pip install needed. The mechanism is a two-layer design:

Installation mechanism

LayerWhatHow
Hook (deterministic)Detects missing depsChecks if .deps-ok exists — if not, signals the agent. Cost: <1ms per prompt.
Skill (intelligent)Installs depsClaude reads the /janus-setup skill, runs pip install, verifies imports, handles errors and retries.

Once installed, a .deps-ok marker file is created and the hook stays silent on all subsequent prompts. To force a reinstall, delete .deps-ok or run /setup.

The hook mechanism is Claude Code-specific (via UserPromptSubmit in .claude/settings.json). Other agents (Codex, Gemini CLI, Cursor, Cline) use a simpler check: they read the .deps-ok marker from their own instruction files and run pip install if it's missing.

Tips

  • When you query the wiki, the agent shows the answer first, then offers to save it as a synthesis page — your explorations compound like any other source
  • Every script in tools/ accepts --help
  • python tools/health.py checks KB integrity; python tools/lint.py finds gaps and suggests what's missing
  • Everything is plain markdown — portable, version-controllable, no vendor lock-in

try it

Ingest this paper and tag it as ai-strategy