← Docs hub

Command cheatsheet

Everything you need on one page. Slash commands work inside Claude Code / Codex CLI; CLI commands run at your terminal.

30-second setup

From Claude Code, run these slash commands in order:

1. /wiki-init — scaffold raw/ wiki/ site/
2. /wiki-sync — ingest sessions from every auto-detected agent
3. /wiki-graph — build the AI knowledge graph
4. /wiki-build — compile the static site
5. Open site/index.html — the site is plain files, nothing to run
6. llmwiki install-automation — hand that loop to a daily job so you never repeat it

Keeping it current

Set the daily job up once and llmwiki keeps itself up to date. The wizard asks what the job should do (collect sessions only, or also summarise them), offers the optional extras, and shows you the exact command line before it writes anything.

llmwiki install-automation                                   # interactive wizard
llmwiki install-automation --yes --job ingest                # collect sessions + rebuild the site; no AI provider, no cost
llmwiki install-automation --yes --job maintain --schedule "0 8 * * 1-5"   # also summarise; weekdays at 08:00
llmwiki install-automation --yes --job maintain --graph builtin --lint-fail errors

--job maintain sends session text to your AI provider — run llmwiki synth --estimate first to see what a run costs. Re-running the command replaces the existing job instead of adding a second one. Full flag table: CLI reference.

For a one-off run of the same pipeline by hand:

llmwiki all                            # sync → synth → build → graph → lint
llmwiki all --no-synth                 # same run, no LLM calls
llmwiki all --skip-graph --skip-lint   # sync → synth → build only
llmwiki all --lint-fail errors         # exit 2 when lint reports an error

One step at a time

CLI commands you'll use most

llmwiki --help lists every subcommand in six lifecycle groups (Start here → Daily loop → Run the loop for me → Look around → Take things out → Rare). One-time vault repairs live under migrate <name>, not top-level migrate-*. These are the ones that carry the daily loop:

Knowledge graph

llmwiki graph                          # builtin wikilink graph (stdlib, zero deps)
llmwiki graph --engine graphify        # AI-powered: Leiden communities, confidence edges, god nodes
llmwiki graph --format json            # json only
llmwiki graph --format html            # interactive HTML only

Install Graphify: pip install llm-wiki-plus[graph]

Graphify outputs to graphify-out/: graph.json, graph.html, GRAPH_REPORT.md. Features: tree-sitter AST extraction, semantic analysis, community detection, confidence-scored edges.

AI-consumable exports

llmwiki build writes every AI-consumable export into the output directory (default site/). There is no separate export subcommand — replace llmwiki export all with llmwiki build.

llmwiki build                          # HTML site + llms.txt, llms-full.txt, graph.jsonld, sitemap.xml, rss.xml, robots.txt, ai-readme.md

Quality

llmwiki lint                           # 17 wiki-quality rules
llmwiki lint --json --fail-on-errors   # CI-friendly
llmwiki lint --rules link_integrity,orphan_detection
llmwiki lint --vault /path/to/vault --fail-on-warnings   # rules switched off in <vault>/llmwiki.json never run

Candidate workflow

llmwiki candidates list                # show all candidates
llmwiki candidates list --stale        # only stale (>30 days)
llmwiki candidates promote --slug Foo  # promote to wiki
llmwiki candidates merge --slug A --into B
llmwiki candidates discard --slug X --reason "hallucinated"

LLM synthesis

llmwiki synth                          # sources + candidates
llmwiki synth --sources-only           # sources only (legacy synthesize default)
llmwiki synth --check                  # probe backend (exit 0 if ok)
llmwiki synth --estimate               # cost (eligible sources) + Candidates (pre-run state)
llmwiki synth --force                  # re-synth everything, then harvest

Auto-tags pages (up to 5 AI tags per page, near-dup rejection, stop-word filter).

Adapters

llmwiki adapters                       # list every adapter + who fires on next sync
llmwiki adapters --wide                # untruncated descriptions

Core (auto-discovered, always loaded):

Contrib (load on-demand with --adapter <name>):

Support map + headless rules: multi-agent-setup.md.

Obsidian integration

# Sync wiki into an Obsidian vault
llmwiki sync --vault "~/Documents/Obsidian Vault/my-wiki"

# Build site from a vault
llmwiki build --vault "~/Documents/Obsidian Vault/my-wiki"

# Use the obsidian adapter to read from a vault
llmwiki sync --adapter obsidian

Flags you'll actually use

Config files

All configurable settings (sessions_config.json)

Three-layer architecture

raw/     IMMUTABLE transcripts (source of truth, never modify)
wiki/    LLM-generated pages (you own this)
  sources/      one summary per raw source
  entities/     people, products, tools (TitleCase.md)
  concepts/     ideas, patterns, decisions (TitleCase.md)
  projects/     codebases and work streams (kebab-case slug)
  syntheses/    saved query answers
site/    GENERATED static HTML (don't edit by hand)

Common recipes

# Daily, hands-off: schedule the whole loop
llmwiki install-automation --yes --job maintain --schedule "0 8 * * 1-5"

# Daily, by hand: the same loop as one run, then open site/index.html
llmwiki all

# Nightly cron (one project)
llmwiki sync --project my-project --no-auto-lint --since $(date -v-1d +%Y-%m-%d)

# AI knowledge graph
pip install llm-wiki-plus[graph]
llmwiki graph --engine graphify

# CI quality gate
llmwiki lint --json --fail-on-errors

# Export wiki to Obsidian vault
llmwiki sync --vault "~/Documents/Obsidian Vault/my-wiki"

# Full site rebuild with AI synthesis (exports included)
llmwiki build --synthesize

Exit codes

See also

• CLI reference -- every flag of every subcommand
• Slash commands reference -- what each /wiki-* does
• UI reference -- every screen on the compiled site
• Upgrade guide -- what changes between releases