llmwiki documentation

None

A local, stdlib-only Python knowledge base built from your AI-coding-agent session transcripts. Install in five minutes, then keep every session searchable, interlinked, and offline. No database, no account, no cloud.

Pick your mode

llmwiki runs two interchangeable ways. Pick one, start — you can switch later.

	API mode	Agent mode
Who calls the LLM	Python + Anthropic API	Your running Claude Code / Codex CLI
API key	Yes (`ANTHROPIC_API_KEY`)	No
Cost	Per token (with cache)	Included in your agent subscription
Concurrency	Batch + parallel	Serial
Best for	Large corpora, cron, CI	Interactive, exploratory

→ Read the full comparison before picking.

Getting started — 5 minutes

#	Tutorial	Time
01	Installation — macOS / Linux / Windows / Docker	5 min
02	First sync — from install to a browsable site	5 min

If it's not working in 10 minutes, open an issue — that's a bug in the docs.

Use with your agent

Claude Code — slash commands, session metadata, /wiki-ingest, /wiki-sync, /wiki-query.
Codex CLI — sync from ~/.codex/sessions/, live-session filtering.
Adapter reference: Claude Code · Codex CLI · Cursor · Gemini CLI · Copilot · Obsidian · OpenCode / OpenClaw · ChatGPT.

Use it locally

Query your wiki — /wiki-query, /wiki-graph, /wiki-lint, /wiki-candidates, /wiki-serve.
Bring your existing Obsidian / Logseq vault — llmwiki sync --vault <path>, non-destructive by default.
Example workflows — four real, end-to-end workflows.

Deploy

Target	Guide
GitHub Pages	deploy/github-pages.md
GitLab Pages	deploy/gitlab-pages.md
Docker / GHCR	deploy/docker.md
Vercel / Netlify	deploy/vercel-netlify.md
PyPI publishing	deploy/pypi-publishing.md
Homebrew tap	deploy/homebrew-setup.md

Reference

CLI reference — every python3 -m llmwiki <subcommand> with every flag and realistic examples.
Slash commands reference — every /wiki-* command used from Claude Code / Codex.
UI reference — every screen on the compiled site, how to reach it, what it shows.
Architecture — three layers (raw/ / wiki/ / site/).
Configuration · Full configuration reference.
Cache tiers — L1 / L2 / L3 / L4 frontmatter.
Prompt caching + batch API.
Reader API contract — stable shapes of every file llmwiki build writes.
Reader-first article shell — opt-in Wikipedia-style layout.
Entity schema — structured model-profile frontmatter.
Adapter authoring — write an adapter for a new agent.

Operate

Command cheatsheet — every slash + CLI command on one page.
Upgrade guide — what changes between releases, migrations, opt-ins.
FAQ · Troubleshooting · Privacy.
Accessibility (WCAG 2.1 AA).
Benchmarks · Competitor landscape.
Maintainers — governance docs at docs/maintainers/.

Contributing

Style guide — how to write docs that match this site's voice.
Adapter authoring — ship a new agent adapter.
Architecture — understand the three-layer model before changing code.
Roadmap · Public roadmap.

What llmwiki is not

It's not a vector database, not a RAG framework, not a hosted service. It compiles markdown from JSONL transcripts, writes a static site, and stays out of the way. The only third-party runtime dependency is markdown.

What's new

See the CHANGELOG. Latest tagged release: v1.3.82.

The version above is substituted from llmwiki/__init__.py:__version__ at build time, so this hub stays current on every release without a manual edit (#457).