| title | llmwiki documentation |
|---|---|
| type | navigation |
| docs_shell | true |
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.
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.
| # | 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.
- 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.
- 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.
| 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 |
- 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 buildwrites. - Reader-first article shell — opt-in Wikipedia-style layout.
- Entity schema — structured model-profile frontmatter.
- Adapter authoring — write an adapter for a new agent.
- 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/.
- 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.
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.
See the CHANGELOG. Latest tagged release: v{{llmwiki_version}}.
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).