oh-my-wikiGitHubInstall

A wiki your AI agent helps build as you work.

The more you work, the more your knowledge stacks up and connects instead of slipping away. Across supported AI agents, oh-my-wiki keeps the sources and summaries you approve, finds the right notes, and brings useful context back when you need it.

$ pipx install oh-my-wiki
Get started →Read the tutorial

AI sessions lose context. Your notes scatter.

AI sessions have limited context, and notes pile up with nothing connecting them. oh-my-wiki recalls relevant wiki pages as you work. On Claude Code and Codex, it can also stage a small, redacted same-project snapshot for the next session without turning that snapshot into a wiki page.

What is an LLM wiki?

A living collection of knowledge your agent keeps with you. Hand it a source, and it preserves the original, then proposes a short summary and useful links for the people, tools, and ideas that came up. After you approve, one source can update several related pages at once, often around 10 to 15.

It works the way you'd hope

Just ask in plain language — "save this", or "what do I know about X?". oh-my-wiki shows you what it will save and how it will connect things, and applies the changes only after you approve.

Just ask

No commands to memorize. Ask the way you'd talk to a person, and oh-my-wiki helps your agent handle the saving and linking.

Nothing changes without your OK

It shows you every change first and applies it only once you approve. Your files always stay yours.

Smarter than a plain wiki

A basic LLM wiki stops at storing and linking. oh-my-wiki goes further — it helps you keep your knowledge checked, updated, and easy to find.

Consistent by default

Every kind of page — person, tool, idea — follows a tidy template, so your wiki stays easier to scan.

Visible confidence

Mark how sure each note is, and retire outdated ones without leaving clutter.

Nothing slips away

It brings notes worth a second look back at the right time, so important things don't fade.

Finds the right note

Full-text evidence and optional local embeddings work together, while conservative filtering keeps weak matches out of your prompt.

Recalls before you ask twice

Native host hooks can bring relevant wiki context into a session, a prompt, or a file operation before your agent acts.

Carries only what matters

Claude Code and Codex can stage a bounded, redacted same-project snapshot locally. It expires automatically and never becomes a wiki page on its own.

Connected to the world

Pull in fresh material from the web, and open your wiki up to other apps when you want.

Suggests useful links

It spots related mentions and proposes useful links; you decide which links are inserted.

See how your topics connect

It uses your wiki links to show clusters, bridge pages, and hubs.

Catches the little details

Jot small facts like a date or status right inside a note, and omw can read them as fields.

Korean, done right

Korean names are matched correctly even with attached particles (josa).

See it in Obsidian or Logseq

Open your wiki, a page, or a search in Obsidian or Logseq with one command — no plugin needed.

Save from the web

Drop in a URL or YouTube link and it pulls the page or transcript straight into your wiki.

Publish only what you choose

Every page is private by default; mark the ones you want public, and only those are shared.

Use the AI you already use

The same wiki works in Claude Code, Codex, Gemini CLI, OpenCode, Hermes, and OpenClaw. Each host uses the native integration it supports; full before-tool recall and staged capture are currently available on Claude Code and Codex.

See it in action

In your AI session, say ingest this and paste a source. The agent saves it, summarizes it, and suggests links for the key people and ideas it finds.
ingest this

Andrej Karpathy calls the LLM Wiki a "compounding knowledge
artifact". Every source gets saved verbatim to raw/, a summary
lands at wiki/summaries/, and the entities and concepts that
appeared get their own pages.
Read the full tutorial

Install

Pick the path that fits your setup, then run omw doctor. The Skills CLI path installs the skill only — say set up omw in your agent to bootstrap the CLI. Run omw setup to create a vault and wire the integrations your hosts support.
pipx install oh-my-wiki        # isolated CLI (recommended)
# or
pip install oh-my-wiki         # into the current environment
omw setup agents               # register the bundled skill