Skip to content
obsidian-wiki

obsidian-wiki

obsidian-wiki

A Claude Code plugin that turns Obsidian folders into a wiki an agent maintains for you

You already save things you mean to come back to: quick notes, PDFs, saved articles, voice memos, exported chats. Most of it you never open again. obsidian-wiki fixes the second half of that habit. You drop a source into a folder, run one command, and the agent reads it and writes it up as short, cross-linked wiki pages: what the source said, distilled, with a link back to the original and a note on how much to trust it. You read and edit the result in Obsidian, exactly like any other note.

Nothing about it is magic and nothing runs on its own. It is a set of commands you invoke, a few folders in your vault, and a plain-markdown output you own. If you deleted the plugin tomorrow, every page it wrote would still be there and still readable.

ZONE 1 · ENTRY POINTS Quick-notes quick-note Articles, PDFs article Transcripts voice-transcript LLM conversations claude-chat READ /ingest classify, dedup, distill WRITE ZONE 2 · STRUCTURED KNOWLEDGE Projects active work Documentation curated articles Resources references, lists People person pages Concepts companies, frameworks ZONE 3 · SERVICE, READ AND WRITTEN BY EVERY COMMAND manifest sha256 ledger log append-only hot.md last 20 touches sources per-source pages archives snapshots feedback behavioral rules LEGEND read write

Why a wiki works well for an LLM

Section titled “Why a wiki works well for an LLM”

A wiki is the format that serves a language model and a human at the same time. The pages are plain markdown with named titles, explicit cross-links, typed relationships, and a note on each claim’s source and confidence. A model reads that natively and can follow the links; you read the exact same file and can edit it.

That is the opposite of a typical RAG store, where your documents are chopped into embedding chunks you cannot read, retrieval returns fragments by fuzzy similarity, and there is nothing to open, correct, or trust. With a wiki you can structure any kind of knowledge for a model while keeping it fully human-readable. See Why a wiki, not a RAG for the full argument.

What it does

Section titled “What it does”
  • Ingests: scans folders you nominate as entry points, hashes new and changed files, classifies them by source type, extracts the durable knowledge, and writes it into your knowledge folders. Each claim carries a provenance marker; each page carries a confidence score
  • Maintains: cross-links pages, surfaces orphans and broken links, flags stale and low-confidence content, tracks each page from draft to reviewed to verified, and proposes archiving projects that have gone quiet
  • Answers and updates: answers questions using only the wiki, folds in targeted updates from a URL or free text, captures the durable parts of the current conversation, and runs web research that gets distilled back into pages

Quick start

Section titled “Quick start”
Terminal window
# In Claude Code CLI
/plugin marketplace add giovi321/obsidian-wiki
/plugin install obsidian-wiki

Then register a wiki and run the daily loop:

/setup-wiki                       # interview: name, root folder, entry points
/ingest <slug> some-file.md       # distil a source into pages
/query <slug> "what did I save about X?"

Full walkthrough: Install in Claude Code and First run.

What it is not

Section titled “What it is not”
  • Not a chat-history dump. Conversation sources score 0.3 by default and are filtered hard; verbatim assistant output is never written
  • Not autonomous. Every command is invoked by you. There is no background indexing, no watcher, no scheduled task
  • Not a replacement for Obsidian. Output is plain markdown with wikilinks and frontmatter on disk; you keep using Obsidian to read and navigate

Where to go next

Section titled “Where to go next”