obsidian-wiki

obsidian-wiki

A plugin for Claude Code that turns one or more Obsidian folders into structured knowledge bases. Drop sources into entry-point folders; the agent classifies them, distills the durable content into wiki pages with provenance and confidence tracking, maintains cross-links, and keeps a manifest so reruns are incremental.

Supports any number of wikis. Each wiki has its own configuration in a CLAUDE.md file at its root and is registered globally so commands can address it by slug.

Quick start

Five steps. Five minutes if you already use Obsidian.

1. Install the plugin. In Cowork: Customize → Personal plugins → Browse plugins → paste giovi321/obsidian-wiki → install. In Claude Code CLI: /plugin marketplace add giovi321/obsidian-wiki then /plugin install obsidian-wiki.

2. Install the required Obsidian community plugins: Dataview, Tasks, Periodic Notes, Front Matter Timestamps, Folder Notes. See Obsidian plugins required for the full list.

3. Register your wiki: in a Claude chat, run /setup-wiki. The interview asks for the wiki name, root folder (point at any existing Obsidian folder, or a new empty one), which entry points to enable (quick-notes, articles, voice transcripts, LLM conversations), which structured-knowledge folders to enable (projects, documentation, resources, people, concepts), and a few thresholds. Pick a short slug (1 to 4 characters); you'll type it as the first argument on every command.

4. Drop a source file into one of the entry points the setup created. A markdown note, a PDF, an article you saved, anything. Then run /ingest <slug>. The agent reads the file, distills the durable content into one or more wiki pages, and moves the source to _service/entry-points/<entry-point>/<YYYY-MM>/. Check the report it prints.

5. Ask the wiki a question: /query <slug> "what did I save about X?". The agent answers using only the wiki contents, citing the pages it pulled from.

That's the daily loop. Add more sources, rerun /ingest <slug>. Run /lint <slug> weekly to surface orphans and broken links. Run /upgrade after a plugin update to refresh the boilerplate. Everything else is layered on top.

Table of contents

• Quick start
• What it does
• What it is not
• Architecture
• Folder structure
• Concepts in plain English
• Install in Claude Code
• Install in Cowork
• Obsidian plugins required
• First-run setup
• Ideal workflow
• Addressing two or more wikis
• The 20 commands
• Command reference
• How ingest works on one source
• Page lifecycle
• Confidence scoring
• Source quality buckets
• Per-operation confidence defaults
• Provenance
• Standard page frontmatter
• Entry-point schema
• Source ID canonicalization
• Manifest schema
• Registry schema
• Log format
• Feedback loop
• Retrieval cost escalation
• Modes of operation
• Customization
• Shared docs
• What happens when the plugin updates
• Custom procedures
• Visibility tags
• Adding a custom command
• Removing a wiki
• FAQ
• License

What it does

Three things, in this order.

1. Ingests: scans folders you nominate as entry points, hashes new and changed files, classifies them by source type, extracts durable knowledge, and writes it into structured-knowledge folders. Each fact carries a provenance marker (extracted, inferred, ambiguous). Each page carries a base confidence computed from the count and quality of its sources.

2. Maintains: cross-links pages, surfaces orphans and broken links, flags stale and low-confidence content, tracks page lifecycle (draft to reviewed to verified), and proposes project archival when activity drops below thresholds.

3. Answers and updates: answers questions using only the wiki, integrates 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.

What it is not

A few clarifications about scope.