Personal Knowledge Compiler

Compile Your Knowledge.
Let AI Maintain the Wiki.

A personal knowledge base that grows smarter over time. Feed it articles, the LLM reads, distills, cross-references, and maintains the wiki. Knowledge compounds — not just retrieved.

Get Started View on GitHub
raw/
Your articles
→
LLM
Reads & distills
→
wiki/
Structured knowledge
→
∞
Compounds forever
Core Operations

Five Commands That Do Everything

No complex UI. Five slash commands power the entire knowledge compilation pipeline.

1
/init

Initialize Wiki Structure

One command creates the full wiki skeleton—directories, CLAUDE.md schema, indexes, and logs. Plugin-native, zero LLM dependency, completes in milliseconds.

2
/ingest

Smart Source Ingestion

LLM reads your article, creates summary pages, extracts concepts and entities, adds cross-references, checks for contradictions, updates indexes—all in one go.

3
/query

Wiki-Based Q&A

Ask questions grounded in your wiki. LLM reads the index, finds relevant pages, synthesizes answers with wikilink references. Good answers get archived as new wiki pages.

4
/lint

Health Check

Periodic wiki checkup: detects contradictions between pages, orphan pages with no links, stale content, missing attribution, and knowledge gaps. Can auto-fix most issues.

5
/scan

Legacy Library Scan

Hundreds of old notes? A lightweight scan reads only titles and first 10 lines of each file, generating an index map. Then selectively ingest what matters.

6
tag audit

Tag Management

AI-powered tag auditing with merge suggestions. Clean up tag sprawl, consolidate duplicates, and maintain a coherent taxonomy across your entire wiki.

Workflow

How It Works

Five steps from raw articles to a living, growing personal knowledge base.

  1. 1
    /init

    Initialize the Compiler

    The plugin creates the full wiki skeleton on your vault—directories, the CLAUDE.md schema, index files, and operation logs. No LLM involved. Instant.

  2. 2
    Drop files in raw/

    Feed It Source Material

    Put articles, papers, book notes, or web clippings into the raw/ directory. Don't worry about organization—the LLM auto-categorizes during ingest.

  3. 3
    /ingest

    Compile Into Wiki Entries

    The LLM reads each article, distills key points, and creates structured wiki pages with cross-references and source attribution. One ingest can touch 10–15 pages.

  4. 4
    /query

    Query Your Growing Knowledge

    Ask questions. The LLM consults the wiki index, reads relevant pages, and synthesizes answers with traceable references. Archive great answers as new wiki pages.

  5. 5
    /lint

    Maintain Wiki Health

    Run periodic health checks. The LLM finds and fixes contradictions, orphan pages, stale content, missing links, and knowledge gaps. Your wiki stays coherent as it grows.

Why Wiki

Wiki Is Not RAG

Two fundamentally different approaches. One resets every query. The other compounds.

Traditional Approach

RAG: Retrieve & Forget

Upload docs → LLM retrieves fragments → assembles answer from scratch → knowledge not retained
  • Knowledge doesn't accumulate between queries
  • Each question re-discovers what was already found
  • No cross-references between documents
  • Contradictions go undetected
  • Quality degrades as document volume grows
vs
The Wiki Way

Wiki: Compile & Compound

Source files → LLM reads once, distills → updates entity/concept pages → knowledge compounds forever
  • Knowledge builds on previous knowledge
  • Each ingest enriches the entire wiki network
  • Automatic cross-references between all pages
  • Contradictions detected and flagged
  • Quality improves with every new source added
Architecture

Four Directories, Four Roles

Clear ownership boundaries prevent wiki decay. Every directory has exactly one writer.

Vault/
├── raw/ # Your source material — immutable, LLM reads only
│ ├── tech/ # Tech articles, papers, tutorials
│ ├── work/ # Work-related documents
│ ├── reading/ # Book notes, podcasts
│ └── general/ # Everything else
├── wiki/ # LLM-maintained knowledge base
│ ├── summaries/ # One summary per source file
│ ├── concepts/ # Cross-source concept synthesis
│ ├── entities/ # People, tools, frameworks
│ ├── methods/ # Reusable how-to guides
│ ├── comparisons/ # Side-by-side analysis
│ └── analysis/ # Deep dives from archived queries
├── drafts/ # Your private scratch space — LLM never touches
├── legacy/ # Frozen archive of old notes — read-only
├── CLAUDE.md # LLM operation schema — the real core logic
├── index.md # Master wiki index
└── log.md # Operation timeline
Ownership Rules
DirectoryWho WritesWho ReadsPurpose
raw/ You LLM (read-only) Immutable source material
wiki/ LLM only Both Structured knowledge base
drafts/ You only You only Private scratch space
legacy/ Nobody (frozen) Both (read-only) Archived old notes
Two Flavors

Plugin or Skill—Your Choice

Same wiki engine, two different interfaces. Pick what fits your workflow.

Obsidian Users

Obsidian Plugin

A unified Wiki panel + Chat panel inside Obsidian. File picker, status badges, one-click init. Communicates with Claude Code or Cursor via ACP.

  • ✓ Graphical interface inside Obsidian
  • ✓ File picker for selecting sources
  • ✓ Cmd+Shift+C to open chat instantly
CLI / Any Editor

Claude Code Skill

A Claude Code skill that works in any directory. No Obsidian required. Initialize a wiki, ingest sources, query, lint—all from the terminal.

  • ✓ Zero dependencies beyond Claude Code
  • ✓ Works in any directory, any editor
  • ✓ Perfect for batch processing

Ready to Compile Your Knowledge?

Start with an empty vault. End with a living, growing personal wiki that compounds over time.

Read the Docs Star on GitHub