LLM Wiki Reader

The fastest, lightest native desktop reader for Karpathy-style LLM wikis.

Built for people (and agent swarms) who live in large, richly interlinked Markdown wikis and want a snappy, opinionated reading experience without the weight of a full note-taking app.

What is this?

If you (or your agents) maintain a Karpathy-style structured LLM wiki — hundreds of interlinked Markdown files with YAML frontmatter, [[wikilinks]], "Top" summaries, and a living raw/ or concepts/ structure — then Obsidian starts to feel heavy.

LLM Wiki Reader is the purpose-built alternative:

• Blazing fast (entire wiki loaded into memory in < 100 ms)
• Read-only by design
• Beautiful typography and "Top" callouts
• Deep support for the exact link patterns your agents produce
• A first-class human feedback loop back into the wiki

It is the reading companion for the age of agent-written knowledge bases.

Standout Features (v0.1.0)

• Instant everything — Full-text search, sidebar filtering, and page loading feel instantaneous.
• Perfect wikilink resolution — Handles aliases ([[slug|Display Text]]), deeply nested SKILL.md files, and legacy relative Markdown links.
• Organized navigation — Pages are automatically grouped by type (Entities, Projects, Concepts, Comparisons, Queries…) with collapsible sections and noise filtering (raw/, _archive/, etc.).
• In-page Table of Contents — Headings are extracted and become a live, clickable mini-map with smooth scrolling.
• Multi-vault support — Remembers your last wiki + a short list of recent ones. Switch between several wikis instantly without restarting.
• Curator Queue (the killer feature) — A dedicated right pane for human feedback. Drop suggestions, corrections, or new ideas per page. They are saved as structured Markdown files (raw/suggestions/) that your agent pipelines can later consume. This is the missing human-in-the-loop for living LLM wikis.
• Backlinks & outbound links — See the full graph neighborhood of the current page.
• "Open in editor" — One-click jump from any page to the real source file in your preferred editor.
• Native & beautiful — Dark theme, excellent typography, proper tables and code blocks. Feels like a desktop app, not a web view.

Why not just use Obsidian?

Obsidian is incredible when you are writing or using dozens of plugins.

LLM Wiki Reader exists for the other 80 % of the time — when you (or a reviewer) just want to read, follow links, search, and give lightweight feedback as fast as possible.

It is dramatically lighter, faster, and more focused.

Installation

All official releases are available on the Releases page.

Windows (recommended)

1. Download the .msi (preferred).
2. Run it — it will install like any normal Windows application.
3. The app appears in Start Menu as LLM Wiki Reader.
4. First run: it automatically looks for your wiki at C:\Users\YourName\wiki. Use the "Choose folder…" button if your wiki lives elsewhere.

Note: Windows may show a SmartScreen warning on first run (common for new apps). Click "More info" → "Run anyway".

Linux

We publish two packages:

• AppImage — Universal, portable (works on most distros without installation)
• .deb — Native Debian/Ubuntu package (recommended if you're on Ubuntu, Pop!_OS, Mint, etc.)

Download from Releases.

For the AppImage:

chmod +x LLM-Wiki-Reader-*.AppImage
./LLM-Wiki-Reader-*.AppImage

For the .deb:

sudo dpkg -i LLM-Wiki-Reader_*.deb

First run auto-detects ~/wiki.

macOS

Best-effort / unsigned universal build

We produce one universal DMG (works on both Intel Macs and Apple Silicon) on every tagged release.

• Download LLM-Wiki-Reader-macOS-universal.dmg
• Open it and drag LLM Wiki Reader into your Applications folder.
• On first launch, right-click the app → Open (or go to System Settings → Privacy & Security and allow it).

This is unsigned for v0.1.0. We plan to add proper code signing + notarization in a future release.

Build from source

See:

• WINDOWS.md — detailed Windows instructions
• The GitHub Actions workflow (.github/workflows/build.yml) for the exact CI commands used for official builds

You need:

• Rust (stable)
• Node.js 20+
• pnpm
• Platform-specific C++ build tools (Windows) or Xcode command line tools (macOS)

The Curator Feedback Loop

One of the most powerful patterns with agent-driven wikis is the human ↔ agent loop.

LLM Wiki Reader makes the human side trivial:

1. While reading any page you can open the right-hand Curator Queue pane.
2. Choose a type (suggestion, correction, addition…).
3. Write your feedback.
4. It is written into your wiki as a timestamped, frontmatter-rich file under raw/suggestions/.

Your agent pipeline (or a human curator) can then pick these up on the next run, incorporate the knowledge, and close the loop.

This is the feature that makes LLM Wiki Reader feel alive even though it is read-only.

Roadmap

• Live file watcher + instant reload when agents write
• Command palette (⌘K / Ctrl+K)
• Lightweight graph view
• macOS code signing + notarization (currently unsigned universal DMG)
• Better query / Dataview-style views

Development

This is a Tauri v2 + Svelte 5 + Rust application.

The high-performance wiki parser and indexer live in the reusable wiki-core/ crate.

Releases are built automatically via the GitHub Actions workflow in .github/workflows/build.yml.

License

MIT

Made with ❤️ for people (and agents) who actually read their wikis.

If this is useful to you (or your agent swarm), consider starring the repo — it helps me understand if there's real interest in a focused wiki reader.

Feedback, ideas, and bug reports are very welcome.