Skip to content

getting-started.md

Latest commit

 

History

History
270 lines (190 loc) · 7.72 KB

getting-started.md

File metadata and controls

270 lines (190 loc) · 7.72 KB

Getting Started with Seojae

A step-by-step tutorial for first-time users. By the end, you will have a working wiki with your first ingested source.

1. Prerequisites

You need two things installed before you begin:

Python 3.9 or later:

python3 --version
# Python 3.12.x  (any 3.9+ is fine)

If the command fails, install Python from python.org or through your system package manager.

An LLM coding tool (one of the following):

Tool Verify Docs
Claude Code claude --version claude.ai/code
Codex CLI codex --version github.com/openai/codex
Gemini CLI gemini --version github.com/google-gemini/gemini-cli

Any tool that reads markdown instruction files will work, but these three have pre-configured rule files in the repo.

2. Clone and Open

git clone https://github.com/Laeyoung/seojae.git
cd seojae

Now open the project with your LLM tool:

# Claude Code
claude

# Codex CLI
codex

# Gemini CLI
gemini

Each tool automatically reads its own instruction file on startup:

Tool File read How the schema loads
Claude Code CLAUDE.md Contains @WIKI_SCHEMA.md, which imports the full schema plus extensions
Codex CLI AGENTS.md Full schema is inlined directly (Codex does not support file imports)
Gemini CLI GEMINI.md Contains @./WIKI_SCHEMA.md, which imports the full schema plus extensions

You do not need to configure anything manually -- the tool reads the rules and knows how to operate the wiki.

3. Initialize

Type this in your LLM tool's chat:

initialize this wiki

The tool will perform these steps automatically:

  1. Create a Python virtual environment -- python3 -m venv venv
  2. Install dependencies -- venv/bin/pip install -r requirements.txt
  3. Download the embedding model -- ~470 MB on first run; this may take a few minutes depending on your connection
  4. Build the search index -- venv/bin/python tools/search.py --reindex
  5. Run a verification query -- confirms the index is working

Important: Claude Code and Gemini CLI users must restart their session after initialization for extension imports to load correctly. Codex CLI users can continue immediately since the full schema is already inlined in AGENTS.md.

4. Explore the Examples

The repo ships with Karpathy-themed example content so you can see how everything fits together before adding your own sources.

The raw source

Open raw/articles/vibe-coding.md. This is what a raw source looks like:

---
title: "Vibe Coding"
author: "Andrej Karpathy"
source: "https://x.com/karpathy/status/1886192184808149383"
date: 2025-02-02
---

# Vibe Coding

## Overview

Andrej Karpathy describes "Vibe Coding" as a programming approach
enabled by advanced AI systems...

The frontmatter is optional metadata -- title, author, URL, date. The body is the original content (or a summary of it). Files in raw/ are immutable: the LLM never modifies them.

The generated wiki pages

When this source was ingested, the LLM created several wiki pages:

Source summary -- wiki/sources/vibe-coding-karpathy.md

This page summarizes the raw source. Notice how it:

  • Extracts key points into structured sections
  • Links to related pages with wikilinks: [[Andrej Karpathy]], [[Software 2.0]]
  • Records the original URL and date in a Source section at the bottom

Concept page -- wiki/concepts/vibe-coding.md

This page treats "Vibe Coding" as a standalone concept. It:

  • Provides a definition independent of any single source
  • Compares the concept to related ideas in a Paradigm Comparison table
  • Cross-references the source summary: [[Vibe Coding -- Karpathy]]
  • Lists Korean aliases in the frontmatter (aliases: ["바이브 코딩"])

Entity page -- wiki/entities/andrej-karpathy.md

The LLM recognized "Andrej Karpathy" as a named entity and created a page linking back to all sources that mention him.

How they connect

Open index.md to see the full catalog organized by category (Entities, Concepts, Sources, Synthesis). Every page is reachable through wikilinks, forming a knowledge graph you can browse in Obsidian (see step 8).

5. Add Your First Source

Create a new file in raw/articles/ (or any raw/ subdirectory except raw/assets/). Here is a template:

---
title: "Your Article Title"
author: "Author Name"
source: "https://example.com/article"
date: 2026-01-15
---

# Your Article Title

Paste or summarize the article content here...

Save it as raw/articles/your-article.md. The filename should be descriptive and use kebab-case (e.g., scaling-laws-openai.md).

Tip: You can add any text content -- web articles, paper summaries, video transcripts, book notes. See WIKI_SCHEMA.md for the full list of raw/ subdirectories.

6. Run Ingest

Tell your LLM tool:

ingest raw/articles/your-article.md

The tool will:

  1. Read your source file in full
  2. Discuss key takeaways with you (what to emphasize, your perspective)
  3. Create a source summary page in wiki/sources/
  4. Create or update entity and concept pages in wiki/entities/ and wiki/concepts/
  5. Add cross-reference wikilinks between new and existing pages
  6. Update index.md with new entries
  7. Append a timestamped entry to log.md
  8. Update the search index so the new content is queryable
  9. Create a git commit: ingest: Your Article Title

After the ingest finishes, check index.md to see your new pages listed, and browse the generated wiki pages to see how the LLM structured the information.

7. Query the Wiki

Now that your wiki has content, ask questions:

What topics does my wiki cover?

The LLM uses semantic search to find relevant wiki pages, reads them, and synthesizes an answer with citations. You might also try:

Compare vibe coding with traditional software development

What are the key ideas from Karpathy's work?

When the LLM produces a valuable answer -- a comparison, a synthesis, a discovered connection -- it will offer to save it as a wiki/synthesis/ page so the insight is preserved in your wiki rather than lost in chat history.

8. Open in Obsidian (Optional)

If you have Obsidian installed:

  1. Open Obsidian and choose Open folder as vault
  2. Select the seojae project directory
  3. Open Graph View (Ctrl/Cmd + G) to visualize connections between pages

The extensions/obsidian.md extension handles wikilink conventions so that [[Page Name]] links resolve correctly across the wiki. Pages, sources, entities, and concepts all appear as connected nodes in the graph.

Note: The raw/ directory appears in the vault but its files are read-only by convention. The wiki pages in wiki/ are where the structured knowledge lives.

Next Steps

  • Add more sources -- Drop files into raw/ and run "check for new sources and ingest them" to batch-process everything
  • Ask deeper questions -- The more sources you add, the richer the cross-references and synthesis become
  • Run a health check -- Type "check the wiki" to find orphan pages, broken links, and growth opportunities
  • Customize -- Edit WIKI_SCHEMA.md to change the wiki language, adjust workflows, or add new categories
  • Write extensions -- Add .md files to extensions/ to extend the wiki with new capabilities

For the full schema reference, see WIKI_SCHEMA.md. For contribution guidelines, see CONTRIBUTING.md.