A personal RAG knowledge base for the Odin programming language applied to game development (Raylib, Sokol, hot-reload, allocators). Built and curated end-to-end with MiniMax-M3 as the structural engineering agent.

This repo deliberately has no main branch on GitHub - the default branch is public (sanitized subset). The author maintains a parallel main branch on a separate private remote with full content (personal notes, dailies, paywalled scraped sources). That private branch never reaches GitHub.

• For users: git clone defaults to public. That's what you want.
• For contributors: the split rationale is documented in _Helpers/docs/007_mixing_public_and_private_history.md.

OdinRAG aggregates, formats, and indexes learning material about Odin game dev so it can be navigated quickly by either:

• a human (via Obsidian, Grep, or INDEX.md files)
• an AI agent (via the .kilo/skills/ kb-navigator skill and the .kilo/agents/odin-gamedev.md subagent)

The repo ships only the curation workflow (scrapers, agents, skills, indexes, doc metadata). The copyrighted content the scrapers target - Skool courses, blog posts, the Karl Zylinski book - is deliberately not redistributed in this repo and stays in the user's local working directory. See SOURCES.md for how to obtain each source legally on your own.

The full tree lives in _Helpers/docs/001_folder_structure.md. This section gives a one-line summary of each bucket; refer to the structure doc for the complete tree.

• odin-knowledge-base/ - Bucket 1 (public, partly gitignored) - scraped KB + Skool courses
• code/ - public code references (examples, gists, vendored templates, personal projects)
• _Helpers/ - Bucket 2 (public) - scripts, meta docs, internal templates, prompts, logs
• _Private/ - Bucket 3 (gitignored, never pushed) - config, planning, raw notes

odin-knowledge-base/courses/ and odin-knowledge-base/docs/karl_zylinski/odin-book/*.md are gitignored by design - they contain copyrighted content under your own subscription. The repo ships sample articles in odin-knowledge-base/docs/{karl_zylinski,gingerbill,jakubtomsu}/; run the corresponding scrapers to populate the full corpus on your own machine.

git clone https://github.com/LaurentOngaro/OdinRAG.git
cd OdinRAG

# Audit before any push (idempotent, exit 0 = clean)
python _Helpers/scripts/diagnostic/audit_public_safety.py

# Scrape the public sources (no paywall, no auth)
python _Helpers/scripts/scrappers/scrape_official.py
python _Helpers/scripts/scrappers/scrape_zylinski.py

# (Skool requires your own paid membership - see SOURCES.md)
python _Helpers/scripts/scrappers/scrape_skool.py

# Format all .odin + ```odin``` blocks in the working tree
python _Helpers/scripts/fixes/format_odin_in_files.py --path odin-knowledge-base

Requirements: Python 3.10+, pip install requests beautifulsoup4 markdownify lxml.

All machine-specific paths and credentials live in a single gitignored file: _Private/.config/user_config.jsonc.

Setup:

cp _Helpers/templates/user_config.example.jsonc _Private/.config/user_config.jsonc
# edit the copy with your actual paths

The loader _Helpers/scripts/lib/user_config.py exposes them to the scrapers and format_odin_in_files.py. Override any value with an env var (e.g. ODINFMT_EXE, BOOK_HTML_SRC). Keep credentials out of the JSON - use env vars like SKOOL_PASSWORD for secrets.

This whole project - the scraper architecture, the frontmatter schema, the agent prompts, the 6 Kilo skills, the subagent design, the lint configs - was designed and iterated with MiniMax-M3, powered via Kilo Code as the agentic IDE. M3 acts as the structural engineer; I act as the domain curator.

Every file in this repo is the result of a prompt + a code review between us. There is no git blame you can read to tell which line was M3's first draft and which was my edit - that's the point.

For the technical story of how M3 is used in this repo, see _Helpers/docs/002_how_minimax_m3_is_used_in_this_repository.md.

For the social posts around the MiniMax-M3 Showcase Round 2, see _Helpers/docs/social/.

This repository's code (scrapers, agents, skills, scripts, configs) is released under the MIT License - see LICENSE.

The third-party content referenced by the scrapers is not included in this repository and remains the property of its respective authors. See SOURCES.md for attribution and licensing.

If this knowledge base saved you a few hours of Odin exploration, you can help it grow by sponsoring its development. The code is open-source, the scrapers are free, no ads and no tracking - your support is what keeps it that way.

Spread the word at zero cost:

• Star the repo to help others find it: https://github.com/LaurentOngaro/OdinRAG
• Watch for new sources / patterns / scrapers
• Open issues for missing topics or broken scrapers
• Cite it in your blog / Reddit answer / Discord when it helps

Every dollar funds one of these (in order of priority):

1. Scraper maintenance - keeping scrape_skool.py, scrape_official.py, scrape_zylinski.py resilient against site changes, rate limits, and bot detection.
2. New sources - book ISBN API, dod-benchmarks tracker, official newsletter parser, etc.
3. AI integration polish - more Kilo skills, smarter subagent routing, frontmatter improvements for Obsidian / RAGnarok compatibility.
4. Documentation - English translations of French-only docs, code samples, video walkthroughs for the trickier topics (hot-reload, allocator patterns).

Sponsoring is entirely optional. The repo and the KB stay MIT-licensed and free to use regardless.