OPEN_SOURCE ↗
REDDIT · REDDIT// 20d agoOPENSOURCE RELEASE
no-magic ships 47 single-file AI algorithms
no-magic is a 47-script, zero-dependency Python collection that reimplements modern AI building blocks from tokenizers and GPT to Flash Attention, DPO, MCTS, and ReAct. It pairs the code with learning paths, Anki cards, an EPUB, and translations so it feels like a curriculum, not just a repo.
// ANALYSIS
This is less a repo than a full AI textbook disguised as code, and the constraint system is the real product: one file, stdlib only, and commentary that explains the math instead of hiding it.
- –The single-file rule removes framework noise, so readers can see the algorithmic mechanics without jumping across helpers or dependencies.
- –The coverage is unusually broad, spanning foundations, alignment, systems, and agents, which makes it useful as a map of the modern AI stack, not just a grab bag of demos.
- –The learning layer matters: 7 paths, 182 flashcards, challenge prompts, EPUB export, and translations turn the repo into something you can actually teach from.
- –The tradeoff is maintenance at scale; 47 hand-rolled implementations plus docs and translations will only stay trustworthy if the repo keeps getting careful review.
// TAGS
no-magicopen-sourcellmagentreasoningresearch
DISCOVERED
20d ago
2026-03-23
PUBLISHED
20d ago
2026-03-23
RELEVANCE
9/ 10
AUTHOR
tom_mathews