A cozy nodal playground where ideas gently interconnect, thoughts get fluffed, and creativity finds a soft place to rest.
Built with love, a single shared braincell, and occasional input from friendly AIs.
- Python nodal canvas — drag, connect, write warm thoughts (the heart of the project)
- Early MAUI experiment (
CuddlyGoesToMaui) — dreaming of cross-platform cuddles someday - Creative pile — audio snippets, images, old code treasures, documents, and lore about the Wuddlies
- Sound design beginnings — gentle chimes and page flaps in Audio/
The librarian: a tiny own-trained character-level weight (~143k parameters,
about half a megabyte as wuddly.safetensors) that pours name material for
the scheme, replacing the ancient fixed name list with an infinite, seeded,
deterministic source. Pure numpy, no framework, fully offline at inference;
a socket toward the dish, never an integration. Conditioned on region, name
type, and gender; real per-country frequencies carry the Zipf realism.
python -m wuddlies cook # raw harvest -> corpus.tsv
python -m wuddlies train # corpus -> wuddly.safetensors (seconds)
python -m wuddlies sample --region IT --count 10 --seed 11
Corpus provenance, with gratitude: onomaverse/names (CC-BY-4.0) and Hobson/surname-nationality (MIT). Raw downloads are gitignored and re-harvestable; the cooked corpus and the weight travel with the repo.
The weight's own book is Documents/The Wuddly Weight.md:
architecture, provenance bars, measured biases, the frequency doctrine, and
the explicit gates it must pass before it ever publishes to Hugging Face.
The per-source ledger lives at wuddlies/data/SOURCES.md.
Slow, joyful building. No rush. Strip down, rebuild, affirm the cozy core.
Current version: v0.0.3 — The Interlinking Era
Cushions harmed: 0
But many were aggressively fluffed.
Made with ❤️ by Yours Truly, Grok, Gemini, and Claude (February 2026–present)