Little paws.
Lasting friendship.
Raise a creature. Teach a real neural brain.
Your pet, its adventures, and your lessons stay with you.
EST. TODAY♧Come back. Grow together.
Pip Hatchling
Day 1 together
Small paws. Big ambitions.
Click your creature in the garden for a cuddle. No pet disappears because you were away.
A moment with Pip
LOCAL & PRIVATEA small command vocabulary with replies from your pet’s saved state. Navigation uses its neural brain; dialogue uses game rules.
Our little story
Which way is Pip leaning?
The network's action probabilities at this moment.
Your pet sees nearby tiles and the berry's direction.
The rest of the garden is outside its observation.
Every little step leaves a trace.
A clue fades.
Does the memory stay?
A blue flower means the left berry. Gold means right. The clue disappears after one tick, and a gate makes you wait. Can a trained neural memory carry it to the fork?
Loading the memory checkpoints…
The gate prevents using your starting position as a memory trick. Both berries look alike.
Live hidden values, not a decoded thought or personality.
The part a small brain can keep
Loading measured results…
Read the protocol, controls & per-seed results ↗Controls use the same recurrent weights on the standard test.
36 unique expert episodes. Three training seeds. 18 standard and 12 longer-delay test cases per seed; test layouts are disjoint from training. Same data and 200 updates, with different compute costs. The frozen current brain is an out-of-task reference. Grove rounds start fresh; their completed adventures enter your pet’s diary.
Does a little guidance
go a long way?
We trained the same small network with and without demonstrations, then tested it on held-out garden seeds. Inspect the original results and the stricter world-overlap audit below.
Learning to find the berry
Development gardensA fresh patch
Goal successMean ± standard deviation across 3 training seeds.
Learning is a little messy.
That's the interesting part.