🐶 Good Boy · teach two dogs your own words

Say something to both dogs. Each dog's trick is one live call to the Jepela API with the same words. Give each a treat 🦴 or a “no” ✋. Biscuit has no memory, so he forgets at once. Mochi has a Jepela memory: a treat writes one line, a “no” writes a rule checked in code. Try your own made-up words, like “banana!”: say them, reward the right trick, and say them again.

Biscuit · no memory

Mochi · Jepela memory

The exam every word Mochi was rewarded for, asked again to both dogs

Reward Mochi for a trick first.

Biscuit, no memoryMochi, Jepela memory

Mochi's memory lines and rules

Mochi's last API call, word for word

No call yet.

Model jepela-english. Nothing is scripted: both dogs ask the same engine the same question, and only Mochi names a memory. Measured on 2026-09-28: words the engine knows are understood without any memory (“sit”, “give me your paw”, “roll over”, “twirl”: all at 100%). Made-up words are guesses without memory (“banana!” became sit at 23%, “abracadabra” fetch at 20%). After one rewarded line each, Mochi got 4 of 4 made-up words right at 95 to 99%, with all four lines in memory at once.

Engine answering · API health check, 09:07 UTC