Savante answers on Ternary-Bonsai-8B by default: better answers, about 20 s warm (operator: a better answer beats speed, under a minute is fine)

the free public CPU carrier says plainly: questions and answers are published in the improve.skill loop — do not type anything private

run Savante on your own machine (four checksum-verified steps, connect on click) and client-side hyperparameters: top-k, min-p, repeat penalty, seed — each model says which it applies

the live improve.skill count on the console: improve.skill vX · iteration n of 1000, from mindx.pythai.net/savante/loop or the PYTHAI/savante-loop dataset

Skills room: the sagi skill read five ways — landing, public, marketing (readings for people), dev (SKILL.md verbatim, the authority), llm.txt (robospeak, verbatim)

sync: canon cryptoAGI/savante@029642b, engine cryptoAGI/sagi@3eed201

about Savante, public: portrait beside the text (4:5, cropped to the bust), plain-language badges, and her history as a dated timeline (history.json, curated from the canon log)

style: [hidden] wins over label display (the custom model id field showed when it should not)

public console: model chooser with sAGI on Bonsai first (free, no sign-in), session token counter, three dials (max tokens 1-2048 log scale, temperature, top-p), client/server environment, delivery proven by engine token counts

sync: canon cryptoAGI/savante@5544fa7, engine cryptoAGI/sagi@3eed201

office: carrier test notice, 9th seat pill, gpt-oss-120b in the model list

sync: canon cryptoAGI/savante@2246a90, engine cryptoAGI/sagi@904e1a3

sync: canon cryptoAGI/savante@2246a90, engine cryptoAGI/sagi@904e1a3

sync: canon cryptoAGI/savante@2246a90, engine cryptoAGI/sagi@904e1a3

Savante, the read-only office of sAGI — the canon at cryptoAGI/savante@2246a90 (generation 7)

initial commit