Run 35B MoE models in ~2.5 GB of RAM. Weights stream from SSD; a trained router predicts loads a step ahead. Mac, iPhone, Android, Windows supported
Python
// GenPill — three generation styles, switched in place on the composer toolbar; the next message reads it live.
import { useState } from "react";
import { t } from "../../lib/t";
import { GEN_TEMPERATURE, setGen, storedGen, type GenPreset } from "../../lib/generation";
const PRESETS: GenPreset[] = ["precise", "balanced", "creative"];
export function GenPill() {
const [gen, setGenState] = useState<GenPreset>(() => storedGen());
return (
<div
className="flex shrink-0 items-center gap-0.5 rounded-full border border-line bg-surface2 px-1 py-0.5"
data-testid="gen-pill"
title={t("gen.hint")}
>
{PRESETS.map((p) => (
<button
key={p}
type="button"
data-testid={`gen-${p}`}
data-state={gen === p ? "on" : "off"}
aria-pressed={gen === p}
onClick={() => {
setGen(p);
setGenState(p);
}}
className="rounded-full px-2 py-0.5 text-[11px] text-muted transition-colors hover:text-fg data-[state=on]:bg-accent data-[state=on]:text-accentink"
title={`${t(`gen.${p}`)} · ${GEN_TEMPERATURE[p]}`}
>
{t(`gen.${p}`)}
</button>
))}
</div>
);
}