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
export interface Bucket<T> {
key: "today" | "yesterday" | "week" | "earlier";
items: T[];
}
const DAY_MS = 24 * 60 * 60 * 1000;
function localDayStart(ms: number): number {
const d = new Date(ms);
d.setHours(0, 0, 0, 0);
return d.getTime();
}
export function bucketThreads<T extends { updatedAt: number }>(
threads: T[],
nowMs: number,
): Bucket<T>[] {
const today = localDayStart(nowMs);
const buckets: Record<Bucket<T>["key"], T[]> = { today: [], yesterday: [], week: [], earlier: [] };
for (const th of threads) {
const day = localDayStart(th.updatedAt);
if (day >= today) buckets.today.push(th);
else if (day >= today - DAY_MS) buckets.yesterday.push(th);
else if (day >= today - 6 * DAY_MS) buckets.week.push(th);
else buckets.earlier.push(th);
}
return (["today", "yesterday", "week", "earlier"] as const)
.filter((k) => buckets[k].length > 0)
.map((k) => ({ key: k, items: buckets[k] }));
}