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
import type { Event } from "../gen/Event";
import type { SystemSnapshot } from "../gen/SystemSnapshot";
export type StreamFrame =
| { kind: "chunk"; data: string }
| { kind: "done" }
| { kind: "error"; code: string | null; message: string };
export type EventFrame =
| { kind: "snapshot"; snapshot: SystemSnapshot }
| { kind: "event"; event: Event }
| { kind: "status"; state: "live" | "reconnecting" | "resyncing" };
export interface ApiErrorShape {
status: number | null;
code: string | null;
message: string;
}
export class ApiError extends Error implements ApiErrorShape {
status: number | null;
code: string | null;
constructor(status: number | null, code: string | null, message: string) {
super(message);
this.name = "ApiError";
this.status = status;
this.code = code;
}
}
export function toApiError(raw: string): ApiError {
try {
const o = JSON.parse(raw) as { status?: number | null; code?: string | null; message?: string };
if (o && typeof o === "object" && "message" in o) {
return new ApiError(o.status ?? null, o.code ?? null, o.message ?? raw);
}
} catch {
/* non-JSON errors are passed through as-is */
}
return new ApiError(null, null, raw);
}