A Windows app that runs large AI models on an ordinary desktop, on your own machine. The headline result is Llama 3.1 70B at full precision, a 141 GB model, on a PC with one 16 GB graphics card.
No. The model runs on your machine. Your prompts and answers stay there.
At full precision, every token RS-Molt writes matches a plain run of the same model. Same model, same answers, word for word. The smaller sizes (Q8, Q4, Q2) trade some accuracy for speed, and the results page shows how often each one agrees with full precision.
Ryzen 7 9700X, RTX 4070 Ti Super 16 GB, 32 GB DDR5-6400, and three NVMe drives: a Samsung 9100 PRO, a Samsung 970 EVO and a Crucial P3. The full list is on the results page.
Estimated delivery is April 2027. Founders get early access builds before that, posted in the feed as they land.
Windows first. Linux is a stretch goal.
No. Results are published with receipts; how they are achieved stays in-house. Achieved using capabilities from our other projects; not part of this campaign.
Pledges open at launch. The payment and refund terms are published on this page before the first pledge is taken.
So the results, the builds, the feed and the live demo can sit right next to it, with no limits on how much of the work we can show.
Zac Henderson, who builds RS-Molt under Raven Sym LLC, a Wyoming company.
