A radical-privacy AI preview from Estonia: Savant promises no logs and no KYC

Tallinn-based Savant has grabbed attention with a preview that pushes privacy in AI services to the extreme: its site promises a chat service with no accounts, no logs and no KYC. The announcement passed 400,000 views on X. An important caveat: the site itself says it is "not live yet" and that nothing is running in production; the company is currently fundraising.

The model family is named after Estonian weather: Lumi (snow) is light and fast; Udu (fog) is the balanced one for code, analysis and long conversations; Torm (storm) is the flagship for deep reasoning and agents. The real pitch is the architecture: queries are layer-encrypted on the device, a third-party relay sees only your IP, and Savant's hardware produces the answer in memory and discards everything. The company says it runs its own servers rather than rented cloud, with AMD EPYC processors and liquid-cooled AMD GPUs booting Libreboot firmware. Accounts are random 16-digit numbers; payments range from Monero to plain cards.

The other side of the coin is worth stating: an AI service that keeps no records and asks no identity is a dream for privacy advocates and a fresh headache candidate for regulators and abuse-watchers. If the service launches as promised, that tension will sit at the center of the privacy-versus-safety debate.