Shadowfetch

Journal beat

Local AI

A thin beat, and the pieces in it are the ones that hold up. Two are open-model releases — a large open push from Moonshot, and a European entrant judged on receipts rather than weights — and the third is the argument underneath both: that a home full of devices which stop working when a remote service does is fragile by design, and that local control is the fix. This is the beat closest to what the studio actually builds, and also the one with the least written about it so far. That gap is worth stating rather than papering over.

These are existing pieces from the Shadowfetch archive, still hosted on the previous site — every link below opens at its original /blog address. Nothing on this page is new writing.

  1. Kimi K3 puts Moonshot’s largest open model push on the web and API

    Moonshot AI says Kimi K3 is a 2.8T, 1M-context multimodal model now live on Kimi.com and API, with full weights promised by July 27.

  2. The Hidden Fragility: Why Your Smart Home Needs Local Control

    Why your smart home's dependence on cloud servers is a security risk, and how to regain control.

  3. Soofi S makes Europe’s open-model race about receipts, not just weights

    A German-backed open model project is notable less for leaderboard claims than for the unusually auditable release package behind them.

Beat assignments on this page are an editorial judgement made by Shadowfetch; they are not part of the original article metadata.