April 10, 2026
Hype or Skynet?
Show HN: QVAC SDK, a universal JavaScript SDK for building local AI applications
Skynet jokes, privacy panic: devs react to a build-anywhere AI kit
TLDR: QVAC launched an open‑source toolkit to build AI apps that run on your own devices with peer‑to‑peer sharing. Comments split between privacy hawks demanding strict permission and zero leaks, and builders cheering the possibilities, cracking Skynet jokes, and asking one thing—when’s the hackathon?
“Show HN” just dropped a bomb: QVAC released an open‑source toolkit to build AI apps that run on your own device — phone, laptop, even servers. It promises chatbots, speech, translation, vision, and peer‑to‑peer model sharing (think BitTorrent for AI), plus plugins and hefty docs. Links: SDK • docs. But the comment section? It’s the main event. One user laser‑locks on copy that read like “without permission on any device,” triggering a privacy flare‑up. The vibe: give us hard opt‑in and clear boundaries, not stealth processes. A QVAC team member slides in with a calm “happy to answer questions,” while the crowd debates control vs convenience. On the comic‑relief flank, a skeptic asks if this is for building “Skynet,” “AI crypto schemes,” or just a basic local chat — a joke that lands because the ambition is huge. Another veteran says the real boss fight isn’t speed anymore; it’s keeping sensitive company code from leaking, even by accident. That anxiety overshadowed the tech stack talk. Meanwhile, builders are itching to ship — “Hackathon when?” sets the tempo — even as launch notes admit rough edges like big bundles and a clunky plugin flow. Verdict: massive promise, bigger questions, and a comments section doing QA with memes.
Key Points
- •QVAC launched an open-source JavaScript/TypeScript SDK for building local-first AI applications across desktop, mobile, and servers.
- •The SDK is built on QVAC Fabric and runs on Bare, a cross-platform JavaScript runtime within the Pear ecosystem.
- •It supports local inference for multiple model types (LLMs, OCR, translation, transcription, text-to-speech, vision).
- •Models can be distributed peer-to-peer over the Holepunch stack, similar to BitTorrent, with a plugin-based architecture for extensibility.
- •Known issues include larger-than-desired bundle sizes, a need for simpler plugin workflow, and a CLI-dependent tree-shaking process.