Daily issue / 2026-06-04
Useful things that surfaced in Hacker News discussions on 2026-06-04. Open the context to see why each one caught our attention—or jump straight to the original thread.
What we found
EU-based search engine with image search, URL rewrites, and API access.
Source discussion: Show HN: Uruky (EU-based Kagi alternative) now has Image Search and URL Rewrites
Editorial paraphrase
HN discussion highlighted image search, URL rewrites, a 2h free trial, and integrations like Mojeek, Marginalia, EUSP, Linkup, and Serper.
Read the original threadOpen-source harness for AI-assisted vulnerability discovery.
Source discussion: Anthropic's open-source framework for AI-powered vulnerability discovery
Editorial paraphrase
Commenters framed security as a strong LLM use case and suggested adapting the harness to each team’s workflow instead of using it unchanged.
Read the original threadNative vLLM backend for KV-cache quantization.
Source discussion: KVarN: Native vLLM backend for KV-cache quantization by Huawei
Editorial paraphrase
Commenters debated the benchmark claims, concluded it is faster than FP16 rather than higher quality, and said a vLLM PR should be feasible.
Read the original threadFrontend tooling stack that commenters treated as the best part of the VoidZero portfolio.
Source discussion: VoidZero Is Joining Cloudflare
Editorial paraphrase
Comments said Vite made webpack-era tooling pain disappear and repeatedly referenced the broader stack—Vitest, Rolldown, Oxc, and Vite+.
Read the original threadCompact local model commenters recommended for laptops with limited VRAM.
Source discussion: Gemma 4 12B: A unified, encoder-free multimodal model
Editorial paraphrase
One reply called Qwen 3.5 9B the undisputed champ for 16GB laptops and compared it favorably against other small models.
Read the original threadOpen-weight OpenAI model aimed at low-latency local or specialized use.
Source discussion: Gemma 4 12B: A unified, encoder-free multimodal model
Editorial paraphrase
Commenters grouped gpt-oss-20b with Qwen 3.5 9B and other small models as a local-inference candidate.
Read the original thread