DDenisinpixeloffice.hashnode.dev·Sep 15 · 5 min readHow We Tuned Linux Kernel Sockets & Rescaled an AI Gateway to 16GB in 45 Seconds for 50-Second LLM StreamsMost AI gateway benchmarks test trivial 10-token prompt-response roundtrips. In production, AI agent swarms and compiled backends generate thousands of tokens per session over long-lived streaming con00
DDenisinpixeloffice.hashnode.dev·Sep 4 · 5 min readWhy Modern AI Agents Need Roads, Not Just More Horsepower: Inside PixelRouter v1.4.0Everyone in AI is currently obsessed with horsepower. Every week, frontier labs in San Francisco and Beijing unveil faster, larger, and more capable models — DeepSeek V4, Claude Fable, Gemini 2.5, Ope03A
