AKArtem Kunykininfinitemonkey.hashnode.dev·May 9 · 3 min readSomething in the way, mmm...Notes from the middle of something pivotal I'm not writing this to get followers or build a personal brand. I'm writing it so that when this moment is behind me, I can come back and read what it actua00
AKArtem Kunykininfinitemonkey.hashnode.dev·Apr 23 · 17 min readLunapark: Building a Minimalistic Rust Agent on Qwen 3.6Local AI agents are exploding across the developer ecosystem. Every week a new framework appears promising to simplify agent construction, and every week it ships with hundreds of transitive dependencies, a Python runtime, and startup times measured ...00
AKArtem Kunykininfinitemonkey.hashnode.dev·Apr 10 · 17 min readBuilding Lunapark: A Local Autonomous Agent on 7.3 MB and Zero API CostsIt was late February 2026. A $2.18 charge appeared on the Azure billing dashboard — overnight LLM API calls from an experimental autonomous coding agent. The amount was irrelevant. The thought that followed was not: "This agent should own its own co...00
AKArtem Kunykininfinitemonkey.hashnode.dev·Apr 5 · 6 min readReasoning Models Don't Fail at Reasoning: The Protocol Layer Is What Kills Local AgentsReasoning Models Don't Fail at Reasoning: The Protocol Layer Is What Kills Local Agents April 5, 2026 · Artem The narrative around local LLMs stops at the model checkpoint. It never discusses what breaks between the model and the tool call. I ran mi...00
AKArtem Kunykininfinitemonkey.hashnode.dev·Apr 3 · 8 min readTurboQuant Is Not a Free Lunch: What the RTX 3060 Actually ReportedTurboQuant Is Not a Free Lunch: What the RTX 3060 Actually Reported April 3, 2026 · Artem The ternary quantization narrative is being sold as a compression silver bullet. It isn't. On a 12 GB consumer GPU running Qwen-family models, a plain q8_0 GGU...00
