4 posts
llm A public benchmark showed 92% of coding-agent tokens were served from cache — a 12.5x cost difference. A teardown of oMLX's prefix cache, the six gaps in it, and the three ceilings that decide how many agents one machine can serve.
llm While running opensre (an AI-SRE investigation tool) locally through oMLX — a macOS-native MLX inference server — on two Mac Minis, the same real Kubernetes investigation task was pointed at two different open-weight instruct models in turn. Same oMLX server, same tool schemas, same alert. Only the
agents Continuing the kri fleet assistant story: the bugs that shipped after RAG landed, the agentic design we red-teamed before building, a 6-persona audit that found the control plane was broken, and the safety model for letting an LLM act on real infrastructure.
mcp Strip away the hype and an AI agent is one simple thing: a language model running in a loop, calling tools, and reading the results until a goal is reached. A chatbot answers once and stops. An agent acts — it decides its own next step based on what just happened in the environment.