02 — Subject

AI

Agents, protocols, and working with models as tools rather than magic.

Less about which model is best this month, more about the shape of the tools: agents, the Model Context Protocol, and the difference between what reasoning models do and what they appear to do.

Articles

Sixteen Times SmallerArticleturbovec, Google’s TurboQuant as a vector index in Rust — what rotate-then-round does, what it costs in recall, who says RaBitQ got there first, and how it fares against FAISS on four CPU cores. A Voice in the RoomArticleLocal voice agents in three shapes — Voicebox, VoxCPM and Dograh placed among Moshi, Unmute, Pipecat and Home Assistant — with the household cascade timed on four CPU cores. An Agent in Your PocketGuideGemma 4 on a phone through LiteRT-LM — the app, the 270M model that only calls tools, and the Kotlin API — with the laptop numbers measured and the phone numbers labelled Google's. Videos from HTMLGuideHow the four narrated explainers on this site were made — HyperFrames compositions and Kokoro narration, rendered on a laptop for nothing — with the pipeline timed. An Agent on Your Own MachineGuideGemma 4 and Hermes Desktop with nothing leaving the room — served by Ollama, Docker Model Runner or vLLM, measured on a 24 GB Mac. One Agent, Four DoorsArticlePackaging an AI agent for the terminal, REST, MCP and AG-UI from one core — with Hermes behind it, and why "AGP" is not a fifth.

AI

Articles

From the newsletter