technical blog
Practical notes from agent systems, local infrastructure, search, memory, and testing.
all blog postsWhen research reports should fail
A small Arcwell research run shows why claim ledgers and skeptic passes matter more than asking a model to write a smoother report.
June 2026Porting agent memory means porting behavior
mem0-rs keeps mem0's memory behavior while cutting the Python runtime cost: lower latency, lower resident memory, and a smaller deployment surface.
May 2026Tests should try to refute the claim
severe-testing is a small agent skill built around one habit: name the claim first, then write tests that would prove it false.
May 2026Coding agents as supervised local services
codex-swift explores what changes when a coding agent is not a one-shot CLI, but a daemon with per-session workers, shared services, OS sandboxing, and durable state.
projects
Public code for agents, local infrastructure, search, memory, and developer-program measurement.
all projectsArcwell
Local-first assistant services built around a recurring lesson: agents need durable memory, visible state, and bounded workflows.
githubcodex-swift
A native macOS coding-agent runtime exploring what changes when the agent becomes a long-lived local service.
githubopen-cloud
A local Cloudflare-shaped platform for making integration state close enough to inspect, reset, and test.
Colour after the fire
CoBrA, Stockholm Design Lab, and Rotterdam's public art route as a postwar argument about colour, freedom, and civic repair.
May 2026Taste takes time
Cloth, proportion, weather, and the slow education of the eye.
May 2026Boat at night
Old skills, cold water, and the private architecture of continuity.
May 2026Paraboots in the rain
French shoes, useful ugliness, and the difference between admiring an object and living in it.