Musings
Things I've built, broken, or thought way too hard about. Mostly the last one.
Equal parts architecture, threat model, and me arguing with myself.
All Posts
-
devlocal, Part 1: The Boundary
Trust zones, one way in, encryption on every hop, and rule sets that each deny on their own. The devlocal network, built to contain north/south and east/west threats alike.
-
devlocal, Part 2: Agents as Tenants
Persistent identity, ephemeral execution. AI agents as first-class users without trusting them: sandboxed runs, credentials that die with the task, one model gateway, and evidence for every action.