deck
macOS command center ActiveA native macOS app that keeps every persistent CLI agent session on one board — active, quiet, and stopped at a glance. Sessions outlive the app, stay on your machine, and work with any command-line agent.
c9r.io
Agent-first software delivery
Tools and experiments for building software where coding agents do the work — and where verification runs fast enough to keep up with them.
Projects
Three attempts at the same question, from three directions.
auth9
Identity platform & SDLC experiment ActiveA self-hosted identity and access management platform — multi-tenant isolation, OIDC and OAuth2, token exchange, hierarchical RBAC — written almost entirely by agents. It exists to test whether an AI-native development lifecycle can polish software the way people do, on a problem where a wrong decision cascades.
orchestrator
Control plane ArchivedA control plane that drove Claude Code, Codex, OpenCode and shell-native agents from declarative workflows. Development has stopped and the repository is archived; the documentation stays online, read-only, for anyone still reading the code.
Approach
What the projects have in common, stated plainly enough to be wrong.
Agents do the work
Not autocomplete around a human author — agents carry the task from plan to patch. The interesting engineering is in what surrounds them.
Verification sets the pace
Writing code was never the bottleneck; knowing it is correct is. Generation that outruns verification just builds a bigger queue of unreviewed work.
Local by default
Sessions, state and source stay on the machine that produced them. Nothing here needs an account to be useful.