All work

HQ

Our own product

Our control room for AI coding agents. They work on schedules; a supervisor decides when to wake us.

HQ dashboard: tasks happening now across projects, each with its agent's progress, and a list of tasks finished today.
Happening now · Mac

01 · The problem

Running several AI coding agents by hand means babysitting terminals, answering the same questions twice and losing work when a process dies.

02 · What we built

A self-hosted control room. Claude and Codex agents take tasks from a crash-safe queue, run on schedules and share memory. A supervisor agent answers what it can and escalates the rest.

03 · The result

Only the decisions that need a person reach us. It runs seven agents across five projects, from a browser or a native Mac app.

More screens

HQ memory: notes the agents share, with counts of notes read in every chat, parked and flagged for a closer look.
Shared memory · Mac
HQ conversations: agent and supervisor messages, waiting for an answer, beside the full thread of one mission.
Supervisor conversations · Mac

01

Supervisor routing

Workers ask the supervisor first; after 10 minutes it escalates to us.

02

Queue and schedules

Tasks survive restarts, and a built-in cron runs the overnight shift.

03

Guardrails that fail closed

Risky commands are blocked, and anything unreadable is refused.

What we did
Architecture · AI agents · Backend · Realtime · Mac app · Security
Built with
TypeScript · Fastify · SQLite · React · Electron · Claude
Platforms
Web · Mac
Status
Internal. In daily use behind a login.

Have something to build?

Pick a package and send your brand, or tell us what you need.