Autonomous Harness

The way we use computers changed. The hardware didn't. Mouse and keyboard held for forty years. Then agents arrived. Typing became the slow way to work. Tracking which agent is doing what, on which machine, became a full-time job — Harness is the device for that. Skip typing, just say one sentence. The work goes to the right agent, on whichever computer or server it lives on. It drives every agent you already use, and the newest ones.

Bring your own agent. Every agent Harness runs today joined through an open protocol. Yours can too. Ship a CLI command or deploy an HTTP endpoint — both open source on GitHub.

Set a Loop. It runs on schedule. Say the schedule once — "every 9am, check the overnight deploy" — and it runs. Every day, every hour, every Monday morning, whatever you set. You get pinged only when there's something worth seeing. Not another Slack ping to ignore.

Set a Goal. It runs until done. Not prompt, then prompt, then prompt. A Goal takes one instruction and runs it to done. Say "get all billing tests passing" once. The agent runs, fails, fixes, reruns, until every test passes. You get pinged when the goal is met.

Tap twice. Say the task. Step away. Double-tap the face and speak: "watch the deploy pipeline and flag anything red" or "keep running the test suite until it passes." Harness dispatches the task to the agent already running on your laptop, your server, or the cloud. Line up several tasks — Harness pings you when work finishes, or when something needs your call.

Every task. Every agent. One glance. The round display carries three things and nothing else: what is running and for how long, down to the second; the finished work, streamed on the moment it returns; and what your harness has spent today, down to the cent and readable across the room. No chat window, no feed, no personality — Harness doesn't talk back, it gets back to you, with one quiet tone when work returns. Voice in, work out, quiet in between.

A circle. A stand. Nothing else. Harness is a single matte-black disc — one material, one colour, no buttons to find and no menus to learn. The whole face is the tap surface. The stand ships in the box and is part of the product: a low, weighted cradle that holds Harness upright at the base of your monitor, facing you like an instrument. USB-C on the bottom edge, a loop up top for the days the desk comes with you.

Your agents. Your machines. Your Harness. Claude Code, Codex, Cursor, OpenCode, Pi, Hermes, Command Code, Devin, and more — the agents you already run, on the machines you already own. Nothing to migrate, nothing to rewrite. When they finish a task, hit a loop, or complete a goal, Harness catches the signal and brings it back to your desk. One face for every agent. Speak into it. See the work on it.

End-to-end encryption. The agent runs where it already runs — your laptop, your server. Files, diffs, commits, logs, the repo itself: all of it stays there. The only thing that comes back to Harness is what got done, and what needs you to confirm. We don't keep a copy of the work either.