Superagent Gives Claude Code an Actual Computer

The pitch is 'Claude Code for the rest of us,' and the actual idea is sharper than that sounds. Claude Code writes code and then hands you scrolling terminal text, forgets everything when you close the window, and makes you alt-tab to check whether the page you asked it to build actually changed. Superagent's answer is to stop making the agent live in a terminal and give it a computer instead: a real browser it drives on your own logged-in sessions, an iOS Simulator it taps and swipes, your files, a notes board it keeps, and routines that run on a timer.

The design choice worth noticing is what they didn't build. There's no Superagent server and no extra subscription. It runs Claude Code locally, on the plan you already pay for, and your work stays on your own disk. The Mac app, the iPhone app, and the relay that puts your phone in the loop are all MIT licensed and open source on GitHub. In a season where every agent product wants to become the cloud you rent, shipping the whole thing as a local, readable, MIT app is a real position, not a footnote.

This is the same thread as the harness conversation we've been tracking for weeks, the argument that an agent is a model plus a harness and the harness is where your leverage actually lives. Superagent is a harness you can see: how it touches your browser, what leaves your machine, how the phone messages are encrypted, all there to read. Codex and Antigravity support are listed as coming. If you've been frustrated that the smartest coding model on your machine is trapped behind a wall of text, this is the most direct fix anyone's shipped. It's on Product Hunt and GitHub under the superagent name.