Superagent gives every chat its own git checkout and a visible browser: Reveneau

Why it mattersTwo agent sessions on one repository normally conflict over the working directory. One checkout per conversation removes that problem by design, so nobody has to remember to avoid it.

Superagent is a macOS desktop application that gives Claude Code a place to work. It launched on Product Hunt this week under the line "Claude Code for the rest of us". The repository is MIT licensed and was created on 28 July 2026. It had 18 stars at the time of writing, so this is early and small.

It runs locally on the Claude subscription you already have. No server in between, no API key, and the source is readable, which matters given what it does with your browser.

The idea worth copying

Every conversation on a git project automatically gets a private copy of the project on its own branch.

Two chats on the same repository can edit, branch and commit side by side without touching each other or your own checkout. When a chat has changes, two buttons appear: Keep adds everything as one change named after the chat, Throw away deletes it. Deleting a chat with unkept work asks first.

The details around that are the part that shows someone actually used it. Branches are named from the conversation, so "Fix the flaky auth test" becomes fix-the-flaky-auth-test, and the name follows renames. Keep merges onto the branch the chat started from, even if the project has changed since. Dependency folders (node_modules, .venv, vendor, target) are shared into each copy, so no reinstall.

Anyone running more than one agent session on a repository has had the conflict this solves. Doing it per conversation, automatically, is a better answer than remembering to make a worktree.

The browser

The agent controls a real browser with your logged-in sessions: open a page, click, type, read it back. It is the browser on your screen rather than a hidden one it describes to you, so you watch it work and can take over.

The constraint the project states is the right one: automation only ever happens in Superagent's own browser pane, never in your personal browser.

Be clear about what this means before running it. An agent with your logged-in sessions can do whatever those sessions can do. Visibility is a real improvement over a headless browser you cannot see, and it is not the same as a permission boundary. There is an Ask mode where the agent checks before acting, and that is the setting to start with.

The rest

An iOS Simulator streams inside the app from its own framebuffer, live. You can tap, swipe and type on it, and so can the agent through tools like sim_boot, sim_tap and sim_screenshot, so it builds, launches and checks its own app.

A companion iPhone app connects to the Mac from anywhere: live transcripts, send a prompt, approve what the agent asks, with a push notification when it is done or needs you.

Claude Code today, with Codex and Antigravity listed as coming. Apple Silicon only.

At 18 stars this is worth reading rather than adopting, and the per-conversation checkout is worth copying whatever tool you use.

Source

pungme/superagent-desktop on GitHub, MIT licence. Star and date figures read from the GitHub API on 31 August 2026
Found via Product Hunt

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.