Berth runs Claude Code and Codex on your own boxes so they survive a closed laptop

What Berth does

Berth, whose site reached the Hacker News front page on 6 October 2026, is a macOS app that runs coding agents such as Claude Code and Codex on development machines you already own rather than on the laptop in front of you. The pitch is simple: shut the laptop and the agents carry on; open it the next morning and their output is waiting.

The problem it targets is that agent sessions are long-running. When an agent runs as a local process, a closed lid, a dead battery or an ended workday interrupts it. According to the project's site, Berth moves execution onto what it calls your own dev boxes and treats the laptop as the place where you steer, observe and respond. The landing page does not describe the underlying transport to those machines, but it does show status examples naming multiple machines, each with its own count of active agents.

Starting an agent session

Per the site, a session begins with a single action: press ⌘N and describe the task in one line. Berth then creates a Git worktree on the machine you choose, launches Claude Code, Codex or another agent inside it, and gives that worktree's dev server a tab of its own. Because each task gets its own worktree, several agents can operate on the same repository without overwriting each other's working copy.

Watching and comparing worktrees

The interface is built around following agent work as it happens. The site says commands and file reads are condensed into single lines while edits appear as diffs, and when the agent asks a question you answer it in place; anything the agent publishes sits one click away.

For parallel work, two worktrees can share the tab strip as colour-coded groups within one tab. A Compare mode goes further, arranging two worktrees in mirrored lanes so their chats, diffs and terminals line up next to each other — handy when two agents have taken different approaches to the same fix.

Customisation and availability

Berth ships with its own themes plus nineteen ports of well-known colour schemes, from Dracula to Rosé Pine, carried through into both the terminal and every diff. You can also set a custom chat background and enlarge the interface with ⌘+. The download covers Apple silicon and Intel Macs, is signed and notarized, and the project is open source.

Why it matters

Long-running coding agents change where compute needs to live. Tools like Claude Code assume sessions that can stretch across many minutes or hours, which sits badly with laptop sleep, battery limits and the habit of closing a machine at the end of the day. Berth's answer — keep the agent on hardware you already control and use the laptop as a window onto it — is cheaper than spinning up a cloud VM for every agent run, and it avoids handing your source code to a hosted service.

The open-source release matters too: when an agent has broad access to your repositories and shell, being able to audit exactly how the app reaches your boxes is a genuine advantage over closed alternatives. Whether the per-worktree model becomes the default way to juggle concurrent agents remains to be seen, but Berth is aimed squarely at a real friction point as agent-assisted development becomes routine.