Herdr 0.9 lets one terminal client manage agents across multiple machines

One terminal, many machines

Herdr, an open-source terminal application for managing AI agents, has released version 0.9 with support for controlling agents across multiple machines from a single interface. According to the Herdr blog, a new herdr machine add command takes an SSH target and makes that machine's workspaces, tabs and agents appear alongside local ones in the same TUI. Agents keep running on their own machines even when the client disconnects.

The project says it has more than 700,000 downloads and close to 1,000 plugins, and that users have been asking for multi-machine support for months. Some had already built workarounds through Herdr's plugin system, but the core application was not designed for the use case.

A rebuilt client-server split

Herdr already had a client and a server, but the architecture tied one client to one server, with everything rendered on the server. Multiplexers such as tmux and Zellij do not support a single TUI client managing independent servers on different machines, which the post describes as a niche use case until agents made it routine. Rather than making one server aggregate all connections, the team moved rendering of the outer UI to the client, while each server continues to own its sessions and supply terminal views. The post says this separation also leaves room for performance improvements as the client and the runtime are pulled further apart.

The thinking behind it

The announcement frames the change around developers starting to treat laptops as clients rather than as the source of compute. Agents frequently run for hours, and the post argues you should not have to keep a laptop open just because an agent is still working. A VPS or a machine such as a Mac mini elsewhere can make more sense than spending a couple of thousand dollars on the laptop you carry around.

Current limits and what comes next

This is described as the first version of multi-machine support. Machines must be reachable over SSH, whether on a local network, over the internet, or through a VPN such as Tailscale. The agent CLI still operates within a single server and cannot yet see agents running on other machines; cross-machine agent collaboration, where agents find and work with each other across servers, is named as a goal for later work. The team says it shipped the feature early to gather feedback from real usage before deciding what to build next.

The next planned piece is Herdr Cloud, which did not make it into this release. It would act as a connection layer, letting users link machines with a single command and a Herdr account instead of dealing with SSH and network setup. Terminal traffic would be end-to-end encrypted between the machines and the Herdr client, and users would still bring their own machines — the cloud service would connect them, not host the agents. Further out, the post raises the idea of moving a running agent session between machines, so the choice of where an agent lives would not have to be made before work starts. A waitlist is open for the cloud component.

Why it matters

Local agent orchestration has so far mostly meant one machine managing itself. Herdr 0.9 points at a different shape: a lightweight client that supervises agents running wherever compute is cheap and always on. That matches how agent workloads actually behave — long, unattended, and better suited to a server in a closet than a laptop in a bag. The gap between this release and the stated roadmap is real: the CLI is still single-server, SSH is still required, and the cloud layer does not exist yet. But making multi-machine control a first-class part of the client, rather than a plugin workaround, is a concrete step toward treating a fleet of machines as a single workspace.