Agent of Empires

You will feel the difference the first time your SSH connection drops. Every agent lives in its own persistent session, so a crashed terminal or a closed laptop costs you nothing and reopening restores the lot. The deciding daily trait is that continuity, because the expensive failure with long agent runs is not a bad diff, it is losing forty minutes of context to a network hiccup.

Pick it if you routinely have three or four agents in flight. Pick Claude Squad if two panes and a keybinding already cover your day.

Start with the exposure. Agents here execute shell commands and touch git, and they are reachable from a TUI, a CLI, an HTTP API and a browser. That is four doors into a process with write access to your source tree, and the sandboxing that would contain it is opt-in rather than the default. A phone-friendly view means the door is sometimes open on an untrusted network.

What it does right: status detection distinguishes running, waiting, idle and error, so a stuck agent is visible rather than silently burning a session.

1. File-level isolation uses git worktrees and multi-repo workspaces, which is the correct cheap answer to two agents editing one tree. 2. Process-level isolation is a separate concern handled by Docker, Podman or Apple Containers, and it is configurable rather than implied. 3. Verification is not addressed at all; nothing inspects an agent's output before a worktree is reused.

Documentation is the repository README, which is thin for a tool with four interfaces. No benchmark is claimed, and none would be meaningful for a supervisor.

Eight months old and 3,190 stars is genuine traction for a supervisor tool, and the category is real because everyone running several agents hits this problem. The trouble is that nothing here is defensible. There is no hosted tier, no price, no enterprise edition and no obvious place to put one, and the pattern underneath is a session manager, which is a weekend of work for a competent team.

Likely path: absorbed as a feature by whichever agent CLI decides sessions matter. Position: pass on the company, use the binary.

The demo is a phone showing four running agents, and it is a good demo. Then the questions start. This installs per developer via a shell script or a package manager, so there is no console, no directory integration and no retention statement, and every engineer's supervisor is invisible to us. Cost is zero for the tool and unbounded for the agents it keeps alive, which is the meter nobody budgets. It runs on macOS and Linux only.

Onboarding is an hour. Not yet, absent a managed deployment story.

This is built the way I would build it. MIT, so a fork survives anything. Installation is brew or a shell script, and the isolation layer accepts Podman or Apple Containers instead of insisting on one daemon, which is the first time a tool on this board has asked what I already run rather than telling me.

The gap is protocol. There is no MCP client and no MCP server, so my existing tool servers do not reach the agents this thing supervises. That is a real limit, and given the licence it is also an afternoon of work rather than a support ticket.