Agent of Empires is not just a TUI wrapper for coding agents anymore. The release stream points at a sharper bet: run many AI coding sessions, then control them from a terminal, browser, or phone without losing tmux persistence.

ToolVitals sees 10 release events in 30 days and 30 GitHub releases in 90 days. The project has 2,074 GitHub stars, a 100 shipping score, a 92 health score, and a 90 ToolVitals score. That is a loud signal for a repo in the automation category, even with 70 data confidence and no 30-day commit count in the payload.

The official site describes Agent of Empires as a terminal session manager for AI coding agents, with support for Claude Code, OpenCode, git worktrees, Docker sandboxing, tmux persistence, and status detection. The GitHub repo frames it more broadly: manage multiple Claude Code and OpenCode agents from either a TUI or web UI, with mobile access and support for tools such as Codex CLI, Gemini CLI, Copilot CLI, and others.

The interesting signal is web and mobile control

The recent releases are not random patch churn. They cluster around making agent sessions controllable outside the local terminal.

Version 1.4.0 added a web command palette, a top app bar, remote access over Cloudflare Tunnel, and web DX polish around errors, version display, security settings, and toasts. Version 1.4.2 added terminal font pinch-to-zoom, local network or Cloudflare tunnel selection, and recovery for dead agent sessions on attach. Version 1.4.3 fixed iOS PWA authentication and mobile keyboard handling.

Then the project kept pushing in the same direction. Version 1.4.4 added mobile-first project creation, profile selection, settings, a web UI package target for Nix, a replacement of xterm.js with wterm, session deletion, stable serve ports, and a clone-from-URL flow. Version 1.4.5 says the web dashboard is no longer experimental, then adds mobile terminal improvements, right-edge diff swipes, and a virtual keyboard bar.

Version 1.4.6 continues the same story: dashboard hardening, Web Push notifications, daemon orphan prevention, merge conflict support in diff view, Tailscale Funnel preference for stable PWA-installable HTTPS, and more mobile fixes. Version 1.5.0 adds full settings in the web UI, URL-based dashboard routing, mobile paste fixes, terminal scroll fixes, selected profile overrides, and TUI navigation improvements like jumping to the next waiting session.

That suggests Agent of Empires is aiming at a specific workflow: developers running several long-lived coding agents in parallel, often in git worktrees, with enough remote UI polish to check and steer sessions away from the machine where they started.

The product bet is sensible. A coding agent manager that only works while you stare at one terminal is useful. A coding agent manager that keeps sessions alive in tmux, isolates them with Docker, tracks waiting states, and exposes the control plane through web and mobile starts to look like an operations console for local agents.

What ToolVitals cannot infer

ToolVitals can say Agent of Empires is active. It can say the project has 2,074 stars, 30 releases in 90 days, 10 release events in 30 days, a 100 shipping score, and a 92 health score.

ToolVitals cannot say the code is high quality. It cannot prove the web dashboard works well for every setup. It cannot measure user satisfaction, revenue, production usage, or whether Docker sandboxing is configured safely by each user.

The 30-day commit count and active contributor count are null in this payload. The GitHub page showed live activity when checked, but the ToolVitals metrics payload does not provide a 30-day commit number, so this post does not use one.

The release notes are strong evidence of direction, not proof of maturity. Web Push, Tailscale Funnel, Cloudflare Tunnel, mobile terminal gestures, PWA fixes, and dashboard hardening all point to serious remote-control work. They do not guarantee that your particular agent stack will behave cleanly under load.

How it compares

Agent of Empires is smaller than the giants in the related set, but it is not quiet. n8n has 186,990 stars and 52 release events in 30 days. LangChain has 136,054 stars and 36 release events. Agent of Empires has 2,074 stars and 10 release events.

That comparison cuts both ways. n8n and LangChain have massive distribution and faster public release volume. Agent of Empires is narrower, but its recent release notes are focused on one concrete job: managing many coding-agent sessions across TUI, web, and mobile.

Skyvern is a closer automation-category comparison by hot score. It has 21,526 stars, a 220.7 hot score, and 19 release events in 30 days. Agent of Empires has a 215.2 hot score and 10 release events. Skyvern has more stars and more recent release events, but it solves a different automation problem.

oxideterm is an interesting contrast. It has 648 stars, 30 release events in 30 days, and the same 100 shipping score. Agent of Empires has fewer recent release events than oxideterm, but more stars and a higher hot score in this payload.

Recommendation

If your team is already running Claude Code, OpenCode, Codex CLI, Gemini CLI, or similar tools in parallel, evaluate Agent of Empires as a local control plane for agent sessions. The strongest reason is not the 2,074 stars. It is the pattern of recent releases around web control, mobile access, tmux persistence, worktrees, Docker isolation, and waiting-state navigation.

If you need audited enterprise workflow automation, start elsewhere. If you need a focused way to supervise multiple coding agents from your terminal and phone, Agent of Empires is exactly the kind of fast-moving project to test in a sandbox this week.

Sources