I built this because I kept losing track of which agents needed attention. Running 4+ Claude Code sessions across terminal tabs, I'd find one sitting idle for 20 minutes waiting for approval while I was focused elsewhere. Desktop notifications didn't help — they'd vanish before I noticed.
Architect solves this with visual feedback. When an agent finishes, its cell changes hue. When it needs approval, it glows. The grid layout means I can see all sessions at once and know immediately where to focus. I also built in git worktree support, so switching between tasks is fast.
It's built on ghostty-vt (the terminal emulation library from Ghostty) and uses hooks from Claude Code, Gemini CLI, and Codex to detect agent state. Written in Zig — partly to learn the language, partly because ghostty-vt is also Zig.
forketyfork•1h ago
Architect solves this with visual feedback. When an agent finishes, its cell changes hue. When it needs approval, it glows. The grid layout means I can see all sessions at once and know immediately where to focus. I also built in git worktree support, so switching between tasks is fast.
It's built on ghostty-vt (the terminal emulation library from Ghostty) and uses hooks from Claude Code, Gemini CLI, and Codex to detect agent state. Written in Zig — partly to learn the language, partly because ghostty-vt is also Zig.
macOS only right now. Linux is planned.
Blog post with more details: https://forketyfork.github.io/blog/2026/01/21/running-4-ai-c...