A terminal for agent fleets
I basically have the final form of my stack built now. Its foundation is a terminal multiplexer with agent fleet management built in.
I was a hardcore Ghostty user for years, and even submitted patches the maintainers adopted. But I needed more than a CLI multiplexer like tmux or herdr could give me, though herdr comes very close.
I built Sox so I can leave my computers running without being tied to my desk. Too often my progress halted because an agent needed my input or approval. So I built an attention protocol that lets one human manage fleets of agents across heterogeneous infrastructure: you attend to the agents that need you most, simply and efficiently.
I made it a terminal multiplexer because today’s agent GUIs are closed, or only work well with one harness. Terminal-native means I choose my harness, and when a smarter model ships, switching workflows is easy.
This has been one of the most fun things I’ve ever built, and I learned a ton. I’m excited for the new terminals coming out: Superlogical’s Rex and Tern, plus omp.
What Sox has today
- Durable shell sessions on macOS and Linux hosts that outlive the app.
- macOS, iOS, and Android clients for attaching to those shells. The mobile clients are in private beta.
- An attention protocol that tells you when an agent needs you.
- The Command Center: your agent inbox and action queue.
- File sharing with agents on remote hosts, over the session’s own secure transport.
- Agent skills for coordination, including messaging between sessions.