umputun/agterm

What it solves

It addresses the difficulty of managing multiple long-lived AI coding agent sessions in a standard tabbed terminal, where it is easy to lose track of which agents are active, blocked, or finished. It provides a structured way to organize these sessions and a programmatic interface for agents to control their own environment.

How it works

agterm is a native macOS terminal (built using the Ghostty engine) that organizes shells into named workspaces and sessions. It exposes a full control API via a bundled CLI tool called agtermctl, which allows scripts or AI agents to create sessions, type commands, read terminal text, and manage window layouts over a local socket. It also includes "agent skills" that can be installed into LLMs (like Claude Code or Codex) to teach them how to use this API to drive the terminal themselves.

Who it’s for

Developers who run multiple AI coding agents concurrently and need a terminal that can be programmatically controlled and monitored.

Highlights

• Control API & CLI: Use agtermctl to programmatically drive the terminal, including creating sessions and reading buffers.
• Workspace Organization: Group sessions into named workspaces (e.g., "work" or "personal") for better context management.
• Agent Integration: Installable skills for AI agents to manage their own layouts and status indicators (active, blocked, completed).
• Advanced Layouts: Supports splits, scratch terminals, and floating overlays for running interactive programs without disturbing the underlying shell.
• Live Restore: Ability to keep primary and split processes alive across app restarts using zmx.