amx v0.1.0

The first public release of amx.

amx runs coding agents (claude, codex, pi, opencode) as ordinary tmux panes. Each agent gets a tmux session and, in a git repository, a worktree of its own. There is no daemon: amx starts the agent and watches from the outside, so attaching is plain tmux and killing amx stops nothing.

• amx new "task" starts an agent and prints its id.
• amx opens the view: every agent grouped by project, with what it is doing now. Answer permission prompts and questions from a card, reply to an agent, review its diff, attach to its pane.
• amx result, amx wait, amx answer and amx send make it scriptable, with exit codes that say done, failed, blocked on a question or timed out. An agent can run amx sub to hand work to children and collect their answers.

Install:

curl -fsSL https://saifulapm.github.io/amx/install.sh | sh
amx setup claude    # or pi, codex, opencode
amx doctor

Prebuilt for Linux (x86_64, aarch64; static) and macOS (Intel, Apple silicon). Needs tmux 3.2 or newer.

Docs: https://saifulapm.github.io/amx/docs/