aster.ax / Docs

Documentation overview

AgentXchange gives remote commands a job identity, retained output and a recorded outcome. People and agents can leave a command running, return in another session, and manage the same execution.

What you can do

When this happens What ax gives you
Your tool call times out or the client disconnects A job ID you can recover, with current status and retained output. Run and recover your first job.
Several agents or sessions have work on one runner Jobs listed by local project, with the recorded harness and session when supplied. Find your work.
You need to stop one build or test run Cancellation by job ID, followed by a confirmed terminal state. Stop the intended job.
A submission reply disappears A saved request ID and exact-retry checks so you can reconcile the original submission. Retry safely.
A tool displays only part of a long log Retained stdout/stderr you can read in pages and resume at a cursor. Recover output.
An agent needs to decide what to do next Structured receipts, states, exit codes or signals, and JSON, CSV or TSV output. Use ax from an agent.

These workflows grew out of repeated incidents in real Codex and Claude Code sessions: self-matching wait loops, cleanup commands that killed their own shell, and remote work interrupted by connection loss. Read the original excerpts and how ax handles each situation.

Start with a complete run

Install ax, then follow your first remote job. You will deploy a runner over an existing SSH connection, submit a command, close the client and recover its result. The walkthrough uses Linux with systemd.

For an agent harness, install the ax skill. It teaches runner selection, submission, recovery and cancellation through the CLI. MCP is also available.

Understand execution and access

Operate and configure