Attaching an editor

ACP is the protocol editors speak to coding agents. apn acp makes a station on another machine look like a local agent to any client that speaks it — Zed, JetBrains, anything else.

How it works

Your editor spawns apn acp and talks ACP over its stdin and stdout. Those frames are piped to the hub, which does the protocol work and routes them to the station. The node dialled out, so this reaches stations behind NAT or CGNAT that your editor could never connect to directly.

apn acp --list                 # stations you can attach to
apn acp --station <id>         # attach
apn acp --station <id> --session <id>   # resume a session

--hub points at a hub other than the default.

This is the one command that needs no node

Every other node verb assumes this machine is enrolled. apn acp does not — a laptop can install apn purely as a client, attach to stations elsewhere, and never be part of the fleet itself.

Tokens

Prefer $AGENTPOD_TOKEN, which is what --token defaults to. A token passed as an argument lands in your shell history and in the process list.

Sessions

Sessions are recorded, and an agent can read back its own transcripts through the MCP tools — agentpod_my_sessions to find a session, then agentpod_my_transcript to read it.