External agents get a terminal: when an ACP agent (Gemini CLI and others) runs commands, LEVEL11's approval rules apply and the output shows as a card in the conversation. Switching the work mode (Chat, Builder, Auto-Run) switches the agent's mode too.
ACP agents from your phone: the remote host's task screen lets you pick a configured ACP agent.
No leftover processes: LEVEL11 keeps track of every program the AI starts and cleans them up when you quit or delete a conversation; processes left behind by a crash are cleaned up on the next launch. Settings lists running processes with a stop button.
Start-up problems in plain words: a corrupt settings file is backed up and reset; an unwritable data folder, a busy remote port or a native module error come with the one thing to do.
Marketplace safety labels: plugin cards show required permissions, risk level and supported version, with a confirmation before installing.