🛠️ Manage daemons

Daemons lists the machines connected to your organization. Open one to inspect its health, available runtimes, resources, hosted agents, capabilities, and visibility.

Status

Transcripts owned by an offline daemon cannot be fetched until it reconnects.

A restart or upgrade appears as Restarting or Upgrading until the daemon is ready again.

Actions

Restart and upgrade use the ordinary daemon edit permission: any Owner or Collaborator who can see the daemon may run them. Viewers cannot.

• Restart (online daemons) — drains work and asks the service supervisor to relaunch the same version.
• Upgrade (online daemons) — installs a selected release, drains and relaunches, then health-checks the result. A failed upgrade rolls back when possible. The entry point is the Update to <version> badge on an outdated daemon.
• Reconnect (offline daemons) — mints a fresh one-time token and shows a command to run on that host. Its identity and agent placements are preserved.
• Delete (offline daemons) — removes the daemon and revokes its keys. Its agents are left unplaced and inactive, so they stop responding until you place them on another daemon. The local process and files remain on the machine until you stop or remove them there.

Placement

Every placed agent runs on one daemon — that machine owns its workspace, runtime processes and transcripts. You can move an agent from its Configuration tab when both daemons are online, ready and compatible with its runtime, model and MCP servers.

A move cold-reprovisions the saved agent definition; it does not migrate daemon-local workspace, memory or transcript bytes. Commit or back up local work first, and expect GitHub workspaces to be cloned again on the target.

Session retention

Each daemon has an Expire sessions setting: choose a common window, enter a custom number of days, or keep sessions indefinitely. It controls how long that daemon keeps finished transcripts and session worktrees. The Control Plane keeps session metadata after the content expires. See Session retention and cleanup for what remains visible.

Troubleshooting

• Stuck “Waiting for daemon…” in Add daemon — the command probably failed in your terminal. Check that Node is ≥ 24.12 (node -v) and that the machine can reach your control-plane URL over HTTPS/WSS.
• Daemon shows offline but the process is running — check the log (npx -y @agentconnect.md/cli status prints its path, default ~/.agentconnect/logs/daemon.log). Repeated connect/handshake failed usually means the key was revoked — use Reconnect for that daemon, or onboard a new one if it was deleted.
• Runtime missing from the pickers — make sure the runtime is installed, authenticated, and on PATH for the OS user that owns the daemon service. Restart the daemon, then check its log if the runtime is still missing.
• Run in sandbox is unavailable — the selected daemon did not pass the live Linux sandbox probe. Check its OS, bubblewrap / ripgrep / socat dependencies, user-namespace policy and logs, then restart it. See Sandboxing.