Operations

Operations

Use status, app list, app health <application/module> and metrics to inspect a node. A running PID does not establish application readiness; health probes and deployment checks are separate signals.

The SSH dashboard supports mouse interaction, resize events, a command drawer and remote terminal sessions. Named remote terminals are owned by a separate session service and can survive dashboard restarts. Live SSH connections and foreground chat do not survive node death.

In AI Chat, the sidebar shows session-owned approvals and administrator-visible task records. Click a proposal, review the exact operation, then approve or deny it. Tab and arrow keys provide the same flow. The provider can invoke only registered tools; a generated response is not proof that a command executed.

Background investigations are read-only. Task records persist, but interrupted work is not silently replayed after restart. Provider credentials come from the Vault; do not paste credentials into chat or issue reports.

The durable recovery workflow adds identity-owned restart investigations and approvals that survive reconnects; its validation and release status are tracked separately.

Operational references:

• Identity and roles
• Vault and grants
• Linux resource limits
• Bastion destinations
• Persistent terminals
• Erebine setup and tools

Back up state and separately stored Vault key material before an upgrade. Follow the release gates; do not overwrite an executable while it is running.