Proxmox Aiops

Disclaimer: Community-maintained open-source project. Not affiliated with, endorsed by, or sponsored by Proxmox Server Solutions GmbH. "Proxmox" is a trademark of its owner. MIT licensed.

AI-powered Proxmox VE VM and container lifecycle operations with a built-in governance harness — unified audit log, token/runaway budget guard, undo-token recording, and descriptive risk-tier labels. Self-contained: no external dependencies beyond proxmoxer and the MCP SDK. Coverage is not yet exhaustive across every Proxmox operation.

Verification status: live-verified against real Proxmox VE 8.4.19 across three rounds — read-only surfaces, the QEMU write surface, and a two-node cluster (quorum, live migration, move-disk, a backup that actually succeeded). Rounds 2 and 3 each found a real bug the mocks could not see. See docs/VERIFICATION.md for exactly what was proven and what is still uncovered.

What works

• CLI (proxmox-aiops ...): vm list/get/config/start/stop/shutdown/reboot/reconfigure/clone/delete/migrate, vm resize-disk/move-disk/agent-ping, vm snapshot-create/snapshot-delete/snapshot-list/snapshot-rollback, backup create/list/restore, ct list/start/stop, cluster nodes/status/task-status/resources/node-status/task-log/next-vmid, ha status/resources, pool list/members, firewall vm-rules/cluster-status, storage list/content, diagnose node-pressure/guest-health, undo list/apply, init, secret set/list/rm/migrate/rotate-password, doctor, mcp.
• MCP server (proxmox-aiops mcp or proxmox-aiops-mcp): 43 tools, every one wrapped with the bundled @governed_tool harness.
• Diagnostics / RCA (read-only): diagnose node-pressure ranks cluster nodes by CPU/memory/root-fs pressure; diagnose guest-health scans VMs/containers for stopped guests, memory saturation, and disks near full. Every finding cites the measured number that tripped it and a concrete action — transparent heuristics, not a black-box verdict.
• Credentials: proxmox-aiops init (onboarding wizard) and proxmox-aiops secret ... manage an encrypted secret store — no plaintext passwords in config.yaml.
• Reversibility: write ops with a clean inverse (start/stop/shutdown/reconfigure/clone/migrate/snapshot-create/move-disk, container start/stop, and restore-into-a-free-vmid) record an inverse undo descriptor; irreversible ops (delete, snapshot-rollback, forced restore) declare none and are tagged high risk. Disk resize is grow-only (shrink refused).
• Async tasks: Proxmox writes return a task UPID — poll completion with cluster task-status / read lines with cluster task-log (the runaway budget guard prevents poll loops from running away).

What this tool does, and does not, decide

It delivers Proxmox VE operations — reads and writes — accurately and efficiently, and records every one of them. It does not decide whether a write is allowed to happen. That is the agent's judgement, or the permission of the account you connect it with: use a Proxmox VE user or API token granted only read privileges (no VM./Datastore. write roles), and the writes fail at the server — the place that actually owns the permission.

So there is no read-only switch, no policy file, no approval gate to configure. The one thing the tool guarantees is that nothing is silent: every call, over MCP and over the CLI alike, lands an audit row in ~/.proxmox-aiops/audit.db, and destructive writes still capture their before-state and record an inverse where one exists.

Each tool declares a risk_level, carried into the audit row as a descriptive tier (none/confirm/review) — so a reviewer can see at a glance that a row was a high-risk delete. It is a label, not a gate.