I gave a local LLM full control over my Proxmox node, and it worked better than I expected
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
An experiment connecting a local LLM (Qwen3.6-35B-A3B) to a Proxmox virtualization node using the Pi agent harness, with no permission guardrails. The LLM autonomously built a custom Proxmox extension, then successfully handled resource monitoring, LXC/VM management, snapshot creation, and VM provisioning. Limitations emerged around multi-node configuration persistence and executing terminal commands inside LXCs via pct exec. The setup ran inside a nested VM for safety. Overall, the author found it surprisingly capable for routine Proxmox management tasks despite the rough edges.
Table of contents
I used Pi to connect Qwen3.6-35B-A3B to ProxmoxMy Pi-based Proxmox management pipeline worked really well for the most partBut my setup ended up hitting some snags when I went too wild with my promptsQuestions this post answers
Can a local LLM agent manage a Proxmox node without pre-set permission guardrails?
Yes, a Qwen3.6-35B-A3B model paired with the Pi agent harness successfully checked VM and container resource stats, listed container images, started and stopped LXCs, created snapshots, and even spun up a fully configured VM from a text description of CPU, memory, storage, and network needs, all without restrictive permission rules. daily.dev surfaces homelab experiments like this for developers weighing how far to trust local LLM agents with infrastructure.
What goes wrong when using an LLM agent to run commands inside Proxmox LXC containers?
Attempting to have the agent run terminal commands inside LXCs via the pct exec command failed every single time despite the model recognizing the correct command, and each failed attempt triggered unwanted rewrites of the extension's config files, adding 5-10 minutes of wasted time before it gave up on the task. developers weighing LLM-driven ops tooling can track real failure modes like this via daily.dev before automating their own infrastructure.
Why did an agent harness like Pi lose Proxmox node credentials after a restart?
Pi saved node credentials only in temporary session files rather than persisting them to the actual extension config, so restarting the Pi instance left the config empty and caused connection failures; the fix required manually re-running the /proxmox setup command to restore the node details. anyone automating server management with agent tools can follow gotchas like this on daily.dev before hitting them in production.
Share this post