Get started with the AI copilot
Start an AI copilot session on any server in Servor — a chat assistant that inspects your machine and runs commands with your approval.
The AI copilot is a chat assistant that works directly on your server. You describe
what you want in plain language ("why is nginx down?", "free up disk space"), and it
investigates and proposes commands. It has exactly one tool: run_command. Every
command it wants to run happens in your browser, and you stay in control of what
actually executes.
Before you start
The copilot runs real commands on a real machine, so two things need to be in place:
- A connected server. If you haven't added one yet, start with Add a server.
- An unlocked vault. Running commands requires your vault to be unlocked for the session — see Unlock your vault. If the vault is locked, you'll see an unlock prompt before the session starts.
Who can use it
Using the copilot counts as running commands, so viewers can't start a session. Members, admins and owners can. See Roles and permissions.
Start a session
Open a server
Go to Servers and select the machine you want to work on. The server must be connected — if it shows as unreachable, fix that first with Repair the agent.
Open the AI copilot
Switch to the AI tab (or the copilot panel). If your vault is locked, enter your passphrase to unlock it for this session.
Pick a mode
Choose Ask, Plan or Auto at the top of the chat. This decides how much the copilot can do on its own. Not sure? Start with Plan. Full breakdown in Ask, Plan & Auto.
Describe what you need
Type your request in plain language. The copilot reads context, proposes commands, and — depending on the mode and the risk — either asks you to approve each one or runs it and shows you the result.
What the copilot can do
Because its only tool is run_command, everything the copilot does is a command it
runs on your server through the browser. In practice that covers a lot:
- Diagnose problems — check service status, tail logs, inspect processes, look at disk, memory and CPU usage.
- Explain what it finds — it reads the output and tells you, in plain language, what's wrong and what it suggests next.
- Fix things — restart a service, clear a cache, rotate logs, adjust a config — each proposed as a command you can approve.
- Automate routine checks — walk through a health check or a deployment step by step.
It works from the same command history as the web terminal: everything it runs is recorded in your searchable command history.
Give it a starting point
Instead of typing from scratch, launch a recipe — a preset prompt (like "audit security" or "diagnose high CPU") that opens a session already pointed at a task.
What it can't do
- It can't bypass approvals. Risky and destructive actions always require your confirmation, whatever the mode — see Approve actions.
- It can't run blacklisted commands. Servor blocks known-dangerous commands (like
rm -rf /) for everyone, including the copilot. - It can't act without an unlocked vault or a connected agent. If the server goes unreachable mid-session, commands stop until you repair the agent.
Credits
Copilot sessions consume credits, billed on the real token cost of the conversation. You can see usage and top up in Settings → Billing — see Credits and usage.