Skip to main content

AI Assistant

The AI Assistant is a chat that can do things, not only answer questions. It works inside your account with shell and file access and the panel's own operations, so "my WordPress site shows a database error" can end with the cause found and, once you approve the fix, corrected.

It knows your account's layout (which sites and services you have) and answers in your panel language. Assistant in the panel's top bar opens it as a side panel that stays open as you move between pages: you can read a log or change a setting while it works, and a page it points you to opens beside the conversation.

Not on every server

Your provider can leave it out of a small server; then the panel doesn't show Assistant. On some small servers it stops when idle: the first answer takes a few seconds, and the chat shows "Starting the assistant…" until it's ready.

The assistant can:

  • Inspect and edit your files, and run commands, as your own system user, with exactly your permissions, inside your home directory.
  • Use panel features for you: create and configure databases and app servers, inspect services. It uses the same operations and validation as the panel interface, and is a client of the same interface your own AI tools reach through the MCP server.

How it asks for approval​

The dividing line is between a read and a write.

ExamplesWhat happens
ReadReading files, listing processes, inspecting configuration.Done freely. This is what makes diagnosis fast.
WriteEditing a file, running a command that isn't read-only, creating or changing a panel resource.The conversation stops at an Approve / Reject card describing the action.

Nothing in your account changes until you press Approve, for each action.

Two more properties complete the safety model:

  • Confinement. The assistant authenticates on the server as you. It can't see or touch anything your own user can't, and it can't open an outbound SSH connection.
  • Audit. Each command and tool call is recorded in your account's audit trail.
Check what you approve

The assistant acts as you. A destructive command you approve is still destructive.

Usage and budget​

The assistant draws on a budget your provider grants, which refills on a schedule.

  • With no budget, the assistant is off entirely.
  • When the budget runs out, it pauses until the refill.

You get a clear message in both cases and can still read past conversations. See Quotas and limits.

Conversations are kept per account. When you're done with one, archive it: it leaves your active list but stays on record.

Prefer your own model?

The same capabilities are available through the MCP server, with your own AI tool and your own budget.