Agent

What the agent can see, send and do

All help topics

The agent explains the console and looks up your engines, and it can make some changes for you, but only after you click Approve on a card that says exactly what will happen.

What it can do
  • Answer how-to questions from the console's own help and the support-center articles, and link the page or article it used.
  • Look up your engines: which ones you have and their state, one engine's details (its state, nodes and subnet), who has access to it and with which role (yours included), that engine's billing (balance and charges), its open approval requests and its proxies. It does this as you, so it sees exactly what you can see in the console and nothing more.
  • Read what the page you're on shows (its headings, buttons and notices), so it names controls the way you see them and can tell you why a button is disabled.
  • Read the apps on your engines: which apps and canisters an engine runs, an app's published interface (its methods), and what the app's read-only (query) methods return when it calls them as you. That can be data the console itself never shows. Reading needs no card; only a change does.
  • Make a change you ask for: rename an engine; give someone access to it or take that access away; approve, reject or cancel an approval request; turn Caffeine on or off; choose the inference provider or turn inference off; or delete a proxy. It first shows a card, "Approve this change?", with what will happen and the engine and person it is about. Nothing changes until you click Approve; Deny cancels it. It only offers what your role lets you do in the console, and when the engine's approval rules apply, it sends a request for approval just as the Team page does.
  • Take you to the right page with its dialog open, for what it can't do in the chat.
What it can't do
  • Pay for anything, upload a file, or delete an engine or an app. For those it opens the page or dialog where you finish it yourself.
  • Do anything your role doesn't allow in the console, or see engines you have no access to.
What goes to the AI model

An empty conversation says it below the suggested questions: "Your question, the page you are on and what it shows, details of the engine you are viewing, and what it reads to answer you (such as an app's interface and the data its queries return) go to an AI model. Answers can be wrong, so check what matters." Everything a lookup returns goes too: your engines' details, your team, your billing, and your apps' interfaces and query results. The request reaches the model through OpenCloud's own servers. The conversation lives only in your open console tab: reloading the page or signing out ends it.

Help only

When no AI model is available, the status reads Help only and answers come straight from the console's help topics and the support center, marked "Answered from the console's help, without the AI model." Nothing goes to an AI model then. Start a new conversation, or reloading the page, checks again whether the model is back.

Still stuck? Contact support