Skip to content

AI assistant

Settings → AI assistant (Tenant Administrator).

Vista Platform is AI-native and never AI-dependent. Every place a model could help has a rule-based behaviour behind it that runs whether or not anyone has configured a model — so a deployment with no AI at all is fully functional, not degraded and not half-built. This page is where you see which of those places have a model available here, what happens in each when one is not, and where you turn the whole thing off for your organization.

The page exists in every edition. It is not an upsell screen: the two switches on it are yours in Core as much as in Enterprise, and the capabilities table is an honest inventory of what this particular deployment does today.

How to get there

Click your profile chip (top right) → Organization Settings → Integrations → AI assistant.

You need the settings.update permission, which the Tenant Administrator role has. Other roles will see an access notice rather than the page.

What the page shows

Model provider

Row What it means
Model provider Which model endpoint answers for your organization, and whose it is: Provided by this deployment (whoever runs the platform set it) or Connected by your organization.
Model The model id that is asked for, when one is pinned.
Capabilities on How many of the capabilities below can answer here right now.

In Core this row reads Not included in this edition: the model clients are part of Enterprise, and the rule-based behaviour in the table below is what answers.

Whoever runs the platform can also set a provider for every organization — see Connecting a model provider.

You will never see an API key on this page, and you will never see the address of the deployment's own endpoint. For a provider your organization connected, the page shows the endpoint's host name and the last four characters of the key, and nothing more.

Capabilities

One row per AI capability, and four columns:

  • Capability — what it is.
  • Where — where in the product you meet it.
  • Without AI — what answers when no model does. This column is the point of the page. Every row has an entry; none of them is "nothing".
  • Status — one of four:
Status Means Who fixes it
On Something can answer through this capability here. —
Enterprise This build has no model client in it at all. A change of edition.
No provider The build has it; nobody has connected a model endpoint. Whoever runs the deployment.
Not yet built Nothing implements this capability yet, in any edition. Us.

"Enterprise" and "No provider" are shown separately on purpose. One of them is a purchase and the other is ten minutes of an administrator's time, and a single grey dash for both would send you to the wrong person.

Your organization's controls

Two switches, both yours, both applying to every capability above.

Use the AI assistant

The kill switch. Turn it off and no generative capability runs for your organization — no prompt is sent anywhere on your behalf, by any part of the platform, even where this deployment has a model provider connected and working.

Nothing breaks. Every capability falls back to exactly the behaviour in the "Without AI" column: the CBOM comparison still writes its summary, the inventory still classifies assets, and the end-of-life catalogue is still consulted. You lose the model's contribution and nothing else.

It is enforced twice — once where each capability is invoked, and again at the boundary every outbound model call passes through — so a capability added in a future release inherits the switch rather than having to remember it.

Record the questions we send

Off by default. Every AI call is written to your audit trail either way, with the capability, the model, who asked, how many tokens it used, and a fingerprint of what was sent. What the fingerprint does not include is the text: a person's question, or a document they pasted.

Turn this on and that text is stored in the audit trail alongside the fingerprint. Turn it off and it is not. That is the whole difference.

Whichever way it is set, secrets are removed from a prompt before anything is sent to a model, and the audit trail records the scrubbed version rather than the original — so switching this on does not put credentials into your logs.

What an asset row carries when it is sent

When you ask a question about your inventory, the rows that answer it travel to the model. They are projected, not sent whole: the model sees class, hostname, display name, primary address, environment, business unit, support group, site/region/zone, ownership, status, risk and dates.

It does not see the owner's email address, free-text descriptions, tags, attributes, or the metadata collectors write. Owner addresses are personal data about a named individual and are removed before the rows leave, so a question like "which production servers have no owner?" is still answered — that part runs in the database and the rows are the answer — while the addresses themselves are never sent to a model provider.

Reasons to turn it on: you want to review exactly what was asked, or you are troubleshooting an answer that looked wrong. Reasons to leave it off: the text people type is the most sensitive thing in this whole path, and an audit trail with a different retention policy from the data it quotes is a second copy of that data.

Drafting on custom policies

A third row appears when authoring on custom policies is switched off across the platform. It is not your setting and there is no toggle on it — it is a temporary, platform-wide pause, and the row states the reason and what is unaffected. The same notice appears on Settings → Custom Policies, which is where the capability itself lives.

Saving

Change either switch and Save changes becomes available. Saving writes to your organization's settings and is recorded in your audit trail against your user, like every other settings change. Connecting or disconnecting a provider saves on its own and is recorded the same way.

If your platform administrator also has a copy of these settings open, the last save wins. Reload the page to see the current values.