Configure AI settings

All AI configuration for your organization lives in one place: Settings → Artificial Intelligence. The provider credentials and budget controls that previously lived on separate pages now open here, so you manage model quality, keys, and spending limits without leaving this section.

Set the Effort dial

The Effort dial sets which level every AI task in your organization runs at by default. It is four buttons, ordered from lightest to heaviest, each naming what it is for:

• Quick: fastest responses for routine tasks.
• Standard: speed and depth in balance. This is the default.
• Deep: deep reasoning for complex work.
• Expert: the most capable model, always reasoning, and the highest cost.

There is one dial, and every AI task in your organization runs at the level you pick here, unless a person’s own Effort choice or a one-off pick for a single run overrides it. (The earlier per-task table, which let you nudge individual tasks up or down on their own, has been removed: Effort is now a single choice of level.) The dial never names a model, because which model backs each level is a separate setting you can override: see Set effort defaults for your organization below.

Choose which models your users can use

Every model your platform enables is available to your organization by default: nothing is removed until you remove it yourself. The Models area of this section lists the models currently available to your organization, each with a remove button on its row.

Remove a model and it disappears from every picker in your organization. To bring it back, choose it from Add from platform catalog, which lists every platform-enabled model you have removed. You cannot remove the last remaining model: at least one must stay available to your users.

When the platform enables a new model later, it arrives already in your organization’s list: if you keep your organization on a fixed, curated set, check back here after a platform update rather than assuming a new arrival stays out until you act.

Set effort defaults for your organization

Each Effort level starts on a platform default model. From the Models area, you can override any level for your whole organization instead: choose the model that runs it, and where that model supports extended thinking, how much budget to give it. You only need to set the levels you want to change: every level you leave alone keeps running on the platform default. The Models area is not offered on the free edition, which runs on its own key and on the platform defaults.

This setting decides which model answers once a level has been chosen, not which level a run uses in the first place: see How DeltaVault picks the Effort for a run below for how DeltaVault picks the level itself.

If a model behind one of your overrides later becomes unavailable, its row shows a stale badge. While the badge shows, that level quietly runs on the platform default instead, so nothing breaks while you pick a replacement or clear the override.

How DeltaVault picks the Effort for a run

There is no personal Effort setting shadowing your organization’s: every control opens at the organization’s level, and nothing a person picks is kept as a saved default of their own. What anyone in your organization can do is choose a different level for a single run, from the AI surface they are working on. The Effort picker appears on the Ask AI chat, the inline edit-instruction line, the Workshop launch form, and the Map and match panel. It is not on every launch form: the Discovery form has no Effort control, and Model review has its own Depth setting instead. The picker lists the same four levels by name, and a hint marks whichever one your organization has set as its default. It does not name models.

What a run gets follows a simple order: the level picked for that run, otherwise the skill’s own authored default, otherwise your organization default, otherwise Standard. A skill’s authored default outranks your organization default deliberately, because it is administrator-set configuration rather than a stale personal choice, so a heavyweight skill authored as Deep never quietly runs Quick. Choosing a level for one run needs no special permission, because a higher level only spends that person’s own budget faster, and the usual cost check still applies before each run. The organization default itself stays under the editor permission and the administrator lock described below.

Set Effort for one run

Some launch points let you set Effort for a single run, without changing your organization default. When you consolidate knowledge documents, the Consolidate dialog shows an Effort picker that starts from the Effort the run would otherwise use, so you can raise it for one demanding merge and leave everything else as it was. When you upload a document and choose “Let AI decide”, an Effort picker appears for that classification run and starts on Quick, because classification is light work. In both cases the choice applies to that one run only.

Bring your own provider key

You can connect your organization’s own AI provider credentials and, where available, pin an AI region. When you bring your own key, AI calls run against your account rather than the platform key.

See AI regions and bring your own key for regional residency details and the step-by-step credential setup.

Set budgets

Budgets let you control how much AI your members can use when your organization runs on the platform key. Allowances are set in credits: you set per-tier monthly credit allowances and concurrency limits, and the per-member tier table maps each member to an allowance based on their tier.

When your organization runs on the platform key, your allowances are capped at the platform ceiling. When your organization brings its own key, the platform ceiling does not apply and your own provider account limits govern instead.

On the free edition the Budgets card is not offered: AI runs on your own key, so there is no platform spend to cap, and what limits your spend is your own provider account.

For what a credit is and how members spend against these allowances, see AI credits.

Lock the section

An administrator can lock the Artificial Intelligence section so only administrators can change these settings. When locked, non-administrators see all values as read-only. This works the same way as the lock on the Conventions page.

The AI recommendation workflow setting, which controls how the map-and-match workbench handles AI suggestions, also lives in this section.