Set the platform's AI limits and models

Cap what the platform's AI key spends each month, choose which model does which work, and which plans may use it.

In short

The platform's own AI key answers people who have no key of their own. In Admin console › AI, "Spending" caps what that key may spend each month, and "Models" says which model does which work and which plans may use the key. People's own keys are never limited here.

Before you start

• Who: only a site admin. The Admin console is in the avatar menu of a site admin and nowhere else.
• Plan: none. A plan's own AI budget is one of its limits; see /help/settings/admin-plan-limits.
• First: the server operator configures the platform key. Without one, "Models" says "No platform key is configured, so only people's own keys answer."

Set the monthly limits

1. Open "Admin console" › "AI". It opens on "Spending".
2. Go down to "Monthly limits on the platform's key". The bars show this month against each cap, and a line says whether platform AI is running or paused.
3. Set "Monthly cap for everyone (USD)" and, inside it, "Free and trial share (USD)". A share of 0 pauses the platform's AI for free and trial workspaces.
4. Set "Pause free and trial at (% of the cap for everyone)" and "Tell site admins at (%)": up to six percents, such as 50, 80, 100.
5. Press "Save limits".

Result: free and trial workspaces are paused first, at their share or at the percent you set. Once the cap for everyone is spent, platform AI pauses for all until it resets on the date shown. People's own keys keep working.

Choose the models

1. Open the "Models" tab. Its first line names the provider and model that answer, and what one typical answer costs.
2. Type a "Light model" and a "Strong model" the way the provider writes them, or leave them empty to use the one shown. Summaries, short answers and everything a free or trial workspace asks run on the light model; generation and design run on the strong one.
3. Under "Model by feature", each feature shows its cost - Light, Heavy or Documentation - and the model it uses in paying and in free workspaces. Type in its "Own model" box to give it a model of its own; empty follows the light or the strong model.
4. Press "Save models".

Choose which plans use the platform's key

1. On the "Models" tab, go down to "Plans and the platform's key". Each plan shows its "AI budget a month" and about how many answers it buys.
2. Turn "Light work" or "Heavy work" on or off for a plan, or for "Trials". Each switch saves at once.

Questions about how the product works are answered on every plan. Work switched off here still runs on a person's own key.

Good to know

• Above the plans, "Light work", "Heavy work" and "Reading a data question" show what a typical and a heavy call really cost in the last 30 days, once there are enough calls to measure.
• Estimated costs on "Spending" use the prices you set with "Add price".
• Each percent in "Tell site admins at (%)" that this month's spending passes opens a case in Admin console › "Abuse", once a month.

If something goes wrong

• "The free and trial share cannot be more than the cap for everyone.": lower the share or raise the cap.
• "Enter a whole percent from 1 to 100." / "Enter up to six whole percents from 1 to 100, separated by commas."
• "Free and trial workspaces are paused": their share is spent. It resets with the month, or raise the share and press "Save limits".

Related

• /help/settings/admin-more-tabs
• /help/ai/ai-usage-report
• /help/ai/providers-and-credentials
• /help/settings/admin-plan-limits