Limits¶

Each provider's window has a Limits tab. What it offers depends on how the provider is paid.

A CLI on a subscription¶

Step down a tier at [ ] % of the plan: once this much of the plan is used, agents picked by tier move one tier down. A whole number from 50 to 100, or empty for never. "A role that always runs on one tier, or on one model, is never stepped down: your own rule wins."

An API key¶

This key is a plan: count no money: its runs show no cost on Usage and count nothing in a project's budget.
Otherwise, At most $ … a day and At most $ … a month, spent at your prices, in every project. Empty means no limit. Past the limit, a new message whose Orchestrator runs on this provider asks you first, a model picked by tier moves to the next one of its tier, and no sub-agent or room member is opened here.

Press Save limits to keep them.

A model server on this machine¶

"A model server on this machine spends no money and has no plan: there is nothing to limit."

When a message would go over a limit¶

When a provider's limit or the project's monthly budget is reached, a new message asks first. It says which limit and how much was spent, then "Send it anyway?". Press Go on anyway to send it, or Cancel.

When an agent would move to a paid provider¶

If KOLONİ would open a sub-agent on a provider that is paid per token, while the chat runs on one that is not, it asks on the Open this sub-agent? card. The card's reason says that the new provider is paid per token and the other is not. Press Open agent or Reject. Your answer counts for that provider for the rest of the turn. In a turn nobody watches, such an agent is not opened.