# Manage the workflow AI budget

Administrators can open **Settings → AI Token Budget** to set the daily ceiling for workflow AI steps across the organisation. This is separate from a conversation's model choice.

## Read the usage

The card shows tokens used today, the effective daily limit, and the number of AI calls. The accounting day resets at midnight UTC. Shared use means another workflow can consume budget while you are inspecting it.

## Set a limit

Enter a whole number of tokens between **10,000** and the displayed default ceiling. The deployment's default is also the maximum an organisation administrator can choose. Select **Save** to apply the change immediately.

If the new cap is at or below today's usage, the confirmation explains that workflow AI will pause for the rest of the UTC day. Choose **Set cap and pause** only when that is the intended outcome. Lowering a limit does not undo tokens already consumed.

## Use the default

Choose **Use default** or clear the input, then save. This inherits the deployment's default rather than storing a fixed copy of today's number. Later default changes can therefore affect the organisation.

## Plan around exhaustion

Some workflows fail or pause on quota exhaustion; a workflow with an explicitly configured degradation contract can use its defined fallback. Topic News can retain snippets where model summarisation is unavailable. Do not assume every AI workflow has that fallback.

Generation can reserve a token allowance before a provider call. A request with uncertain usage can retain its reservation until resolved. Check its saved receipt before starting another billable generation.

If Save fails, read the error and refresh the current usage and policy. Do not claim the new cap is active until the saved value reflects it.