Obviously in order for your AI agents to make decisions, they need a periodic update. But, this doesn't have to happen all the time, and indeed the frequency that it occurs should vary depending on how important the agents are.
So, brain components periodically request that they get an update. The frequency at which that occurs and quickly it happens can vary. These are mostly controlled by the values in Settings.
Brains always request that they get an update on a fixed tick period. How frequent that is mainly depends on how close they are to any player. In Settings there are update request intervals rates for Near, Medium and Far ranges. Anything beyond "Far" is considered "Out of Bounds" (see below).
In addition, brains can receive early updates if their perception triggers a change (does not apply when they're out of bounds).
Agents outside the "Far" range will never request an update. If they were running an action and that had ongoing behaviour they can keep doing it to completion, but they'll never change their mind and if the action completes they won't do anything further.
The "Out Of Bounds Distance Check Interval" determines how often the brain checks to see if it's now within a range that it should get an update.
A brain will also not update if the agent has any of the tags defined in
PreventBrainUpdateIfAnyTagsin the Brain Config; which you can use to stop the brain updating on stun / freeze states
The main hard limiter for the AI update process is the "Brain Update Frame Time Budget Milliseconds" setting. Brains are queued for update periodically, and every frame that queue is processed. but as soon as the amount of (real) time taken equals or exceeds this frame budget (default 0.5ms, or about 3% of a 60Hz frame), then further brain updates are deferred until the next frame.
You can use this value to limit how much time the AI process can take from your frame budget, heading off spikes.
If an action is already running and is not interruptible, we abandon the update and continue to let it run.
All Action Defs in the Brain Config are combined into one big list, ordered by Priority Group.
For each priority group starting with the highest priority (lowest number), we run queries and generate contexts for each action, and score each using considerations. If the action (and context) matches the currently running action, there could be inertia applied.
If any of the action scores come out as non-zero, then an action is picked from that priority group (based on the action choice method e.g. Highest Score) and none of the lower priority groups are evaluated.
That action is then chosen to run. If another different action is already running, it is cancelled. If the one running is the same action (and same context) as has been picked again, then the action is allowed to continue (and is told of this).