Govern · Governance Mode
Off, Reactive, or Proactive. You decide how much an AI can do before a person sees it.
Most AI tools give you a single switch: on or off. Governance Mode gives you three real states, so the answer to "how cautious should this agent be" can be different for a marketing FAQ bot and a regulated-advice one.
Why a switch isn't enough
Caution has a cost either way.
Turn every guardrail into a hard stop and you've built an agent too slow and too cautious to be useful. Turn them all off and a mistake reaches your customer before anyone sees it. Governance Mode exists because most companies need both answers, for different agents, sometimes for the same agent on different days.
The three modes
What each one actually does.
- Off
- Watch words are recorded but never act on a reply. Used while you're still tuning what "wrong" looks like.
- Reactive
- An agent answers first. A match is flagged and logged the moment after, for a human to follow up on.
- Proactive
- A match holds the reply before it ships. The one mode where a customer never sees the flagged answer at all.
What counts as a "match" is set by Watch word feeds; a held reply lands in human-in-the-loop review.
FAQ
Companies setting this up are asking
Can Governance Mode be set per agent, not just company-wide?
Yes. A company sets a default, and a Team or Agent below it can run stricter, but never looser. That is the same inheritance rule the rest of the platform uses for guardrails.
Does Proactive mode make every response slower?
Only the ones a watch word actually flags. An ordinary answer ships the moment it is ready; a hold only ever applies to the reply that tripped a rule.
What triggers a hold in Proactive mode?
A watch word feed match (see Watch word feeds), configured to warn, block, or hold for review rather than pass silently.
Decide how cautious each agent should be, before it ships an answer, not after.
Twenty minutes with the founder. No slides. The real product.
