Govern · Human-in-the-loop review (HITL)
A person sees it before it ships. That's the whole feature.
Human-in-the-loop, or HITL, means exactly what it sounds like: for the answers that matter, a person is in the loop, not just watching a dashboard after the fact. Augela builds this as a real queue with a name, not a slide about "responsible AI."
Why "we review things sometimes" isn't a system
Oversight that depends on someone remembering to look isn't oversight.
A lot of AI deployments promise human oversight and deliver a dashboard nobody has time to watch. HITL review here is a mechanic, not a promise: specific triggers put specific answers in front of a specific person, before or the instant after they ship, with the context needed to actually judge them.
"…and yes, we can match that price for your renewal."
How it works
The mechanics behind that one card.
- Where it can sit
- On a thumbs-down, a watch word match, a pipeline handoff you mark sensitive, or every reply under Proactive mode.
- What a reviewer sees
- The full context, and Test in the agent: one click opens its chat with the question typed in.
- Review, then Fix
- The fix goes to the agent that answered: company memory, its AI Profile, or a rule. A fix that changes nothing is refused.
- Who can do it
- Anyone with the Content Reviewer or Tutor permission. Admins have it automatically.
What triggers a hold is set by Governance Mode and Watch word feeds; what a fix teaches becomes Company memory. The same queue brings "Still true?" checks when a memory item is due, and marks the documents behind a wrong answer so the source gets fixed too.
FAQ
Companies setting up review are asking
Is human-in-the-loop review the same thing as Insight elevation?
They're related but not the same. HITL review is where an answer is checked and fixed. Insight elevation is how a good find rises from one agent to its team or the whole company. Both feed Company memory.
Does every agent need a reviewer watching it constantly?
No. A checkpoint is switched on for specific triggers: a watch word match, a pipeline handoff you've marked as sensitive, or Proactive Governance Mode. It is not switched on for every message an agent sends.
Who can review? Does it have to be an admin?
Anyone an admin gives the Content Reviewer or Tutor permission. A reviewer sees exactly the conversation and context that produced the answer, not just the answer itself.
What happens to a rejected answer?
It never reaches the customer. The reviewer fixes it in two steps, Review then Fix, and the fix goes to the agent that answered: as company memory, as an instruction in its AI Profile for tone or style, or as a rule. A fix that changes nothing is refused.
Put a person in the loop exactly where it matters, not everywhere, not nowhere.
Twenty minutes with the founder. No slides. The real product.
