Auxot
Tutorials Docs Pricing Login

Know what the Escalations bell is for

Open **Escalations** once, learn what a paused agent is asking you for, and tell the difference between a human-waiting card, a generic error, and a red **System Health** tile before you blame the model.

Plus: three Admin Agent prompts that separate escalation noise from outages, coach a one-minute response, and explain why a teammate sees cards you do not.

Audience Everyone
Time ~4 min
Prerequisites You can sign in and open **Chat** ([Say hello to the Admin Agent](/tutorials/say-hello-to-admin-agent)). Helpful: you already know where **System Health** lives ([Take Auxot's pulse in 10 seconds](/tutorials/take-auxots-pulse)) and you skimmed the left menu once ([Find the right screen for your next question](/tutorials/find-the-right-screen-for-your-next-question)).
You'll end up with A clear mental model of **Escalations** as the human pause inbox, the two actions you can take on a card, and three Admin Agent prompts you can reuse when Slack threads panic about *the AI being stuck*.

When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.

Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.

Why this matters

Agents keep answering in Chat until something in their instructions or a workflow tells them to stop and wait for a person. When that happens, Auxot opens an escalation: a card that says, in effect, I will not keep going without you.

If nobody teaches the team what that means, people treat the bell like spam, or they confuse it with model is down because the thread went quiet. The bell is not a popularity contest. Someone configured the agent to stop and wait for a person here on purpose.

Today, you open Escalations once, read the empty state or a real card, and learn the two buttons humans actually use. The next time a customer-critical thread pauses, you know which surface to open first.

Escalations are team-scoped in normal setups: you see cards for agents on teams you belong to. If your coworker sees a card you do not, that is often membership, not a broken inbox.

Quick start

  1. Open Escalations: find the Escalations entry (bell in the chrome or Escalations in the left menu, depending on your layout) and open it once. The page route is /app/escalations if you land there from a link.
  2. Read the empty state: if the list is clean, read whatever copy the page shows so you know what healthy looks like compared to quiet because you are in the wrong team.
  3. Name the two human actions: on any open card, locate Respond & Resume (you type guidance, then the agent continues) and Dismiss (you handled it elsewhere and you want the agent to move on without more back-and-forth from you).
  4. Contrast three surfaces: in your notes, write one line each for when you open Escalations versus System Health versus Audit Logs (Take Auxot’s pulse in 10 seconds, View your audit logs).
  5. Ask the Admin Agent for a one-paragraph definition: paste prompt 1 from The agent can do that? below so vocabulary matches how your org actually talks.

Done? You can explain to a coworker, without opening a laptop, why a paused chat thread might still be healthy if an escalation is waiting.

The agent can do that?

You have seen the page once. These prompts ask the Admin Agent to teach the room using Auxot vocabulary, not generic AI anxiety. Paste them in Chat → Admin Agent.

1. Separate escalation, outage, and audit trail

In four bullets, explain for a non-technical manager: (1) what an Auxot escalation card is, (2) how that differs from a red tile in System Health, (3) how that differs from a row in Audit Logs Jobs, and (4) what still belongs in Chat when none of the above apply. Use our product labels: Escalations, System Health, Audit Logs, Chat.

Why it’s non-obvious: Stuck thread sounds like one problem. Splitting human pause from platform problem from history saves hours of wrong-page debugging.

2. Coach a sixty-second response

I have one minute before a meeting. An escalation card says [paste the header or reason text]. Give me three sentences I can paste into Respond & Resume that are safe when I am missing half the business context, plus one sentence for when Dismiss is more honest than guessing.

Why it’s non-obvious: People freeze because they think approvals need essays. Short, honest guidance often unblocks the agent faster than a perfect policy memo you never finish.

3. Explain missing cards across teammates

Teammate A sees an escalation from agent X. Teammate B sees an empty Escalations list. List the most likely Auxot reasons (team membership, org, role, dismissed already) in order, each with the Settings or left-menu page to verify. No speculation about bugs until those checks fail.

Why it’s non-obvious: Silent I do not see it escalations to engineering are usually team membership and role, not a bug in Auxot itself.

Go deeper

Who configures when agents pause

Teaching an agent when to call for a human lives in agent instructions, workflows, and related product surfaces. Admins start with Set up an Escalation. This tutorial only covers what responders see when an escalation arrives, so the first card is not a surprise.

How this relates to heavier approval programs

When you need tiered policies, timeouts, and repeatable reviewer packets across tools, you graduate into Require human approval before risky actions and Run a workflow. Escalations are still often the first human-shaped signal people feel in Chat.

Severity icons are advisory

Escalation rows may show severity the agent reported. That severity icon is the agent’s own triage label. It does not route anything automatically. Humans still read the card before they act.

Walkthrough

Step 1: Land on /app/escalations

Use the bell or Escalations in the left menu. Confirm the URL if your layout sent you through a deep link. You are learning the inbox, not clearing every card on first visit.

Step 2: Read one row like a receipt

If a card exists, read Reason slowly enough to answer whether this is a judgment call (escalation) or a broken tool (often System Health or Audit Logs next).

Step 3: Practice the two buttons on a low-risk card only

If you have permission and the stakes are low, try Respond & Resume with a short, explicit instruction. If the situation already ended in email, Dismiss may be kinder than fiction.

Step 4: Cross-check System Health once

Open System Health in a second tab. If nothing is red, return to the escalation story instead of digging through provider dashboards looking for problems that are not there.

Step 5: Peek Audit Logs if history is the fight

If the debate is who triggered what, open Audit Logs and use the tab that matches jobs versus threads versus events (View your audit logs). You are looking for what actually ran and when, not for a hunch about why something failed.

What’s next

Reference

Browse all tutorials