Skip to Content
User guidesOverview

Overview

The overview is the home page. It shows how your agents are doing right now and what needs your attention.

Activity

The top of the overview

  1. Model calls. Model calls across all agents over the last 24 hours, in 10-minute columns. The greeting above it says how many agents are running.
  2. Runs. Runs started in the last 24 hours, with a bar for each hour and the count for the last hour.
  3. Guarded tools. How many of the tools your agents used in the last 24 hours are wrapped with a guard. A tool without a guard can only be recorded, never stopped.
  4. Block rate. The share of guarded tool calls that a guard blocked, per day over the last 30 days. A day above the dashed line, at 2%, is worth a look.
  5. Guard decisions. Calls blocked, and calls that asked a person, in the last 24 hours.

The overview updates itself every few seconds while it is open. Before your first run, every chart and list here is drawn empty.

SDK problems

A panel below the charts appears only when something needs fixing:

  • A policy file or signature feed that failed to load. Each message shows how many times it was seen and when it was last seen. The SDK keeps using the last good version until the file loads again.
  • Events lost. How many events SDKs dropped in the last 24 hours while they could not reach the backend, and when the last ones were lost.

What needs attention

The lists on the overview

  1. Approvals waiting. Up to five calls waiting for a person, with the count of every open request beside the title. Influenced by shows an untrusted origin that a value in the call came from. No longer waiting means the process that asked has stopped, but the request is still open. Click a row to answer it on the Approvals page.
  2. Recent runs. The latest runs, with their status, a strip of their guard decisions, their cost and when they started. Click a row to open the run.
  3. Incidents. The latest incidents, with their cause and the state of their replay.
  4. Decision log. The latest guard decisions of the last 24 hours: the time, the guard and its decision, the agent, the tool and the reason.
  5. Agents and guards. The agents that ran in the last 30 days, and whether each is running or idle. Below them, how many decisions each kind of guard made in the last 24 hours.

View all beside a list opens its full page.

Last updated on
Getting startedRuns