Tracking runs

Every time an automation runs it records what it did, what it found, and what it cost. The run history is how you trust an automation, you can see it is working, and see what it surfaced without waiting for a notification.

Prerequisites

Any workspace member can view run history. Creating or changing automations needs the Editor role.

What a run records

• Status: Pending, Running, Completed, Failed, or Stopped. A run that started a deep web research shows Waiting for research until the report arrives, then continues

• Output: The full result the agent produced

• Findings: The specific things worth attention, extracted from the output

• Duration and cost: How long it took, and the token usage it consumed

• Tool calls: What the agent actually did, which queries and tools it ran

Because a run captures the exact prompt it used and the tools it called, a surprising result is auditable: you can see how the automation reached it, not just what it said.

Stopping a run

A run that is queued or in progress can be stopped, for example when you started it by mistake or it is taking far longer than usual.

1. Open the automation, or the run itself from its run history.

2. On the run’s status panel, click Stop run.

3. Confirm “Stop this run? It ends now without a result.”

The run ends straight away and is marked Stopped, “This run was stopped before it finished”. It produces no result, and the automation runs again at its next scheduled time, or when you trigger it manually. If the run finished before your click arrived, Adverity Atlas tells you “This run has already finished.”

Only the automation’s creator, or an Admin, sees Stop run.

Findings and delivery

When a run finds something, it delivers it two ways:

• Posts to chat: The run opens (or continues) a conversation with its result, so you can ask follow-up questions right there

• Notifies the creator: Whoever created the automation gets a notification linking to the run, so a finding is not missed

Both are set per automation, on the Output section of its create and Edit panels, see Creating automations. The notification is sent on every completion, whether or not the run found anything. Every run is also recorded as an execution on the Task Executions page, described below.

Task Executions

An automation run is one of the background tasks Adverity Atlas records for your workspace. Open Tasks from the account menu (the avatar in the top bar) to reach Task Executions, described in-product as “Background tasks for this workspace.”

The list shows one row per task, with these columns:

Column	What it shows
Task	The name of the task, with a one-line description of what that kind of task does
Status	Pending, Running, Completed, Failed, Timed Out, or Cancelled
Error	The failure reason, where there is one
Created	When the task was queued

Narrow the list with the two filters above it, All Tasks and All Statuses. The All Tasks filter lists the kinds of task this page actually holds, named the way the rows are, for example Chat response, Run automation, Profile table, or Fit marketing mix model. Every option returns at least one row, so narrowing can never leave you looking at an empty page. A workspace’s own Tasks page lists only what can appear in a workspace, so account-wide work such as data source syncing does not clutter the choices there.

A running total sits above the list with a breakdown by state, and it follows whatever you have filtered by, so you can see at a glance how much is queued and how much is actually moving. The list refreshes on its own while you watch it.

Tip

A large pending count with nothing running is the signature of a backed-up queue. Filter to Running to confirm before assuming a single automation is at fault.

Admins see an All workspaces link in the page header, which opens the same view across every workspace in the tenant. It is the place to answer “what is this tenant doing right now” rather than one workspace at a time.

Health and auto-pause

A failed run always notifies the automation’s creator, whatever the Output settings say. After five consecutive failures the automation is paused automatically and its creator is notified again. This stops a broken automation from running, and costing, indefinitely. Fix the cause and re-enable it.

Aggregate cost across all automations is visible in Cost monitoring, broken down by workspace and model, useful for spotting a single expensive automation.

What’s next?

• Cost monitoring, automation spend across the tenant

• Creating automations, adjust or add automations after reviewing runs

• Artifacts, outputs a run can save

Was this page helpful? Yes No