Executions

The Executions page (/admin/executions, shown as Executions in the sidebar) shows the history of scheduled tasks the organization’s agents ran. Each row represents one run — a time a schedule fired and the agent responded.

What an execution is

An execution is one full run of an agent responding to a schedule (cron, interval, trigger). It includes:

• the initial scheduled call;
• the agent’s iterations calling tools;
• the final response or the error.

Executions coming from regular conversations (Hub, public channel, WhatsApp, API) do not appear here — for those, use the Conversations page or Analytics.

Filters

Four filters cut the noise (labels exactly as shown on screen):

• Agent: pick a specific agent (default: All agents).
• Status: Active, Completed, Pending, or Failed (default: All statuses).
• Start date and End date: limit the interval, with date picker.

Filters combine. Every change resets to the first page.

What each row shows

The table shows, per execution:

• Agent: agent name. If the agent was deleted later, you see Agent removed in grey.
• Schedule: name of the schedule that fired the run (taken from the conversation title).
• Date/Time: when the run started (dd/MM/yyyy HH:mm).
• Status: colored badge — green Active, blue Completed, yellow Pending, red Failed. When Failed and there is an error message, an alert icon appears next to it — hover for the full error tooltip.
• Credits: how much the run consumed. Shows — when not yet calculated.
• Actions: the eye icon (View conversation) jumps straight to that execution’s conversation.

Debugging failed executions

When an execution shows Failed:

1. Hover the alert icon to read the short error message in the tooltip.
2. Click the eye icon to open the full conversation.
3. In the conversation, expand each tool call to see the input sent and the response from the external service.

Common causes of failure:

• expired credential on a connected tool (reconnect in Tools);
• external service timeout;
• token limit exceeded;
• prompt error that makes the LLM answer outside the expected format.

Pagination

The table shows 20 executions per page. The footer shows N executions total on the left and navigation arrows on the right with current / total.

Best practices

• Set up a weekly routine to review Failed executions of the main agents.
• If a schedule keeps failing at the same step, suspect the involved tool first — open the tool’s account details to check for credential errors.
• Use the Agent filter when you already know only one agent is having trouble, instead of scanning the whole table.
• Credit spikes on specific runs usually mean tool loops or inflated context — worth reviewing the agent’s prompt.