Agentic Executions

The Agentic Executions section gives you visibility into all autonomous agent jobs triggered for your data set — their status, results, and the internal conversation log produced during each run.

History Dashboard

Navigate to Configure → Agentic Executions in the left sidebar to reach the history dashboard.

The dashboard lists all past executions for the current data set, newest first. You can filter the list by:

Agent — narrow to executions from a specific configured agent
Status — filter by pending, in_progress, finished, or failed

Filter state is preserved in the URL, so the view survives a page refresh and can be bookmarked or shared.

Launching an Execution

Click New Execution to open the launch form. The form walks through three steps:

Select an agent — choose from agents configured for the current data set.
Select an execution type — after choosing an agent, the available execution types are populated dynamically from the server. Each type shows a name and description to help you pick the right one.
Provide input — fill in the fields derived from the execution type’s input schema. If the selected agent supports file uploads, a file attachment field also appears.

Field-level validation errors and API errors are displayed inline. On success, you are taken directly to the new execution’s detail view.

Execution Detail View

The detail view shows the full record for a single execution:

Status — refreshes every 3 seconds while the execution is running. Polling stops automatically once a terminal state (finished or failed) is reached.
Result — shown for finished executions: the structured JSON output returned by the execution definition.
Failure details — shown for failed executions: the failure code and the LLM-generated explanation of what went wrong.
Conversation link — a link to the internal conversation created for this execution, giving full visibility into the agent’s message loop, tool calls, and intermediate reasoning.

Execution Conversations

Each agentic execution creates an internal conversation to drive the agent’s message loop. These conversations do not appear in the standard agent conversation history — they represent an implementation detail, not a user-initiated dialogue.

You can inspect an execution’s conversation via the link on the detail view. The conversation is read-only: the text input and send button are disabled, with the placeholder “This conversation is read-only.”