Executions & traces

The executions list shows executions newest-first, with:

• Status — Running, Needs approval (paused on a human decision), Completed, Failed, or Pending (queued to start).
• The execution's reference and the agent or workflow it ran — with its project, node count, the step it's on, tokens, and the start of any error.
• Trigger — what started it: Manual, Scheduled, Test, or Solution. See Triggers.
• Duration — or, while it runs, how far along it is, as a percentage.
• Started time.

You can filter by status (All, Running, Needs approval, Completed, Failed — each with a count) and search by name, or by id. Ids are typed by what was dispatched — workflow_execution_<hex> for a workflow, agent_execution_<hex> for an agent — and they're the same ids the API returns; to search by one, paste the hex part after the prefix.

Select any execution to open it.

A running execution reports how far along it is: the share of its workflow's steps that have finished. A step that fans out — a batching agent or a sub-workflow — fills in its share gradually as its items or steps complete, so the number moves smoothly rather than jumping. You'll see it in the executions list and on Console Home.

Open an execution that's still going and you'll see it happen in real time:

• a Running · live indicator — or Waiting for approval, with the tool it's paused on and a reminder to decide from the bell or the chat,
• each step as it starts and finishes, with done-of-total counts for batches,
• each tool call appearing as it's invoked, and marked done when its result returns,
• the agent's output streaming in as it's generated (its reasoning appears in the finished trace),
• and any error shown immediately if it fails midway.

When the execution finishes, the view switches automatically to the full trace below.

A finished execution opens as a two-pane trace explorer.

On the left is the step tree: every node that ran — agents, integrations, components — and, under each agent, the tool calls it made, shown as Tool Use: rows. Any row with something beneath it collapses and expands on click.

• A tool that's a workflow or agent names the version it ran (v3, v1 · Live, or Draft) and expands to show that child execution's own steps inline, all the way down.
• A batching agent lists each item — Item 1 of N — with its own steps.
• Tools the model provider ran itself, such as tool search, appear as tool rows too.
• A paperclip marks every step a file passed through.
• Failed steps, and tool calls that failed or were denied by a reviewer, are marked in red.

On the right is the detail for whatever you select:

• For a step: its type, start time, duration, tokens (in and out) and cost, the model for an agent, any files, and Input and Output tabs. An agent's Reasoning appears on its Output tab.
• For a tool call: its Input and Output, and for a workflow or agent tool, Sub-trace to open the child execution in place (Back returns).
• Any error is shown here too.

Together they let you follow exactly what happened, in what order, with the data at each step — the primary way to understand and debug an execution.

The detail view's header carries the execution's essentials at a glance: status, its name and reference id, what triggered it, when it started, how long it took, and the tokens and cost it used — including any child executions. If the run paused for approvals, a strip shows each request: who approved or rejected it, when, and how long the run waited.

Once an execution finishes, Copy trace JSON and Download trace JSON export the whole trace — handy for sharing a run or attaching it to a ticket.