--- title: Agent instance page description: "What the Agent instance page covers: the strip above the tabs, the charts and cards under it, and how the tabs fit together." editUrl: true head: [] template: doc sidebar: hidden: false attrs: {} pagefind: true draft: false --- An agent instance is one run of an agent. You'll come here from the [Agent page › History tab](/web-app/agent/history) to see how a run unfolded: its status and timing, its risk score, the spans it recorded, and any alerts, quality verdicts, and token costs that came out of it. While the run is **Active**, you can also terminate it from the action menu. ![The top of the Agent instance page for a failed run, with markers on the Failed status badge (1), Risk profile (2), the Risk score card (3), Span activity (4) and Cost breakdown (5).](../../../assets/screenshots/web-app/agent-instance-page.png) 1. [Status](#above-the-tabs) - where the run is in its lifecycle. 2. [Risk profile](#above-the-tabs) - the profile the run is scored against. 3. [Risk score](#risk-score) - the run's total score and risk level. 4. [Span activity](#span-activity) - the run's spans by status over time. 5. [Cost breakdown](#cost-breakdown) - tokens and estimated cost of the run's model calls. ## What you can do here - Check a run's status, purpose, and timing at a glance. - See the run's risk score under the agent's [risk profile](/platform/concepts/risk-profile). - Follow span activity across the run, and see the tokens and estimated cost of its language model calls. - Reveal sensitive values recorded for the run. - Terminate a run that is still **Active**. - Start a hosted run and send it messages. - Open a tab to read the spans, alerts, risk, quality verdicts, cost, and details. ## Above the tabs The strip at the top names the run as "*agent name* instance", with the instance ID above it. It stays fixed while you switch tabs and shows: - Status — Preparing, Pending, Active, Complete, Failed, Cancelled, or Terminated. Only [judge runs](/platform/evals#judge-runs) are ever Preparing. Hover over a Terminated badge to see the reason given for termination. - Purpose — test runs carry a **Smoke test** or **Eval** pill beside the status. Live runs have none. See [Instance](/platform/concepts/instance) for what each purpose means. - Time — Complete and Failed runs show the start and finish times. An Active run shows its start time, a Cancelled run shows "Cancelled at" with the time, and a Terminated run shows "Terminated at" with the time. A Pending run shows "Ready to start", and a judge run that is still preparing shows "The input is still being prepared." - Summary — when the agent's metadata schema has a template, a one-line summary of the run's metadata. - Registered and Last updated — when Prefactor first recorded the instance and when it last changed. - Risk profile — the profile assigned to the agent, linked through to the profile. It reads "None assigned" when the agent has no profile, and "Unknown profile" when the assigned profile can't be loaded. **Show sensitive information for this instance** appears when the run's recorded input, result, or metadata contains [sensitive values](/platform/sensitive-data). Selecting it reveals those values in the summary and on the tab you're viewing. They're hidden again when you change tab. ### Terminate While the instance is **Active**, the action menu at the right of the strip holds **Terminate**. It opens the "Terminate agent instance" dialog, which asks for a **Reason**. The dialog's **Terminate** button stays disabled until you enter one. Confirming sets the status to **Terminated** and signals the running agent to stop. How the run winds down depends on the agent's SDK integration; see [Handle instance termination](/sdks/handling-termination). Prefactor does not close spans on termination: they keep whatever status they reached, and the agent's integration winds down anything still in flight. :::danger You can't undo termination. The run stays Terminated, and the reason you enter is kept with the instance and shown on the [Details tab](/web-app/agent-instance/details). ::: ### Hosted runs A run of a hosted agent version has extra controls. While it's Pending, **Start** begins the run. While it's Active, **Send message** opens a dialog where you type the next user message, and **Finish conversation** ends the conversation. Both stay disabled until the agent is waiting for input. ### Risk score When the agent has a risk profile, a **Risk score** card sits beside the strip. It shows the run's total score and risk level, with a bar that places the score against the profile's Medium, High, and Critical thresholds. The [Risk tab](/web-app/agent-instance/risk) breaks the score down by span type. Only spans whose type declares data categories or actions are scored, so until the run records one, the card reads "No risk value until there are spans". ### Span activity The **Span activity** chart counts the run's spans by status over time: active, complete, failed, and cancelled. It covers the run itself with an hour either side, and never less than 24 hours, so a short run sits near the right-hand end. The bucket size changes with the length of the window. ### Cost breakdown When the run recorded token usage on its language model calls, a **Cost breakdown** row shows four cards: **Input tokens**, **Input (cached) tokens**, **Output tokens**, and **Total cost**. The [Cost tab](/web-app/agent-instance/cost) splits the same totals by provider and model and explains how the cost is estimated. ## Tabs The page opens on the Activity tab. Some tabs only appear when the run has something to show in them. - [Activity tab](/web-app/agent-instance/activity) — the spans the run recorded, as a conversation or a raw list. - [Alerts tab](/web-app/agent-instance/alerts) — the alerts raised on the run. A red count on the tab shows how many are still active. - [Risk tab](/web-app/agent-instance/risk) — how the risk score was calculated. Appears when the agent has a risk profile. - [Quality tab](/web-app/agent-instance/quality) — the quality evaluations and judge verdicts for the run. - [Cost tab](/web-app/agent-instance/cost) — tokens and estimated cost by provider and model. Appears when the run recorded token usage on its language model calls. - [Details tab](/web-app/agent-instance/details) — the agent, version, and timestamps, plus the input, result, and metadata recorded for the run. ## Related - [Investigate a run](/guides/investigate-a-run) — find a run and work out what happened in it. - [Instance](/platform/concepts/instance) — what an instance is and the states it moves through. - [Handle instance termination](/sdks/handling-termination) — how the running agent finds out and responds. - [Risk profile](/platform/concepts/risk-profile) — how risk levels are derived from spans. - [Agent page › History tab](/web-app/agent/history) — the list of runs you came from. - [Agent page](/web-app/agent) — agent-level context, deployments, and configuration.