Agent Runs
Find any run your agents did, see how it ended, and open it to read what happened.
Overview
Every time an agent works, it creates a run. A run starts when an agent answers an email or a chat message, runs a scheduled task, or does its own background work. The Agent Runs page lists those runs and lets you troubleshoot any one of them without leaving the page.
The page has three panes:
- Run list (left). Search, filters and runs grouped by day. Each row says what the run was about and how it ended.
- Run view (middle). The selected run's header, its summary, its issues and a likely fix, a "Where it went" breakdown of time and cost, and a trace of every step.
- Step inspector (right). The selected step's input, output, prompt, response, tokens and raw log entries.
A typical use: find one run an agent did, read the steps it followed at a glance, then open the details and raw log of any step without losing your place in the list.
This article covers the run list. Reading a run covers the run view and the step inspector.
Watch: Agent Runs: see what an agent did (3:47)
Where to find it
| Hub | Menu path |
|---|---|
| Governance | Governance > Activity History > Agent Runs |
| Boardroom | Boardroom > Monitor > Agent Runs |
| PA Hub | PA Hub > Monitor > Agent Runs |
In the Boardroom and the PA Hub, Agent Runs is an advanced page. It shows in the menu only after you turn on Show advanced pages under Profile > Preferences. The page itself is always there, so a saved link still opens it.
You see the runs of the agents you supervise and of your own Personal Assistant. Platform Admins see every run in the tenant.
Layout by screen width
- Wide screens (1200px and wider). The three panes sit side by side. Each pane scrolls on its own.
- Narrower screens. The run list sits above the run view, and the page scrolls. The step inspector opens as a panel that slides in from the right when you click a step.
The list header
The top of the list shows:
- Agent runs, with a count of the runs that match your filters, and a Refresh button.
- A summary line, for example "3 need a look · 1 failed · $4.12 spent". It covers the whole time range, source, agent and search, whichever chip is active.
Filtering the list
Time range
| Option | Shows |
|---|---|
| Past hour | Runs from the last 60 minutes |
| Past day | Runs from the last 24 hours (the default) |
| Last 7 days | Runs from the past week |
| Custom range | Runs between a start date and an end date you pick |
If a time range holds more runs than can be searched at once, a line says so and the list shows the newest ones. Narrow the time range to see older runs.
Source
The Source control picks what kind of run to show:
| Source | What it shows |
|---|---|
| All sources | Every kind of run (the default) |
| Agents | Runs by your agents |
| Dream cycles | Each agent's background dream cycle runs |
| Weekly self-reviews | Each agent's weekly self-review runs |
| Mailbox scans | Mailbox scan runs |
| Safety Gateway tests | Safety Gateway test runs |
| Outermind support access | Every time Outermind support read your customer-hosted data |
Mailbox scans, Safety Gateway tests and Outermind support access show only for Platform Admins, because those runs belong to no agent.
Agent
When Source is Agents, Dream cycles or Weekly self-reviews, an Agent control appears. Pick All agents or one agent. The list offers only the agents you can see.
In the Boardroom, the menu link opens the page with Source set to Agents and the agent set to your AI Chief of Staff. Pick All agents to see every agent, and the page remembers that choice when you reload.
Search
The search box reads "Subject, person, tool or run ID". It matches:
- Subjects, such as an email's subject line or what the run was about.
- People, such as a Personal Assistant's owner or an agent's name.
- Tools the run used.
- Run IDs. Paste a full run ID to open that run, even when it is outside the current time range.
Press / to jump to the search box. Press Esc in the search box to go back to the selected row.
Quick filter chips
Four chips narrow the list. Each shows its count.
| Chip | Shows |
|---|---|
| All | Every run except skipped ones |
| Attention | Runs that failed, were escalated, finished with issues, are waiting, or were refused by a policy |
| Failed | Runs that failed |
| Waiting | Runs that are waiting, were escalated to a person, or are queued to retry |
Cancelled and deferred runs show only under All.
Skipped runs
A skipped run is one that started but had nothing to do. These runs are hidden by default so they do not crowd the list.
- With the Skipped toggle (the eye icon) off, each day with skipped runs shows one striped row under its day header: "12 runs had nothing to do · Show". Click Show to add that day's skipped runs to the list in time order. Click Hide to fold them again.
- With the Skipped toggle on, skipped runs show inline under every chip, in time order.
The page remembers the Skipped toggle for you in this browser.
Reading a row
Runs are grouped under day headers such as "TODAY · 18 runs · $5.55". The day and the times follow your tenant's time zone.
Each row has two lines:
- Line 1: a status dot, the run's title and its clock time.
- Line 2: a short badge, what started the run, how long ago it started, its duration and its cost, then issue icons.
| Part | What it means |
|---|---|
| Status dot | Green means it completed. Amber means it completed with issues or was stopped by a policy. Red means it failed. Violet means it is waiting or was escalated. Grey means it was skipped, cancelled or deferred. Blue means it is still running. Hover over the dot to read the status in words. |
| Title | What the run was about, such as the email subject. When there is no subject, the title comes from the agent's own summary or from what started the run. |
| Badge | CoS for the AI Chief of Staff, PA for a Personal Assistant, SME for a specialist agent, AIPSO and Agent for other agents. Runs with no agent show Dream, Review, Scan, Gateway or Support. Hover over the badge to read the full agent name. |
| Owner | A Personal Assistant run shows its owner's name after the badge. |
| Duration and cost | How long the run took and what it cost. A cost of $0 is left out, and a cost under half a cent shows as "<$0.01". A run that is still going shows "running" and a timer. |
| Issue icons | A red icon for an error, a violet hand for something held or waiting for a person, and an amber triangle for a warning. Hover over an icon to read the issue. |
Some rows look different:
- Agent message rows show the badge "Agent message" with a dimmed, dashed style. They record a message one agent sent another, with no run behind it. They have no duration or cost and are not counted in any total.
- Privacy-locked rows belong to a Personal Assistant whose privacy policy hides run details. They show a lock icon, "Details hidden by your PA privacy policy", the status, time, duration, cost and the owner's name, and no issue icons.
Moving through runs
- Click a row to open that run in the run view.
- Newer and Older, in the run header, move to the next run in the list's order.
- Scroll down to load more runs. The list loads 50 runs at a time.
When you change a filter, the open run stays open. If it no longer matches the filter, no row is highlighted and the run header reads Not in this filter. A skipped run opened from a day's fold reads "Skipped run" and its day instead.
Keyboard shortcuts
Shortcuts work while your focus is inside the page, and not while you are typing in a field.
| Key | Action |
|---|---|
j / k | Open the older / newer run |
| ↓ / ↑ | With focus in the list, move to the next / previous run |
| Home / End | With focus in the list, move to the first / last loaded run |
| Enter or Space on a skipped-runs row | Show or hide that day's skipped runs |
/ | Jump to the search box |
| Esc in the search box | Go back to the selected row |
A key legend at the bottom of the list reminds you: j k runs · ↑ ↓ list or trace · 1-5 tabs · / search.
Live updates
The list updates on its own while you watch.
- A new run appears at the top of today's group when the list is scrolled to the top, the chip is All and the search is empty. Otherwise a "N new runs" pill appears at the top of the list. Click it to scroll up and add them. The open run does not change.
- A run that finishes updates its dot, duration, cost and issues in place.
- While any run in the list is still going, the list refreshes about every 12 seconds.
- Click Refresh to reload the list at any time.
A custom range that ended before today does not update.
Sharing a view
The page's web address holds the time range, source, agent, chip, search and open run. Reload the page or share the address to show the same view. To share one run or one step, use Copy link in the run view or the step inspector. See Reading a run.
What a Personal Assistant owner sees
In the PA Hub, Agent Runs shows only your own Personal Assistant's runs. The Source and Agent controls are hidden because they are fixed to your assistant.
- If your Personal Assistant is not set up yet, the page says "Your Personal Assistant has not been activated yet."
- If your assistant's status cannot be loaded, the page says "Failed to load Personal Assistant status."
Your organization's PA privacy policy decides how much of your assistant's runs other people can see. Depending on the policy, they see a privacy-locked row with no content, or they do not see the run at all.
Runs in the agent admin page
An agent's admin page has a Runs section that shows a compact version of this list for that one agent. It has the time range, the chips and the Skipped toggle, but no Source, Agent or search control. Click a row to open the run on the Agent Runs page.
Auto-resume
When a run hits a temporary problem, such as an LLM provider rate limit or an exhausted credit balance, the platform can queue it to retry on its own instead of failing it.
- A run that is queued to retry shows under the Waiting chip.
- Its run view lists an issue that says which attempt is next, for example "Will retry (attempt 2 of 5)".
- A run that ran out of retries lists the issue "Gave up resuming".
If you see repeated retries for the same run, the problem may not be temporary. Check your LLM provider's account balance and status.
Common tasks
Find failed runs
- Pick a time range.
- Click the Failed chip.
- Click a run to read its issues and likely fix.
Find what an agent did for one email
- Type part of the email's subject, or the sender's name, in the search box.
- Click the run.
- Read its summary and trace in the run view.
Check costs
- Read the summary line for the total spent in the time range.
- Read each day header for that day's total.
- Open a costly run and read Where it went to see which part of the run cost the most.