Observability
Dashboards: your own views of the numbers
Pages you make yourself: pick from 17 widgets (workflow runs, conversations, tokens and cost, experiments, scores, alerts, credits and traces), arrange them, and watch them for today, 7 days or 30 days. Everyone in the workspace sees them.
Guide5 min readUpdated 1 November 2026
Dashboards, under Observe in the left menu, are pages you put together: pick widgets from a catalogue, arrange them, and see live numbers side by side. A workspace can have many, one for each team or agent, and everyone in it sees them. The numbers come from what SentientOne already records (workflow runs, conversations, replies, experiments, scores, alerts and credits) and are worked out each time the page loads; a dashboard keeps only its name, description, period and widgets.
Create a dashboard
- 1
Open Dashboards
Click Dashboards under Observe and choose New dashboard. New dashboard is also in Search (⌘K).
- 2
Name it
Type a Name (up to 80 characters) and, if you like, a one-line Description.
- 3
Choose the default period
Today (the last 24 hours, hour by hour), 7 days (the default) or 30 days (day by day, in UTC). It's what the widgets count over when the dashboard opens.
- 4
Pick the widgets
Tick as many as you like, up to 24. Each card draws a sample of the widget; the numbers on this page are made up.
- 5
Create it
Choose Create dashboard, or press ⌘↵ (Ctrl+Enter on Windows). It opens with live numbers, the widgets arranged so that each row is full.
You're done when
The dashboard shows your widgets with numbers, and the Dashboards page lists it with its widget count, its period and who made it.
The widgets
17 widgets in six groups. A small widget takes one column, a chart or a list two, a table the full width: four columns on a wide screen, two on a tablet, one on a phone. A widget's title opens the page with the full picture, and its corner says what it covers: the period, or Now, Upcoming or Latest for the ones that don't count over one.
| Widget | Group | Shows | Period |
|---|---|---|---|
| Workflow runs | Workflows | Live runs of your workflows, how many succeeded and failed, and a trend | Yes |
| Failed runs | Workflows | Live runs that failed, out of all runs | Yes |
| Run success rate | Workflows | The share of finished live runs that succeeded | Yes |
| Run duration p95 | Workflows | 95% of finished live runs took less than this | Yes |
| Runs over time | Workflows | A chart of succeeded and failed runs per hour or day | Yes |
| Next scheduled runs | Workflows | Published workflows with a schedule, soonest first, with each one's next run | Upcoming |
| Conversations | Agents | Conversations with your agents on every channel, and a trend | Yes |
| Conversations over time | Agents | A chart of new conversations per hour or day | Yes |
| Tokens and cost | Agents | Tokens your agents used, the estimated cost at each model's rates, and the replies | Yes |
| Agents by environment | Agents | A table of the release each agent runs in Staging, Production and your other environments | Now |
| Experiment pass rate | Evaluate | The share of passing scores across experiments finished in the period | Yes |
| Latest experiments | Evaluate | A table of your newest experiments, their status and headline score | Latest |
| Online evaluation scores | Evaluate | The average score each online evaluator gave live replies | Yes |
| Review queue backlog | Evaluate | Replies waiting for a person, across your review queues | Now |
| Open alerts | Alerts | Alerts that are firing, with their last value | Now |
| Credits | Credits | The organization's balance, what this workspace spent this month, and its monthly limit if it has one | Now |
| Trace p95 and error rate | Tracing | From traces: 95% of requests finished faster than this, and the share that failed | Yes |
Eight widgets can show one agent instead of all of them: Conversations, Conversations over time, Tokens and cost, Experiment pass rate, Latest experiments, Online evaluation scores, Open alerts, and Trace p95 and error rate. The first three can also show one environment, such as Staging. A widget that shows one agent or environment says which on a chip under its title.
Change a dashboard
Choose Edit. Nothing is saved until you choose Done:
- Rename it or change the description in the header, and use the period buttons to change its default period.
- Add widget opens the catalogue. Search it by name or by what a widget shows, and pick an Agent and an Environment first for widgets that can show one (the others ignore them). Each Add puts a widget at the end, drawn as an outline of its shape until you save; then it shows live numbers.
- Each widget has a handle to drag it onto another widget (or use the arrow keys on the handle), Move left and Move right, Resize (small, medium or large, as the widget allows) and Remove.
- Done saves everything (⌘↵, or Ctrl+Enter on Windows). Cancel puts it all back as it was.
If someone else saved the dashboard while you were editing it, Done says so instead of overwriting their changes. Choose Reload, then edit again.
Period, agent and refresh
- The Today, 7 days and 30 days buttons switch the period for you, not for everyone. The page's address keeps it (
?period=30d), so you can bookmark or share that view; the dashboard's own period drops it again. - With an agent picked in the top bar, the widgets that can show one agent show that one, unless a widget was added with its own agent. The others say Whole workspace on a chip, so you can tell which numbers the agent applies to.
- Refresh reads the numbers again (numbers under 30 seconds old are reused, so a second Refresh straight after the first shows the same). Auto-refresh reads them every 60 seconds while the page is open and on screen, not while you edit; your browser remembers the choice for every dashboard.
- Each widget loads on its own, so one that is slow or fails says so in its place while the others show.
Who can do what
| Workspace role | Dashboards |
|---|---|
| Viewer | Sees every dashboard, switches its period and refreshes it. No Edit or delete. |
| Editor or admin | Also creates, edits and deletes dashboards. |
Delete a dashboard
Choose More actions (⋯), then Delete dashboard, and confirm. It's deleted for everyone in the workspace; the runs, conversations and scores it showed aren't touched.
Limits
Up to 24 widgets on a dashboard and 50 dashboards in a workspace. Names up to 80 characters, descriptions up to 300.
If something looks wrong
A widget says “Tracing is off”
Trace p95 and error rate reads the trace store. Where traces aren't recorded, such as a test environment, it says Tracing is off; the other widgets aren't affected.
A widget took too long to load
It says so in its place while the others show. Choose Refresh to try again.
A widget says its environment isn't in this workspace any more
The environment it showed was deleted. Remove the widget, or add it again with another environment.
Conversations doesn't match Home
The widget counts every conversation in the workspace, workflow chats included, like the Conversations page; Home counts your agents' conversations. Both use the same periods.
Related
Still stuck?
Email us with what you tried and any request id. A person replies.