Skip to main content

Overview

Dashboards allow you to organize and view multiple reports in a single place, giving you a comprehensive overview of your voice AI performance. Create separate dashboards for different teams, use cases, or monitoring needs. Define a dashboard in code (recommended), or build it in the dashboard UI.

Manage it with the CLI

List dashboards and read a dashboard’s numbers from the CLI:

Creating Dashboards

  • Multiple Dashboards: Create separate dashboards for different use cases (e.g., Production Monitoring, QA Review, Executive Summary)
  • Add Reports: Select which saved reports to include in each dashboard
  • Reorder: Drag and drop reports to organize your dashboard layout

Property Overrides

Filter an entire dashboard by specific call properties without creating new reports:
Common Use Cases:
  • View all reports filtered to a specific business name or customer ID
  • Analyze performance for a particular agent or integration
  • Compare metrics across different property values
  • Quickly switch between filtered views without leaving the dashboard
When you override properties at the dashboard level, all reports on that dashboard automatically update to reflect the filter, making it easy to drill down into specific segments of your data.

Set or missing

A custom property filter can also check whether the property is there at all. Pick the property, then is set or is not set; there is no value to enter.
  • is set keeps conversations that have a value for the property, an empty string included.
  • is not set keeps conversations missing it: never sent, sent as null, or removed later.
A property no conversation has ever carried is missing on all of them. The filter reads as, for example, plan_tier is not set. It works on a widget, on a comparison group and on the whole dashboard. The calls list has the same filter.

Filter by another metric

Narrow a widget to the conversations where a different metric came out a certain way. For example, show the pass rate of Retrieved accurately only over the calls where Information given is true: calls where the agent gave no information can’t retrieve it accurately, so they would only drag the rate down. The Another metric filter is in the filter menu of:
  • The widget editor, to narrow one widget. It is offered on every widget except uptime.
  • Comparison groups, so each group can be a different slice, for example “Information given: true” against “Information given: false”.
  • The dashboard filter bar, to narrow every widget on the dashboard at once.
To add one:
1

Choose the metric

Open the filter menu, choose Another metric and pick the metric to filter on. Each metric shows its output type (Yes / no, Category, Scale, Number, Count, Time or Text).
2

Choose a side (optional)

If the metric scores each participant separately and isn’t fixed to one side, choose Either side, Agent or Customer. Either side matches when either participant’s value matches.
3

Set the condition

The condition depends on the metric’s output type:
The filter shows as the metric’s name and the condition, such as Information given = True (using the metric’s own labels), with the side in brackets when you picked one: Information given (Agent) = True. A conversation matches on its latest value for the filter metric. Conversations the filter metric hasn’t scored are left out. A threshold can be the filter metric too: choose Passed to keep the calls that passed it, or Failed for the calls that failed it. Each call is evaluated the same way as the threshold’s own pass rate, so the calls a Passed filter keeps are exactly the ones its pass rate counts as passing. A call with no values for the metric the threshold checks is in neither set. A threshold has no side to choose: it reads the side it was defined on.

Worked example

You score calls with two metrics: Information given (yes / no) and Retrieved accurately (yes / no). You want to know how often the agent gets the information right when it gives any.
  1. Add a widget for Retrieved accurately. As a yes / no metric, it shows the share of calls where it came out true.
  2. In the widget’s filters, choose Another metric, then Information given.
  3. Choose equals, then the metric’s true label.
The widget now shows how often Retrieved accurately is true over only the calls where Information given is true. To compare against the rest, add a comparison group for each value of Information given instead. Over the API, the same filter is a METRIC_VALUE filter: see Query analytics.

Managing Dashboards

  • Default Dashboard: Set which dashboard loads by default
  • Quick Access: Switch between dashboards from the navigation
  • Update Reports: Add, remove, or reorder reports as your needs change

Dashboards over the API

Everything on a dashboard is also available over the API, the SDKs, the CLI and the MCP server. A dashboard made in Roark reads over the API, and one made over the API opens in Roark. Read a dashboard’s numbers. GET /v1/dashboard/{dashboardId}/data runs every widget and returns each one’s series exactly as the dashboard shows it: over the dashboard’s saved window and with its filters applied. Pass from (and optionally to) to read another window, and timezone to align the buckets.
A widget that can’t be computed on its own (say its metric was deleted) carries error and no series; the rest of the dashboard still returns. GET /v1/dashboard/{dashboardId}/widget/{widgetId}/data reads a single widget. Create and edit dashboards. A widget is an analytics query plus a name (unique on the dashboard), a display (LINE, NUMBER, RADAR, STACKED_BAR, TABLE) and a size (SMALL, MEDIUM, LARGE, EXTRA_LARGE):
PUT /v1/dashboard/{dashboardId} changes only the fields you send. Sending widgets replaces the list, matched by name: a widget that keeps its name keeps its id, a new name adds a widget, and a name you leave out is removed. Dashboard-level filters must be conversation-level (AGENT_PROPERTY, CALL_PROPERTY, MODALITY, DATASET or METRIC_VALUE). On a field a widget already filters, the dashboard’s filter wins. Reading needs the metric:read permission; creating, editing and deleting need metric:create, metric:update and metric:delete.

Dashboards as code

Dashboards can live in git with the rest of your config as a kind: dashboard resource. A dashboard created by config shows Managed by config in Roark: its title and widgets are read-only there and over the API (edits return 409), so they only change through your config. Anyone viewing it can still pick another time range and filter it. Duplicating a managed dashboard gives you an editable copy.

What’s Next

Reports

Create the reports that power your dashboards

Query analytics

Ask for the same numbers from code or an agent

Dashboards as code

Define dashboards in YAML

Live Monitoring

Monitor calls in real-time