A custom dashboard is one you build yourself: you describe a chart in a sentence, an assistant generates it, and you arrange the results on a page. Custom dashboards belong to the project you are working in. A custom dashboard sits beside the built-in dashboard rather than replacing it, so its Business Metrics, AI Agent, HTTP Metrics, Intents and Cost Analysis tabs stay where they are. Each of those tabs has its own page, listed on Overview. What both dashboards need is on Dashboards.
Dashboard in the left menu is a dropdown with two entries:
| Entry | What it opens |
|---|---|
| Overview | The built-in dashboard and its tabs. |
| Custom | Your own charts. Carries a Beta badge. |
📝 Note: Custom dashboards are available in the Premium edition. The Chart Assistant also needs a language model, set under Settings > Custom Dashboard (LLM configuration, model and reasoning effort). Until one is set, Create Chart with AI and each chart's Configure Chart are disabled. Existing charts, the filters, rename, resize, reorder and Delete All still work.
Opening the Custom dashboard
Go to Dashboard > Custom. Before any chart exists, the page shows an empty state.
1. No charts yet: The empty state. It is replaced by your charts as you add them.
2. Create Chart with AI: Opens the Chart Assistant. The same button sits in the page header.
Creating a chart
Select Create Chart with AI. The Chart Assistant opens on the right, with the subtitle Describe the chart you want and it is added to the dashboard. Every chart you create here is added to the dashboard you have open.
Describe the chart in one plain sentence and send it. For example:
Show the breakdown of conversations by request category as a donut chart
Generation takes several seconds, sometimes up to about a minute. The chart then appears on the dashboard and the assistant replies. The screenshot shows an example; the reply's wording is generated and varies.
1. The reply: Describes the chart that was made.
2. Chart chip: The new chart's title and a counter, N/20: the number of charts the dashboard now holds.
3. The message box: Its placeholder suggests a next chart, ending in "(press Tab)". Press Tab to put the suggestion in the box, or type your own. The box also has an attachment button and a microphone.
Chart types
Fourteen types are available: bar, stacked bar, line, area, pie, donut, table, big number, horizontal bar, funnel, polar area, heatmap, treemap and Sankey.
Name the one you want in your sentence, as the worked example above does with "as a donut chart". If you do not name a type, the assistant picks one from your description.
💡 Tip: Ask for one chart per message. A sentence that names one data subject, one grouping and one chart type gives the assistant everything it needs and gives you a reply you can check at a glance.
Choosing a field to group by
What decides whether a chart is worth looking at is the field you group on. A field with a fixed set of values groups into a chart you can read; a free-text field does not. A Category AI Insight holds exactly one value from a set its author defined, so a chart grouped on it comes back with one slice or bar per value in that set. A free-text insight such as Topic or Summary holds a different sentence for almost every conversation, and a chart grouped on it arrives with a legend nobody can read. The assistant will not refuse the request or warn you; it simply draws it.
Custom attributes behave the same way. For anything you intend to count or compare, group on a Category insight, on a custom attribute with a value list, or on a built-in field with a fixed set of values.
Moving, renaming and removing a chart
Hover a chart card and its controls appear.
1. Drag handle: Move the chart to another position on the dashboard.
2. Configure Chart (magic wand): Opens the configuration panel for this chart.
3. Rename (pencil): Changes the chart's title. It changes nothing else: not the data, not the grouping, not the chart type.
4. Delete (trash): Removes this chart from the dashboard.
5. Resize handle: In the bottom-right corner of the card. Drag it to change the chart's size.
Changing a chart
Select the magic wand on a chart to open Configure Chart. Use it to change the chart type, what it groups by, its filters or its title. A new chart is asked for from Create Chart with AI.
1. Scope: Names the chart the panel is working on. Changes apply to it and to no other.
2. The reply: Here the request was for a single colour shade, and the assistant confirms the chart now uses a monochrome palette. The chart redraws.
3. The message box: Describe the change in a sentence.
A change the chart configuration cannot express is not silently dropped. The panel comes back with a card that names the limitation, explains it, and offers two ways forward with one of them already selected. Nothing is applied until you confirm. Two limitations you are likely to meet this way:
- A bar chart cannot print its values at the end of each bar. The card offers a Table view of the same data instead.
- One chart cannot use two custom AI insights at the same time. The card offers to pair one of them with a standard conversation field instead, which is how a Sankey of request category into resolution type gets built.
The wording of these cards is written fresh each time rather than picked from a fixed list, so the heading and the option labels on your screen will not match anyone else's word for word. What stays the same is the shape: the limitation, the reason, the choices, and nothing drawn until you pick one.
Closing the Chart Assistant or Configure Chart discards its conversation; reopening starts from the panel's opening message.
Filtering the dashboard
The filter panel on the left applies to every chart on the dashboard at once, not to the chart you last touched. The exception is a chart built on data table rows: the filters do not narrow it, and it always shows the whole table.
1. Date range: Yesterday, Today, Last 7 days or Last 30 days, or a range you set. The page opens on Last 30 days and returns to it on every visit.
2. Channel: Limits the charts to the selected channel. Your selection is kept between visits.
Delete All, in the page header, clears the whole dashboard. It asks for confirmation first, and it removes every chart on the page, not only the filtered ones.
What you can chart
Charts are built over five data subjects. The assistant chooses the subject from your description, and the more precisely you name what you want counted, the less it has to guess.
| Subject | What it carries |
|---|---|
| Conversations and sessions | The channel, the project, the client and the conversation's own identifiers. Then, for each session: when it started and finished, how long it ran, its sentiment score and the sentiment it ended on, its resolution type, how many times it was handed to a person, how many messages came from the AI agent, from a live agent, from the virtual agent and from the user, how many times the AI agent was invoked, whether it was a voice conversation, and its topic. AI Insight values and Custom Attribute values hang off this subject. |
| Flow executions | Which flow ran, with its name and type, where it sat in the sequence, when it started and finished, how long it took in total and how long in itself, and the name of the flow that called it. |
| Node executions | The node's type and its display name, the output port it left by, when it was entered and left, how long it ran and how long it spent waiting, and the node before it together with that node's own output port. |
| Ratings | The score, its comment, whether a comment was left at all, and when the rating was given. |
| Data table rows | The columns of the table itself, under the names you gave them. |
Two things to keep in mind when you read a number off one of these charts.
The session fields are per session, and one conversation can hold more than one session. A total over a session field is therefore a total over sessions, not over conversations, and the two will not match.
Transition questions are answerable here. A flow execution carries the flow that called it, and a node execution carries the node before it with the port that node left by, so a chart can follow one step into the next: node type into output port, or one flow into another. That is what a Sankey over this data draws.
Two things that change what a chart shows
- A chart grouped by a Category AI Insight leaves out conversations that have no value for that insight. The chart's total is then lower than the number of conversations in the date range, and the difference is the conversations the insight did not classify. See AI Insights.
- A date series follows the width of the range. Over Last 30 days a line chart plots one point per week; over Last 7 days it plots one point per day. Asking the assistant for a different grain does not change it, so change the date range instead.
Limits
| Item | Value |
|---|---|
| Charts per dashboard | 20 |
| Date range | Up to 3 months; default Last 30 days |
| Chart types | 14, named under Chart types above |