Cost dashboards
The Dashboards tab of FinOps → Cost & savings lays cost out over days and weeks: what a cluster’s compute cost, where it went, and what grew. It holds Seylo’s FinOps templates, read-only, and your organisation’s own cost dashboards. Like the rest of Seylo, a panel never invents a figure: a period with no data is not drawn, a figure that cannot be priced is not shown as zero, and the panel says why.
Pick a cluster at the top of a dashboard, or Whole cloud bill (no cluster) for the whole-cloud panels alone. Refresh reads it again.
The cluster templates
Section titled “The cluster templates”These read the selected cluster’s compute cost. Open one to see it with the cluster’s figures.
| Template | What it shows |
|---|---|
| Clusters | Compute cost over the last 30 and the last 7 days; idle capacity over 30 days and per day; cost by namespace and by workload over 30 days; daily cost by namespace; and requested against used by namespace over 7 days |
| Weekly trend | The cluster’s compute cost for the last complete week, against the week before and against the 8-week average; weekly compute cost for up to 26 weeks; and weekly cost by namespace |
| What grew | Last complete week against the average of the 8 weeks before it, by namespace and by workload, biggest rises first, as a bar and as a table; and namespaces against the week before |
Weekly trend and What grew also have whole-cloud panels at the top, and five more templates are whole-cloud only: see whole-cloud spend.
Idle is the cost of node capacity no workload requested. It belongs to no namespace, so idle panels show figures only to people who can see the whole cluster. Someone limited to some namespaces sees only those namespaces’ figures, and the panel says so.
Requested against used shows average cores and GiB: what the pods asked the scheduler for against what they used. It has no prices.
Where the prices come from
Section titled “Where the prices come from”Cluster panels show compute only: the CPU, memory and GPU allocated to workloads (the larger of what was requested and what was used), from OpenCost in your cluster. Seylo prices them node by node with the same rule as the Cost tab: OpenCost’s price where it agrees with the price book or comes from your live price sheet, rescaled to the price book where it differs, and otherwise not priced, with the reason. With your Azure bill connected, the price book holds what your bill says each machine type cost. Storage, public IPs and egress are priced on the Cost tab’s table and are not in dashboard panels yet. Figures are in US dollars.
OpenCost must be on in the cluster’s seylo-observability chart. Until it
is, each panel says OpenCost is not running in this cluster. Turn it on with:
and shows the command.
How periods are counted
Section titled “How periods are counted”- Days and weeks are complete. A day is a UTC day; a week runs from Monday 00:00 UTC to the next Monday. Today, and this week, appear when they end: a half-finished week beside full ones would look like a fall in spend.
- History starts where the data does. OpenCost works each period out from the cluster’s metrics store, so history goes back only as far as that store keeps data (see retention). A 26-week trend usually has fewer weeks of history. Periods before history, and the one history starts part-way through, are not drawn - not drawn as zero - and the panel says where its history starts.
- A comparison is arithmetic. The last period minus the one before it, or minus the average of the periods before it. Nothing is projected, smoothed or filled. A row with an unpriced part shows no change, rather than a misleading one, and says why.
- A table or chart lists the biggest groups; the rest are summed into one Other row.
Each OpenCost read is kept for 5 minutes, so a dashboard opened twice in a row reads once.
When a panel has nothing to draw
Section titled “When a panel has nothing to draw”A panel says what is missing instead of drawing zeros. Under each panel:
- where its history starts, and what a comparison is against;
- Not drawn:, with each group that could not be priced and why;
- How this is counted, which opens the panel’s notes: what it covers, every reason a figure is not priced, and, for people who can see the whole cluster, which price each node used.
When there is nothing at all, the panel says why instead, for example:
| What the panel says | What it means |
|---|---|
OpenCost reported no allocation in this window. |
OpenCost has no figures for the window, or none in the namespaces you can see |
The last complete week (...) has no figures: ... |
The week the panel shows is not in history, or OpenCost reported nothing for it |
No week before the last one is in history, so there is no baseline to compare with: ... |
A comparison needs at least one complete period before the last one |
You cannot see namespace "..." in this cluster, so this panel has nothing to show you. |
The panel is filtered to a namespace outside your access |
If OpenCost cannot be reached, the panel says why and what to run, as the Cost tab does.
Whole-cloud spend
Section titled “Whole-cloud spend”Some panels show your whole cloud bill rather than one cluster’s compute: the top of Weekly trend and What grew, and the templates Week at a glance, By subscription, What moved, Cost drivers and Projection.
They read your Azure bill once your organisation has an Azure subscription connected as a cloud account: no cluster, OpenCost or billing export is needed. See your Azure bill.
- The bill as billed. Each day’s cost per service, meter category and location, in the bill’s own currency. Seylo converts nothing. Subscriptions billed in different currencies are shown apart and never added together.
- Up to 190 days of history. Seylo reads the bill every six hours.
- Every connected subscription, not only one cluster’s. A subscription that could not be read is named under the panel, as not in its figures.
Without a cluster. Choose Whole cloud bill (no cluster) in the Cluster list to draw the whole-cloud panels on their own, for example before any cluster is connected. Only whole-cloud panels are drawn this way; a cluster panel asks you to choose a cluster. With namespace scoping on in your organisation, this choice is refused: choose a cluster you can see the whole of instead.
What the bill cannot answer. Seylo reads the bill per service, meter category and location. A panel that groups or pins by anything else says why instead of drawing:
| What the panel says | Why |
|---|---|
This panel groups or pins by a single resource, and Seylo reads your Azure bill per service, meter category and location, not per resource. |
Resources are not in the daily bill Seylo reads |
This panel groups or pins by a tag, and Seylo does not read tags from your Azure bill yet. |
Tags are not in the daily bill Seylo reads |
This panel groups or pins by invoice entity, which Seylo does not read from your Azure bill (it reads each subscription). |
Group by subscription instead |
With no Azure subscription connected, each of these panels says why and how
to connect one, and shows no figure - never zeros. On
Whole cloud bill (no cluster) it says
No Azure subscription is connected to this organisation, so there is no bill to draw.
A whole-cloud figure belongs to no namespace, so only people who can see the
whole cluster will see one; anyone else sees Whole-organisation spend: ...
instead.
Your own cost dashboards
Section titled “Your own cost dashboards”With releaser:
- Copy to your dashboards on a template, or Customise it, to make it your organisation’s own. Nothing is saved until you copy or customise.
- New dashboard starts from scratch. Build with AI writes one from a description, using your organisation’s model, as on observability dashboards.
- The editor has three tabs: Assistant, Panel (a form: what the panel is Of, its Kind, what it Shows, Group by, Steps, Window and what to Compare with) and the dashboard’s JSON. Save checks every panel and says what to change: a trend needs steps, a stat shows one figure, and a comparison needs a window one period longer than the periods it averages.
A cost dashboard a coding agent saves through Seylo’s MCP server is listed
under Proposed by coding agents, and nobody else sees it until a
releaser selects Publish, or Discard draft. Every panel runs with the
access of the person looking at it.